you2idea@video:~$ watch gcuR_-rzlDw [6:23:04]
// transcript — 4782 segments
0:00 Hey, welcome to the definitive course on vibe  coding for beginners. I use anti-gravity,  
0:04 Gemini 3.1 Pro, and Claude Code in my life every  day to manage a business that does over $4 million  
0:09 a year in profit. I also teach around 2,000 people  how to use technologies just like this to improve  
0:14 their lives, both personal and business. So, the  way this course is going to work is it'll start by  
0:18 teaching you guys how to set up anti-gravity,  Gemini, and then Claude Code in a simple  
0:22 development environment. And then we're going to  jump right into the building. And we'll start with  
0:26 simple portfolio sites before working our way  all the way up to full-fledged SAS apps with  
0:30 authentication, databases, and so on and so forth.  At every step along the way, I'm going to pause  
0:35 and then show you guys some of the principles  behind what I just built. And my recommendation  
0:38 is if you want to get the most out of this course,  you'd actually build alongside me. Okay, so no  
0:42 fluff. Here's what we're going to cover. We'll  start with the simple toolkit. So anti-gravity,  
0:46 Gemini 3.1 Pro, and Claude Code. I'll give  you guys the best development environment  
0:49 available in probably around 15 minutes. Then  we'll build your first vibecoded product,  
0:53 which in our case is going to be a personal  site designed with anti-gravity and Gemini.  
0:56 You'll have this live on the internet, including  deployment in about 15 minutes. After that,  
1:00 I'll pause to talk authentication and security cuz  I think it's one thing to build simple portfolio  
1:04 sites and stuff, but if you wanted to push that  out to a fullyfledged app with authentication,  
1:08 databases, and a back end, a big issue in vibe  coding is security, and I want to make sure  
1:12 that we cover at least the 8020 of that. Then  we'll build another product, which will be a  
1:15 client-f facing dashboard with authentication, a  database, and a very clean UI. And at that point,  
1:19 since we'll have gotten our hands dirty, we  can then cover foundational concepts behind  
1:23 modern software applications, including hosting,  databases, the major software design patterns,  
1:28 versioning and GitHub repositories, the major  frameworks currently available, then also how to  
1:32 use agents to vibe code all this stuff way faster.  I'm then going to give you a design specific deep  
1:36 dive, which is how to go from a rough idea to a  polished professional design and anti-gravity.  
1:40 Then we'll build our next product, which will be  a lead generation SAS app that scrapes, enriches,  
1:44 and then cues up the sending of fully automatic  emails. This is an actual system I use production.  
1:48 After that, we'll do an AI based SAS app that  uses AI to generate thumbnails. It's actually  
1:52 really similar to the system that I used to create  the thumbnail for the video you're watching right  
1:55 now. So, pretty meta. Once we're done with that,  I'm going to teach you guys some debugging and  
1:58 iteration techniques. You guys will probably have  seen me do a lot of this and hopefully have done  
2:02 a lot of it yourselves by now. But, I'm going  to give you guys a simple framework that you  
2:05 could use to steer back Gemini and then claude  code anytime it starts veering outside of your  
2:09 uh recommendations. And then finally, we'll build  a content automation SAS app that repurposes a  
2:14 single piece of content across multiple platforms.  That's another system that I actually use. Before  
2:18 we wrap up, we'll talk about the economics of  vibe coding, which is how to price, package,  
2:22 and then sell what you build. And that's important  because I find a lot of people watching this are  
2:25 probably going to be doing so not just for their  own hobbies, but for financial reasons. We want to  
2:29 make money with the things that we build. Okay, so  I care a lot about making money with this stuff.  
2:33 I want to show you guys how to do that as well.  Hopefully that makes sense. Let's not waste any  
2:36 time. Let's strap in and get started. Okay, so  how do you actually set up anti-gravity? Just  
2:40 head over to anti-gravity.google. They actually  have like the Google domain name, which is  
2:45 hilarious. and then just click on this little  download button. If you guys are on a Windows,  
2:48 it'll be download for Windows. Since I'm on a Mac  OS, this is download for Mac OS. Now, if you guys  
2:52 don't know which one of these to select. It's  pretty easy. Most modern computing applications  
2:56 nowadays have the ability to type like about  and then you can just go about this Mac,  
3:00 about this PC or whatever. And then it'll tell you  whether or not you are an Apple silicon or Intel  
3:05 silicon for instance with this line right here.  If I was on Intel, it would actually say like an  
3:09 Intel M whatever. Because I'm on an Apple silicon,  it says Apple M4 Max. And so this is the one that  
3:15 I need to download. If you guys are on Windows,  double check to see if it says x64 anywhere.  
3:19 If you guys are on Linux, you probably know more  about computers than I, so I'll stop embarrassing  
3:22 myself. So I'm going to give this button a quick  click. That is going to download the anti-gravity  
3:27 installer. On Mac, what ends up happening is we  get a little thing in the top right hand corner  
3:30 here, which is anti-gravity. And then all you guys  have to do is just drag this anti-gravity app into  
3:34 the applications folder and then let go. I already  have anti-gravity installed, so I'm not going to  
3:38 do that part. But that that's the installation.  It's about that easy, at least on Mac. On PC,  
3:42 you're going to have a little screen that pops up.  Just click accept a bunch of times until you're  
3:45 done. Okay. After that, what you're going to get  is a page that looks something like this. So, I'm  
3:48 just going to zoom in to make things a little bit  more visible. And hopefully my big head doesn't  
3:51 get in the way. But basically, um the way that  anti-gravity works is you have to connect to your  
3:55 Google account in order to get up and running.  And once you connect your Google account, you  
3:58 can also use um you know, Gemini as a model. Uh  the reason why this isn't popping up on the right  
4:03 hand side as you see is because we haven't logged  in with Google yet. So, what I'm going to do next,  
4:07 I'm going to click this login with Google button.  I'm then going to sign in and click sign in. and  
4:10 then it'll successfully authenticate you and then  take you back to the main menu. After that's done,  
4:14 you'll see up at the top we have a bunch more  features. In order to open up the AI feature,  
4:18 just click on this one on the right. It's  called your agent. There's also a little hotkey,  
4:21 which I use pretty often, which is just  um option uh commandB. And by doing this,  
4:25 I can make this pop out really quickly and  easily. Obviously, if you're on Windows,  
4:28 things will be a little bit different, but it's  not that bad. Now, you're inside of anti-gravity.  
4:32 You can actually have a conversation. So, I'm  going to say, "Hey, what's up?" I'm going to  
4:35 run through the UX of how to communicate with your  agent in a second, but I just wanted to show you  
4:39 guys how easy it is to actually get the AI part  up and running. Um, Gemini 3.1 Pro High should be  
4:43 provided to everybody, although with different  usage limits depending on whether or not you  
4:47 upgrade their plans and so on and so forth. And  you also have other models too, like 3.1 Pro Low,  
4:51 Gemini 3 Flash, Sonnet 4.6, Opus 4.6, and so on.  Okay, cool. So, that's just getting anti-gravity  
4:57 up and running. The next thing I want to do is I  actually want to explain the UX. After I explain  
5:01 the UX, I'll also talk about Claude Code, which  is the alternative chat window, kind of like this  
5:05 little agent chat widget, um, except for the Cloud  series of models that performs really, really well  
5:10 in specific applications, which we're also going  to use in conjunction with Gemini in this course.  
5:13 So, really, I want you to think about anti-gravity  as being divided into three panes. On the leftmost  
5:18 pane, you have basically the explorer. And for  the most part, you're only really going to be  
5:22 using the Explorer. You're not really going to  use the rest of these things. In the middle,  
5:25 you're actually going to have your file editor.  And right now, we don't have any files open,  
5:28 which is why we just see the anti-gravity logo.  And then on the right hand side, you actually have  
5:31 like your agent chat, your ability to, you know,  talk to AI. And this is how most like integrated  
5:36 development environments are nowadays. Um, I say  most because anti-gravity isn't really technically  
5:39 an IDE. There's like a couple of additional  features here, but I don't want to be pedantic.  
5:43 For all intents and purposes, this is an IDE. And  this is how it works. So, in order for us to see  
5:47 the things on the lefth hand side here, this  explore, we actually do need to open a folder,  
5:50 which is why it's asking us to do that. So, I'm  going to click open folder. And then I just set up  
5:54 this little demo for a video I recorded yesterday  called website builder. So, I'm just going to open  
5:57 that up. Okay. And now hopefully you guys could  see the page is transformed in this middle section  
6:01 is our our file explorer. So because I don't  really think files are as important these days  
6:06 to explore because AI does most that for you.  I usually make the lefth hand side or the right  
6:10 hand side way bigger. A lot of the time I actually  just issue this middle pane entirely actually and  
6:15 then just chat with the agent. But I'll I'll show  you guys how that works later. For now let's learn  
6:18 this on the left hand side here. As you see this  is just a file explorer like anything else just  
6:22 like a finder on your computer or whatever. You  know you have the exact same thing here. So, you  
6:27 know, in my case, I'm scrolling through my Mac uh  files and and folders and stuff like that. Well,  
6:32 that's basically exactly what anti-gravity does.  It just does it for you in a slightly different  
6:35 user experience. And so, folders in this case have  this little folder icon, right? Um specific types  
6:41 of files which have different file endings have  other logos and stuff. This is a markdown file,  
6:46 which is why it's an M and then it's a down. Uh  this is an HTML file, which is why it looks like  
6:50 that. And, you know, you don't actually need to  know a ton of programming in order to get all of  
6:53 this stuff. To be honest, um I don't have a formal  programming degree or anything like that. I still  
6:58 understood this stuff just fine after spending a  little bit of time inside of this dashboard. So,  
7:01 just to prove to you guys that, you know, we're  in a file. Uh if I click on this index.html here,  
7:05 you see that I actually have like a website file  set up here. HTML is the um ending for websites  
7:11 basically. And uh you know, there's a bunch  of different text here in this middle editor.  
7:15 Doc type, HTML head, meta, right? I don't actually  know what most of this stuff means. I just have my  
7:20 agent deal with it. But it is important from time  to time to at least be able to poke into the file  
7:24 and and do things and stuff like that. Just like  you can do with any file explorer, you can create  
7:28 new files here. So I can make like another file  MD. Then I could write stuff. Hey stuff. Okay.  
7:34 Um you can also delete files. So rightclick  and then you can you know delete, rename,  
7:38 open timeline. You can do all sorts of random  stuff with this. But I just want you to treat  
7:43 everything on the lefth hand side as basically  your average run-of-the-mill file explorer. And  
7:46 then the stuff in the middle you can almost treat  like like notepad on Windows or like Apple notes  
7:50 in the middle. Um it's just because in programming  typically specific character sequences are tied to  
7:56 specific pieces of functionality. You'll see here  that like it's quite colorful like link tags look  
8:00 a specific way. This thing looks another way. If  I go into I don't know this Tailwind config. You  
8:06 see we now have the color purple and and yellow.  Basically they just make it look like a freaking  
8:11 rainbow. And that's cool cuz why would I be  working on something gray dull and featureless  
8:15 if I could be working on a rainbow instead? On the  rightmost side of the file explorer, you also have  
8:19 the ability to like um have like a quick highle  overview of the file. So that's kind of neat.  
8:23 I don't know if you guys could tell super well  here, but this is basically like a I don't know,  
8:27 it's like a a level of abstraction up. It's like  a window that allows us to scroll up and down.  
8:30 This is useful if you're looking for specific  things within the file. Although I'll be honest,  
8:33 I mean like nowadays, as I mentioned, I don't  really use the editor too much. And then up at the  
8:37 very top right hand corner, this is relevant if  you run any markdown files. If you click on this  
8:42 little button here, and by the way, most files  nowadays where you communicate with the model  
8:46 are going to be markdown files. Instead of it  looking kind of like dull and drab and you know,  
8:51 all this tiny text here, if you click on this  little button here, you'll actually open what's  
8:54 called a markdown preview. And then this markdown  preview will actually like make the text the way  
8:57 that it's supposed to look. So every time in  markdown, you have this little like number  
9:01 sign. What that actually refers to is that refers  to like a heading. And that's like the really big  
9:04 heading. If you have two, then that's a little  smaller heading. Then you have like text down  
9:08 here. So, most of the time, this is what you guys  are probably more familiar with. And to be honest,  
9:11 a lot of the time, if I just want to like take  all this dense stuff and then make it look a lot  
9:15 better, I'll go back to um my browser and then  I'll go to Google Docs. And then what I'll do  
9:20 is I'll actually paste this in. And then I'll  just change the text color to black. And then  
9:25 now everything here just looks like a big Google  doc. And I can read through it. And you know,  
9:28 I see like orders of um hierarchy and and stuff  like that. Not super important, but good way for  
9:32 you to like quickly take a look at prompts.  Okay, a couple of quick hotkeys here. Um,  
9:36 you can go forward just uh with this. You can go  backward with this. It's the exact same hotkeys  
9:40 that you use on um, you know, Windows or Mac to  like change tabs and stuff cuz basically what  
9:45 we're doing is we're just changing tabs. So,  if you're alt tabbing, you can alt tab through  
9:49 files like that. Super easy. And then you'll also  notice that there's just different types of files  
9:53 depending on what you're looking at. So, we talked  about markdown. This is a JSX. This is a CSS. This  
9:58 one over here though is actually a Gemini specific  one called walkthrough where it runs you through  
10:01 what the AI uh basically wants to do next. It  basically teaches you what it's doing. It's kind  
10:06 of neat and we're going to cover a lot more about  that later. But there's cool features like review,  
10:10 you could submit a comment, you could ask the  model for stuff and and download it as well which  
10:14 is kind of cool. Okay. And then finally just to  round out our AD20 on the right hand side we have  
10:18 the agent panel. And this is where I'm going to  spend most of my time because as mentioned this  
10:21 is where like AI does most things for us nowadays.  And we don't actually have to worry too much about  
10:25 the folders or the files or the code. What we  worry about is we just worry about specifying what  
10:29 we want in natural language. That's really what  like the vibe and vibe coding is all about. Okay.  
10:33 So, starting at the very top right hand corner,  if you want to start a new conversation, you just  
10:37 click this button. Now, we don't actually have any  text in our current conversation, so I can't start  
10:41 a new one. If I just write a, then I can click  this. Now, I'm in a new conversation. You'll  
10:45 also see down here that our history is populated.  And we can see 4 seconds ago, I had the letter A.  
10:50 Uh 24 hours ago, I asked some questions about  background images loading and stuff like that.  
10:53 really quick and easy way to create a new convo.  Um, you can obviously also just use hotkeys. So,  
10:58 command shift L will just automatically pop one up  and then, you know, every time you type, you can  
11:02 just make a new one and so on and so forth. Over  here are past conversations and so this will have  
11:07 uh basically a list of every conversation I've  ever had with AI. So, refining prompt for website  
11:11 generation 47 minutes ago. This is something that  I produced. Uh, when you click on that button,  
11:16 it'll say, "Hey, should we open in the current  window? Should we open in a new workspace?" So,  
11:19 what I'm going to do is I'm going to open in a  new workspace. It'll open up a new folder. And  
11:22 now it'll actually pull through my conversation as  well as the specific file that I had open when I  
11:26 did this. In my case, I was preparing a website  uh prompt basically a website generator prompt  
11:31 for this course uh which is right over here and I  was just having a back and forth for that. On the  
11:36 right hand side, we also have additional options  which allows us to do things like customize and  
11:39 MCP servers. Customizations are pretty cool.  Basically what you can do is you can define  
11:43 highle rules that guide the behavior of your AI  model. So, we're going to get into ways to steer  
11:48 things as I mentioned in the um sort of outline of  this course, but rules are really quick and easy  
11:52 ways that you can either set globally or within  the specific folder that you're operating, okay,  
11:57 to either control all instances of every line of  code or every conversation you have with Gemini  
12:04 or you can do that within a fold. So, you can  see here I technically already have a rule and  
12:08 it's called cinematic landing page builder  role. Your role is to act as a world-class  
12:12 senior creative technologist and lead front-end  engineer. you build highfidelity cinematic one  
12:16 to one pixel perfect landing pages. So this is  sort of like my highle um workspace rule that  
12:21 helps constrain the behavior of the model.  And you know this is actually pretty long,  
12:25 right? Like you could scroll down this for quite  a while. Um this is one of the reasons why I have  
12:28 pretty good vibecoded quality because I basically  put in a lot of style guide and highle information  
12:33 to help the model you know make quick decisions on  the fly without having to ask me for stuff. So you  
12:37 can define that both globally or a workspace. If  you just click the button then you can enter the  
12:41 information right over here. You click this button  and you can just um create your new rule. So maybe  
12:45 I want it to be called design guide. Oh, and you  have to use um you know hyphens in between each.  
12:51 You can't actually have any spaces. I'm just using  a little voice transcription tool. In this case,  
12:54 this one's called whisper flow. Um and then you  know you can basically specify an activation mode.  
12:59 Is it always on? Is it manual? Do I set it? Do uh  I leave it up to the model or do I do glob which  
13:06 applies specifically to files that match some file  pattern? In practice, I never use glob. I never  
13:11 use model decision. I never use manual. I always  use always on. So in this case, I'd say, hey,  
13:15 when you design websites, I want you to stick to a  minimalistic, sleek aesthetic. I want you to build  
13:19 things that are significantly closer in design to,  let's say, the Apple homepage than anything else.  
13:24 And I can populate that. If I wanted to run this,  I would just save this file. And as you can see,  
13:27 we've now added that as a rule. Now, I  don't actually want this rule here. Um,  
13:31 so what I'm going to do next is I'm just going  to delete this. The way you do so is every time  
13:35 you generate a rule, it's now created inside of  agent/ruules folder. So if you want to delete it,  
13:40 you just delete that. And now we're back to just  that one rule which is called the gemini.mmd. This  
13:44 gemini.mmd is basically just the parent rule that  always applies if you have a file called capital  
13:49 gemini.mmd in your workspace. All Gemini models  will just always take that as a rule. Um multiple  
13:54 different ways to define rules and stuff like  that as well. Not super important. There's also  
13:58 workflows which I think are probably nowadays  better understood as skills where basically  
14:02 similarly to rules you can define behaviors or  specific sequences of steps that you want your  
14:06 agent to take every single time. So these can  be pretty cool and you can again define them  
14:10 as global or workspace. So I could call this I  don't know code review here. I could say this  
14:15 workflow defines a code review process that goes  through the code and then maximizes the security  
14:20 of the vibe coded app. To start with, I want you  to reference this document. This document includes  
14:25 information about all sorts of high ROI security  practices for vibe coded apps. After you're done,  
14:29 I'd like you to go through each of these step  by step. Uh make small incremental changes to  
14:33 the codebase. Try not to, you know, do everything  everywhere all at once. and then verify with me  
14:38 at the end that it works by asking me to test  it before we push to production or anything  
14:42 like that. And so, you know, in this case, what  I've done is I've I've built some quick little  
14:48 rule here. And then maybe over here is where I  put like my security stuff. And now, you know,  
14:53 what I have is I have a quick little workflow  that after I'm done building my vibe coded app,  
14:57 I can just say, "Hey, I want you to run the code-  review workflow." and then it'll just go through  
15:01 this standard operating procedure, the sequence  of steps, more similar to like the way that a  
15:06 business would operate with checklist style  tasks. Again, I don't actually want this. Um,  
15:10 so what I'm going to do is I'm going to go  back here. I'm just going to delete this.  
15:12 That'll delete the code review customization.  Now, there's also some other customizations  
15:16 like MCP servers. I'm going to talk a little bit  more about MCP servers later. But the way I want  
15:20 you to treat MCP servers for now is basically  it's just a way to define workflows. It's just  
15:25 um workflows that somebody else did for you.  And so these MCP servers essentially allow  
15:29 you to connect to different tools and then run  standardized workflows on those tools. So for  
15:33 instance, if I wanted to use a server like a  PostgresSQL, there's a pre-existing workflow  
15:38 bundle here called CloudSQL for PostgrSQL. This  is a specific database language and specific  
15:44 type of database. This basically just defines  a checklist series of steps that the model will  
15:48 always do to communicate with us that I don't  know can translate my natural language queries  
15:52 into that if I wanted to. Same thing with Looker.  Same thing with GitHub, Neon, Stripe. Basically,  
15:58 these just allow us to communicate with different  tools in a very predictable way, which as I'm sure  
16:02 you can imagine if you're building high quality  vibe coded apps can be pretty useful. That said,  
16:06 there are also some downsides to MCP server.  So, I don't really use them a lot in practice,  
16:09 and we'll cover all that stuff later. For now,  I don't want to overwhelm you. Okay, next up, we  
16:14 have the title of the current folder. As you can  see, this one's just called website-builder, not  
16:17 rocket science. Then, we have the ability to add  some context here to our prompt. So, if I wanted  
16:21 to, I could add media. I could add mentions. And  I could also add workflows. Remember the little  
16:25 workflow that I put together earlier and then  deleted? If I had workflows, every time I wrote  
16:29 slash, I would automatically be able to add them  here. What's cool though is you can also mention,  
16:33 and when you type mention or or you just press at,  then you can actually attach files, directories,  
16:37 uh, code context items, rules, even terminals,  pre-existing conversations, and then also MCP  
16:42 servers into the context. And the reason why this  is valuable is because, you know, I could just  
16:47 copy everything in this file and paste it in, or I  could just say, "Hey, I want you to reference this  
16:51 gemini.md." And then it would do the same thing,  but as you can see, I'm not really polluting this  
16:54 big window. It's very clear and clean. After that,  you have the mode. Now, there are two conversation  
16:59 modes available to Gemini 3.1 Pro right now.  There's planning and fast. This may change at  
17:03 a certain point in time, but for now, this is more  than enough. In general, you always want to start  
17:07 in planning anytime you're designing or building a  new feature. Planning is good because it basically  
17:11 allows the model well almost forces the model  to do a lot of thinking before it actually like  
17:15 you know stretches its fingers out in the world  and starts manipulating files and stuff. It then  
17:19 actually presents you a workflow and says hey this  is what I'm planning on doing. Are you sure you  
17:24 want me to do this? And then you can set your own  behavior to either auto approve which is just have  
17:28 it run on its own um decided plan or you could  actually stop it and say hey I didn't really like  
17:33 what you did here. I want to make a minor change.  Next up you also have fast. fast just allows it  
17:38 to execute things directly and go really quickly.  And so this is actually pretty useful if you're  
17:42 doing quick one-off tasks. Hey, I want you to  change this file format to this file format. Hey,  
17:46 I want you to change this markdown file to a txt  file. Hey, I want you to, I don't know, refactor  
17:50 XYZ really quickly with this um, you know, brief  little variable in mind or something like that.  
17:55 Okay, so in general, I always start with planning  and then at the end um, you know, when I'm just  
17:58 doing one-off tasks and stuff, if I need to, I'll  use fast. Underneath, we have the model picker,  
18:03 which I've already covered. includes the specific  one you want to pick and choose and actually have  
18:07 in communication with. Remember how earlier I  said the agent is just like a wrapper for the  
18:12 um specific model you're talking to. Well, this  is how that works, right? This chat window isn't  
18:16 communicating with all models right now. It's just  communicating with one, but I can hot swap them  
18:20 anytime. Instead of Gemini 3.1 Pro high, I could  easily choose Gemini 3.1 Pro Low. And you'll find  
18:25 that every model has different pros and cons.  If you want a quick example of pros and cons,  
18:29 just go to Google and type Gemini 3.1 Pro High  versus Low. And what you'll do is you'll have um  
18:34 you know basically benchmarks showing you how high  quality the models are at different things. Uh  
18:39 use usually some sort of like graph here that runs  through hey you know highs 44.4% on this benchmark  
18:45 lows 39.8% on this other benchmark and so on and  so forth. I will say in general I tend to just use  
18:50 like the best models currently available and to be  honest that's usually the most recent model that  
18:54 has been pushed out by the big three companies  right now which tend to be Google um OpenAI and  
18:59 then Anthropic. Uh I was all about you know claude  uh Opus 4.6. six just last week. Um, now you know,  
19:04 Gemini 3.1 Pro High seems to be very, very good  at like design specific applications. So, what  
19:09 I'm doing is I'm passing off my design stuff to  Gemini 3.1 Pro High. Then I have Opus 4.6. Kind of  
19:14 clean stuff up on the back end. Yeah, this is how  you select through a bunch of different models.  
19:17 And as you can see, it's just like Lego blocks.  They're hot swappable. Gemini 3 Flash is super  
19:21 super quick. Sauna 4.6 is a really long context  window. A lot of these have pros and cons. Um,  
19:26 nobody really uses this in practice though. Okay.  And then finally, um you can click this button to  
19:30 replace your texting with basically talking.  So what I'm doing right now is I'm just having  
19:34 a normal conversation with it. When I press  this button, um you know, and then I I send it,  
19:40 Gemini has a built-in uh mechanism to just  transcribe what I just said and then turn that  
19:44 into text. The reason why this is valuable and why  you're seeing a lot of people use voice dictation  
19:48 tools nowadays, it's because on average, people  speak much faster than they write. If I'm writing  
19:51 at 50 to 80 words per minute, I can speak at like  150 to 200 words per minute. So, it's obviously  
19:56 a lot faster and a lot quicker. Now that I've  actually sent a message, you could see how the  
20:00 UX has changed a little bit. We now have a thought  little tab. This basically runs you through the  
20:05 internal thoughts of the model. The reason why  they make this present and visible to you is  
20:08 because if it's making any sort of like egregious  error, you can just quickly dive in and say,  
20:12 "Okay, so why did it make that error?" This gives  you some transparency, some accountability, and  
20:17 then you can, I don't know, adjust your feature  prompting style to be better to solve that issue.  
20:20 It's also good because sometimes the models are  lazy and they don't really respond because they're  
20:24 trying to conserve tokens or whatever the heck.  This is a quick and easy way to like make sure  
20:27 that the internal thing is not uh the user asked  me this question. I don't really want to spend a  
20:31 lot of tokens. So I've decided to give them a much  simpler answer. This is how you can actually just  
20:35 figure out what's going on under the hood. The  text that is not sort of this darker thinking  
20:39 tab is uh the response. And so in this case to  get started on building a cinematic landing page,  
20:43 please please answer the following questions. The  reason why it's running me through that is because  
20:46 I've defined that as basically my highle workflow  rule uh or sorry just rule in gemini.mmd. I'm  
20:53 saying hey you know I want you to act as a world  class senior creative technologist when the user  
20:57 asks something immediately asked exactly these  questions and then it's asking me these questions.  
21:02 So basically this is taking everything from that  initial configuration file and then despite the  
21:06 fact that my voice message said nothing of any  import uh it just injected that up at the very top  
21:11 and now it's trying to you know do my my higher  level rule instructions. Okay so underneath you  
21:15 have good or bad. Uh this is just a way that  you can help train the Gemini models on like  
21:19 hey I like this response. Hey I don't like this  response. I never really use that in practice.  
21:23 And now you can see that there's also a bunch of  additional tabs that I've opened up here. On the  
21:26 lefth hand side there's changes overview. if your  request changed any files um you know on your work  
21:32 inside of your folder that's actually going to pop  up right here and then you can just take a look at  
21:35 them in practice nowadays basically nobody takes  a look at these they're just like oh yeah man you  
21:39 figure it out you got it uh next up is terminal  so sometimes Gemini or other models will launch  
21:44 terminal instances for you can actually see the  background processes running in that tab artifacts  
21:48 or files that are part of the conversation so you  know I added this gemini.md file for instance I  
21:53 mentioned it right and then over here we have  the browser and this is kind of neat you can  
21:57 actually launch a browser are inside of uh Gemini  nowadays. Um this browser, you know, I'm not going  
22:01 to say it's like the highest quality browser on  planet Earth, but it can be pretty cool. If I go  
22:05 over here, you could see um we're going to keep  it. We're now opened on the anti-gravity uh docs  
22:12 page for browser. And so, you know how some things  in life you can only really do on a browser? You  
22:16 can't just do everything via API or you can't  do everything like through a chat window. This  
22:20 is basically a quick and easy way that allows you  to do that. And I'm not going to say that this is  
22:23 like super super valuable or really really  all that helpful right now, but you know,  
22:27 it's pretty cool and obviously they're trying  to push more towards browser automation. Okay,  
22:31 so hopefully you guys now know everything about  how to use anti-gravity and then the Gemini series  
22:35 of models. What I want to do next is I just want  to give you guys a quick run through of Claude  
22:38 Code because we're also going to be using that.  Basically, the Claude large language model lives  
22:42 within the Claude code chat software which we also  wrap inside of anti-gravity. And so just the exact  
22:47 same thing that we just did with the agent chat,  we can do with the Claude code chat as well. So  
22:52 what I'm going to do next is I'm actually going  to exit out of this view. Okay? And then I'm going  
22:55 to sign up to Claude. Now, in order to use Claude  Code, you do need an account. So what you have to  
22:58 do is you have to go over to claude.ai. That'll  take you to a page that looks like this. And  
23:02 then you can enter some information here and then  continue with your email. I don't actually have an  
23:07 email called asf asdf asdfgmail.com. If I would,  I'd probably be a billionaire. Um, basically  
23:12 what'll happen is they'll send you a link over to  your email. You can click on that link and then  
23:16 it'll sign you in. From there, you have a brief  little onboarding that you need to fill out and  
23:19 then that'll take you to the most important part,  which is the pricing. Now, the way that the plans  
23:24 work for Claude Code is you need to select between  the pro or the max plan. Pro plan is $17 a month  
23:31 right now. The max plan starts at 100 and it can  be significantly higher. Unfortunately, right now  
23:36 they don't allow you to use quad code within a  free plan. That's just part of the limitations  
23:40 of using the cloud family of models and it's just  always a trade-off with these things, right? Like  
23:44 if you think about it, Gemini allows you to use  things for free. Uh Google, sorry, allows you  
23:48 to use Gemini for free despite not having to pay  any money, but then you have quite limited usage.  
23:52 Whereas Cloud Code requires you to sort of put  a certain minimum amount of money, in this case  
23:56 17 bucks a month in order to use it, but then it  tends to be a lot better at specific applications.  
24:00 So you don't have to necessarily sign up for  this, but if you did want to sign up for it,  
24:04 if you wanted to use um you know, Vibe Coding in  a way that I'm going to show you in this course,  
24:08 you would have to have some sort of subscription  ongoing. I don't want to push you guys to sign up  
24:12 to anything. I'm not affiliated with, you know,  Anthropic. I'm not affiliated with Claude Code.  
24:16 I'm not pushing any specific product on you in  this course. I will just say that I, as a business  
24:20 owner, uh, you know, I get a tremendous return on  investment from this stuff. Just a few months ago,  
24:25 in order to hire a developer at the level of Cloud  Code or, you know, Gemini with an anti-gravity,  
24:30 I would have had to spend 20,000 to $30,000 a  month. I would have had to go down to Silicon  
24:34 Valley and find like a super cracked dev and then  hire them on my team. And those people would work  
24:39 eight hours a day max. Um, this stuff does 24  hours a day. This stuff you can plug in APIs. It's  
24:44 totally automatable. This stuff gets smarter every  day, week or month. And so this is really like the  
24:48 technology of the future. Um, I probably get at  least 100x ROI on the money that I spend and I'm  
24:52 spending like $200 a month, right? So that's  that for Claude. Once you're done signing up,  
24:57 um, what'll happen is we can go back to  anti-gravity, go to this bottom lefth hand  
25:01 tab here called extensions, and then click cloud  code for VS Code. Now what this is is just like  
25:06 the agent chat on the right hand side is the way  to communicate with the Gemini series of models.  
25:10 This is built by Google. Okay, this is built by  Enthropic and it's just a little extension chat  
25:14 that you can put into in gravity. And so what I've  done here is I've already installed it, but you'll  
25:18 have to install it. When you get it installed,  what you can do is you can click on this little  
25:21 button in the top right hand corner. That'll open  up a clawed code window just like this. Okay. And  
25:26 as you see, very similar UX to what we had earlier  with the agent chat. Just this time now it's  
25:30 sort of like cute and orange and there's a little  space invader over here. So really quickly to run  
25:34 you through it, if this is the very first time  you've done this, you do need to go /lo. Okay,  
25:39 that'll allow you to log in to your Claude AAS  subscription which you just put together a moment  
25:43 ago. You can also sign in through the anthropic  console or these other thirdparty providers. Most  
25:46 people for cost purposes are going to choose this  as the cheapest one. After that, you just write  
25:50 a message like, "Hey, what's up?" Now, just like  you were communicating a moment ago with Gemini,  
25:54 now what we're doing is we're communicating  with a different model. And you can find out  
25:58 which model by clicking the bottom right hand  corner and then going to switch model. And so,  
26:01 right now, I'm using this default recommended  one, Opus 4.6, most capable for complex work.  
26:06 There's also another one, Opus 1 million context.  There's Sonnet, Sonnet 1 million context, Haiku,  
26:11 and then uh some custom 4.6 model down here.  I think if you guys are keen viewers, you'll  
26:16 probably say, "Well, wait, Nick, didn't we have  the ability to communicate with the Claude series  
26:20 of models down here?" And the answer to that is  yes. Uh for whatever reason, I guess just because  
26:24 of the nature of their business arrangement, um  inside of the agent chat and anti-gravity, Google  
26:29 actually allows you to communicate with the Claude  series of models as well. But in my experience,  
26:33 and I don't know if this is because they're  trying to handicap it or something like that,  
26:35 but like in my experience, it just doesn't work  as well inside of that agent chat. And the UX is  
26:39 actually much better within the Claude chat.  So, what I typically do is I just communicate  
26:42 with the Cloud series of models in Claude Code  and then I communicate with the Gemini series  
26:46 of models in their built-in anti-gravity agent  chat. And that works really well for me. Um, also  
26:51 it doesn't it like simplifies my payments. So, I'm  like paying one provider for one sort of model and  
26:55 then another provider for another sort of model.  Obviously, your mileage may vary with that. And  
26:58 maybe it's just the vibes and vibe coding that are  getting to me. I much prefer to communicate with  
27:01 Claude in the Claude code extension. Anyway,  hopefully it's pretty clear. Same thing. You  
27:06 have little text windows, right? This is its  reply down over here. Just like in um Gemini,  
27:11 we had a few modes like planning and so on and so  forth. In this one, we have um ask for edits, edit  
27:15 automatically, plan mode, and then also bypass  permissions. Uh most Gemini is automatically  
27:20 set to bypass permissions now. So, you don't  actually have to like manually set it. Ask  
27:23 before edit is just where if you want to change  the pre-existing files on the lefth hand side,  
27:27 you need to just like verify with me first. So, if  I say change readme.md to include a snippet about  
27:35 Nick Sarif, okay? And then I ask it to do this,  what it'll have to do is it'll have to go find  
27:40 some information about Nick Sarif. Then it'll have  to read it. Then it'll have to make some proposed  
27:43 adjustments. But do you see how now we have this  little panel that's popping up saying, "Hey, do  
27:47 you want to make this edit?" So I could say yes.  I could allow all edits this session. And I could  
27:51 say no. Or I could tell Claude what to do instead.  I'm just going to escape out of it. That's how Ask  
27:54 Before edits works. Most people are going to use  edit automatically because then you can change  
27:58 pre-existing files without actually having to like  do that little, you know, hey, should I change  
28:02 this? Uh because nowadays these models just move  so much faster than human beings ever can. There's  
28:06 just no need to. And then finally, you have plan  mode, which is similar. It just does a lot of  
28:10 thinking. Doesn't actually do anything until  it presents the plan to you. And then finally,  
28:13 a bypass permissions, which is uh you know my  yolo mode where I just say do it again and then  
28:18 it's not going to ask me for anything. It's just  actually going to go edit the file. Right. So  
28:21 now it says built by Nick Sarif. I'm a software  engineer and automation consultant specializing  
28:25 in AI powered tools and web applications. Thanks  so much. The software engineering title has never  
28:30 been easier to get. Okay, everything else you  are going to learn by doing. Hopefully, this  
28:33 has been a comprehensive setup and you guys now  know how anti-gravity works in conjunction with  
28:37 the agent chat and then the Gemini 3.1 Pro series  of models. You also know how anti-gravity works in  
28:42 conjunction with Claude Code and then the Claude  series of models. And you also understand the  
28:47 interface, at least the 8020 of the interface that  we could use to build our very first project. Next  
28:51 up, what we're going to do is we're going to build  a highquality, simple portfolio site in just a few  
28:56 minutes that I'm then going to push live onto the  internet. I'm doing this because I want to show  
29:00 you guys how simple and easy it is to actually  develop like actual web applications and web  
29:04 properties with Gemini 3.1 Pro High. And I'm also  going to include all the prompts and everything  
29:08 you need to do the same thing on your end down  below in the description. So, just check that free  
29:12 folder out. Okay, so the simplest and easiest way  that I have found to set up websites with Gemini  
29:17 3.1 Pro high is you literally just provide it a  simple copy paste prompt which I have down below  
29:24 in the description and you guys can use as your  end all be all or you guys could use as a base to  
29:28 make your own prompt on top of. Okay, store that  as a gemini.md and then just fire it off. So, what  
29:34 I've done here is I basically open up a new folder  called website builder-demo with the gemini.mmd  
29:39 with this big long prompt. And I'll run through  the prompt in a second while I'm actually creating  
29:43 the website. But then I say build me a website.  Now, what this does is this forces the model to  
29:51 act as a world-class senior creative technologist  and lead front-end engineer. Basically, we're  
29:55 saying you build highfidelity cinematic 1:1 pixel  perfect landing pages. Every site you produce  
30:00 should feel like a digital instrument. You must  eradicate generic AI patterns, including typical  
30:05 structural layouts like a standard hero, three  column features, and CTA. Every site you build  
30:09 will use a vastly different structural layout,  typographic scaling, and interactive paradigm  
30:13 based on the chosen archetype and structural  typology. Kind of a mouthful, right? Now,  
30:18 I'm not going to run through all the rest of  these, but basically, to make a long story short,  
30:21 this section here is just guiding it through a  brief little Q&A so that you can choose, you know,  
30:26 the highest quality website design for you. This  over here has a bunch of different aesthetic  
30:30 archetypes depending on what you're looking at  which we can select. Down below we have some  
30:35 structural typologies including bento grids,  split screens, infinite horizontal structures,  
30:40 linear narratives, and then application shell  which is kind of neat. Then we have some brief  
30:45 little rules about different tools to use just  based off of my own understanding of tooling. Um  
30:50 as well as um some double-checking protocol down  here. And on the right hand side, as you can see,  
30:55 after I entered that little uh message, it's now  asking me for the brand name, the core thesis, um  
31:00 aesthetic archetype, structural typologies, three  core pillars, and then the ultimate conversion  
31:04 goal. So, all I'm going to do now is I'm going  to close out of this. Okay. Make this way way  
31:09 bigger just so I could see like most of it. And  then I'm just going to enter my information. So,  
31:15 uh my brand name, just so people are all on  the same page as me, is called Leftclick. And  
31:19 basically what we are to zoom way in there because  the website looks like trash right now is we're  
31:23 the definitive AI growth partner for fastmoving  B2B companies. This is a website that I actually  
31:28 had a real developer like I paid somebody money  to put this together for me cuz I didn't have  
31:32 the time. And the whole idea is we're like an  AI automation boutique shop that helps people  
31:37 generate revenue using high growth systems. This  is typically stuff like cold email systems. It's  
31:41 stuff like speed to lead systems for paid ads  and so on and so forth. So it's pretty neat,  
31:45 right? It's pretty sexy. We got the cool photo  with Hormosi and Ovens over here. That's uh won me  
31:49 some goodwill for sure. Basically, what I want to  do is I just want to have AI try and generate uh  
31:53 this website instead of me. And then I just want  to compare and contrast the results. You'll see  
31:57 there's some mild things here like there's a cool  little animation that kind of follows my mouse  
32:00 around. That's kind of neat, right? Obviously  the text and the fonts and stuff that those are  
32:03 very sexy. So, do you think AI could do as good  of a job? Well, let's find out. I'm going to  
32:07 answer some information here. Brand name is called  Leftclick. We help high growth B2B companies scale  
32:14 revenue using outbound methods like cold email.  Two, pick best fit. Three, pick best fit. Four,  
32:23 super low cost of acquiring a customer. Um, AI  based personalization, which lets us attack at  
32:31 both scale and high levels of perceived  personalization. And then trusted vendor.  
32:41 We've worked with multi-billion dollar portfolio  companies. And then ultimate conversion goal is  
32:48 book a growth mapping call 30 men consult. Just  going to enter this information now. And I'm going  
32:56 to have the model do everything else for me. The  first thing that's going to pop up is this little  
33:00 thinking tab. And so what it's doing if I scroll  to the very top is it's focusing on refining the  
33:04 initial approach particularly concerning file  operations. It's prioritizing the use of more  
33:08 specific resources to manage and manipulate  information and so on and so forth. These are  
33:12 all of its own thoughts. And basically the way  that this works is at every step it's basically  
33:17 like another it's like another run. It basically  takes the prompt, it sends it, it does a bunch of  
33:21 thinking and then it sends that back to another  instance and says what's next over and over and  
33:25 over and over again. So that's what's happening  down over here at the bottom. Once you're done  
33:29 something that requires a fair amount of planning,  it'll actually pop open a planning uh section. And  
33:33 this basically runs through the files that were  edited as we saw earlier. Okay. the progress  
33:38 updates of that, some additional thinking, and  then finally a task list for what it's planning  
33:43 on doing. When tasks lists pop up, they pop up in  the middle over here. And as you can see, we have  
33:48 a leftclick dashboard landing page. It's going  to start by scaffolding the application using  
33:51 a particular library. Then it's going to install  dependencies, configure Tailwind, then it's going  
33:56 to build a core layout, and then, you know, go  through and do things in a really high quality  
34:00 way. Okay. So over here, what it's doing now is  it's running a terminal shell. Terminals just look  
34:05 like this. uh running background command. Anytime  you see anything like this with this little  
34:09 terminal icon, it's basically just doing the thing  in the terminal for you. Um you can actually just  
34:13 jump to that anytime you want to. And here you can  actually see like what it's communicating with,  
34:18 what it's doing. You can also see the artifacts.  So there's now two artifacts that it's created.  
34:22 One's called uh the task and the other's called  the implementation plan. Both of these here are  
34:26 basically just resources for you to make things a  little bit more visible. Like logically speaking,  
34:31 this model is doing a lot under the hood, right?  writing hundreds, thousands of of words basically  
34:35 every few seconds. Um I have no insight into this.  This is all just occurring behind the scenes.  
34:39 And so one of the big issues right now with vibe  coded apps is there's just no interpretability or  
34:43 accountability about like what the model's doing  underneath um you know behind everything. Well,  
34:47 this implementation plan, the whole idea is it  just like shows you behind the scenes what's going  
34:51 on. Well, it's decided on some AR aesthetic and  typology decisions like um B2B SAS dense. We want  
34:57 gunmetal safety orange steel. The typography is  going to be IBM Plex suns, IBM Plex mono, right?  
35:03 These different fonts and so on and so forth. The  application shell is going to look like this. The  
35:07 proposed changes are going to look like that. As  you saw a moment ago, we just added a file to our  
35:11 folder called leftclick- dashboard. If I give this  button a click, you'll see we now have a bunch  
35:15 of files that it's written. The model just wrote  all of this while I was just talking to you guys,  
35:19 and it's currently writing more, right? Which is  pretty neat. I don't know how any of this stuff  
35:23 works, to be honest. And I don't need to anymore.  I just vibe code. If I like it, fantastic. If not,  
35:28 I I'll ask for changes and I'll show you guys  what that loop looks like in a second. But yeah,  
35:32 I mean, you know, as you guys see, a big chunk  of this is just shephering AI and making sure  
35:36 that it's operating in in the right direction, the  direction that you want, as well as looking over  
35:40 things like implementation plans, intermediary  files that are created, and stuff like that to  
35:44 just ensure that like everything is going the way  that you initially planned. What I'm going to do  
35:47 next is I'm going to go grab a sip of water and  then I'll come back and hopefully this website  
35:50 will be ready to go. Okay. And here is the result.  Deploy the growth engine. We build growth systems  
35:55 for high- rev B2B agencies. Pure information  density, zero latency scaling. Wow, that's  
35:59 cool. Ooh, I love that button. That's really nice.  Uh, if you guys watched one of my recent videos on  
36:04 website design, this is the same approach. So,  I actually really like the background image as  
36:08 well. And so, I uh turned I basically gave it the  ability like the prompt that you guys are going to  
36:12 use if you copy the prompt and use it in your own.  Basically gave it the ability to like find stock  
36:17 image websites and stuff like that. And that's how  we end up with like this really nice clean like  
36:21 glass ISO form thing up at the top. then have some  cool uh you know sections here. Rapid revenue,  
36:28 lowest cost, proven vendor. That's really cool. We  focus on systemic growth. I like that. I like the  
36:32 blue this time. We got some cool animations and  stuff like that. You will find, you know, because  
36:37 we're using the same icon libraries and stuff,  some of these animations will be pretty similar.  
36:40 If you redo this five or six times, but yeah, I  mean like we just we just one-shoted a really cool  
36:45 website, right? Um I'm just going to do the same  thing now a couple times just to show you guys  
36:48 that you don't have to stick just to the style.  You could do it in a bunch of different ones,  
36:51 too, if you wanted to. Okay. So, I'm just going  to open up a new folder here. And then I'm just  
36:54 going to make one website builder new. And then  I'm just going to operate inside of this folder.  
37:00 Okay. This is another rendition. So, I think this  is supposed to be like kind of earthy and healthy,  
37:05 which is why it says leftclick with the globe.  That is definitely not our brand, but what are  
37:09 you going to do? We're the architect of outbound  scale. We weaponize personalization by designing  
37:14 and deploying outbound architect for multi-billion  dollar portfolios. Kind of really like that  
37:18 section too much. Here's probably the weakest bit,  though. This stuff looks really cool. Like like I  
37:22 I really like this. Look at that. This looks nice  as well. You know, it's like something it came up  
37:27 with to showcase how we're verifying decision  maker profiles. We're trusted by the best. You  
37:31 know, we're architecting the revenue engine. So,  this is a great example for me to show you how  
37:34 you can change websites to make things better  or uh I don't know, you know, include sections  
37:38 that you want. So, what I have here is I have a  voice transcription tool. And you know what? I  
37:42 I I don't really like I don't know, I just don't  really like the layout here. I think personally  
37:46 the navbar is kind of an issue. So, what I'm  going to do is I'm going to go and I'm going  
37:50 to see if I could find a navbar from something  else and I'm just going to include a screenshot.
37:57 So, like this navbar is actually really  cool, right? Like this this is clean. So,  
38:00 why don't I just take this navbar from my former  site? Okay. And then why don't I also just create  
38:06 a big voice prompt about all the things I want to  change. Hey, I have a navbar that I've supplied  
38:11 you with. I want you to make it that sort of  pill style because I think it's really high-end  
38:15 and clean. I don't need the Earth logo next to  left click. You can just use uh left click. Um,  
38:20 we should have buttons on the hero header. Also,  just make the hero header way bigger. I want it to  
38:24 be like huge. I want the content in there to  just pop and and explode. There's some weird  
38:29 numbers next to the hypereefficient acquisition  graph. Um, they're outside of the bounds of that  
38:34 little box. So, let's fix that up. Um, let's add  another section between scale without compromise  
38:40 and trusted by the best. That just talks about  me. My name is Nick Sarif. I run Leftclick. I've  
38:46 helped uh I don't know over 400 500,000 people  now use AI automation and then outbound methods  
38:53 to improve their lives and their businesses.  Um you can do a quick little Google about me as  
38:57 well if you want. And then down at the bottom, um  let's just add a footer in addition to the CTA so  
39:03 that it's just a little bit thicker and there's  more going on. Okay. And then I'm just going to  
39:06 dump that right here. And I'm just going to dump  that right here. So boom. Let's just remove that.  
39:12 Uh and then Yeah. And then I'm going to paste.  I guess I got to paste the image, right? Yeah.  
39:15 There you go. Cool. So, now that I've I pumped  all this stuff inside, um I'm just going to wait  
39:19 for it to finish the change and then I'm going to  double back. Meanwhile, I designed another website  
39:23 simultaneously uh for my content writing business.  Once I can copy. So, I like this. This is really  
39:28 clean. We then have that little pillow up top  with similar three card design here with some  
39:32 text. That's neat. Uh we also have these cards  that kind of come up. I really love these cards.  
39:37 These are really clean. And then down over here at  the bottom, we have some pricing. So, that's nice.  
39:42 Anyway, I just wanted to show you guys that while  we're waiting for the other leftclick site. Okay,  
39:45 and I can see it's saying the dev environment  is still running. So, I'm just going to copy  
39:49 this. Go back here and paste this in. And this  is way bigger. Look at that. This is huge. Nice.  
39:56 This is significantly better. I'd say that's  about the size that I wanted. And then up here,  
40:01 I think it's basically just one for one copied the  font. So, I'll just make one more change, which  
40:05 is to make sure that the style of this aligns with  the style of the site. Hey, just want to make sure  
40:08 the style of the pill navbar aligns with the rest  of the style of the site. Super clean. Nice job on  
40:15 everything else. Oh, and I'll supply you an image  that you could use. Um, for me, that will help.  
40:21 Really like the footer as well. It's pretty clean.  Let's make the green just a little bit darker,  
40:26 actually. Um, that's probably the last thing that  I want to change. Everything else here looks nice,  
40:30 though. Man, there's a lot of images of me out  there. We'll go large. What makes sense? Which  
40:34 one makes me not look like a seven-year-old?  probably this one. Let's just do that. So, I'll  
40:38 go here. I'm going to copy both the image and the  agent, I think. So, we'll just dump that in. So,  
40:45 I'm just going to copy this here, and then I'll  paste this in. Okay. And then, yeah, I should be  
40:51 good. And this is the finished product. Uh, what I  like is when I load it, there's that little bubble  
40:56 that goes out. You know, it expands at scale. Uh,  navbar is looking nice and clean. When we mouse  
41:00 over it, it's green. We have everything that I  asked for, including a picture of myself with  
41:04 like slightly darker green. And yeah, I mean,  like, you know, is this again going to win any  
41:08 awards? Probably not, especially since website  design is so commoditized these days. But now  
41:12 that I've showed you how to build sites, not just  one site, multiple sites, by the way, and probably  
41:16 a hundth of the time it would have taken you to  do uh websites the old school way. And I know  
41:20 that cuz I used to be a website designer. I used  to get paid for stuff like this. Melts my heart.  
41:24 Let me show you how to deploy it and how easy it  is to actually get a website like this up live on  
41:27 the internet uh with Vibe Coding. So, basically  what you're going to need is you're going to need  
41:31 a hosting provider. Now, I use Netifi because it's  simple. It's uh free for a couple of sites anyway,  
41:36 but you can also use Verscell and there are a  few other ones that are like this. Basically,  
41:40 what happens is they just take these sites that  we just put together. Okay, so whatever this,  
41:44 you know, link is here that I'm opening up  and then it takes the static files on your  
41:50 computer that you're currently accessing at  localhost 5173 and then it just like puts  
41:54 them on a web accessible URL. So, we're going to  have something like gentlebeaver uh.netlifi.com  
42:01 and we don't need to spend any money in order to  get that done. It's completely free for us. Um,  
42:05 it'll only cost us money if we require like a  custom URL like leftclick.ai for instance. So, I'm  
42:11 going to go over to netifi.com. Uh, and then I'm  going to, you know, if it's your first time there,  
42:15 sign up. I'm just going to click login. And I  have a bunch of different projects down over here,  
42:18 um, that I've hosted on Netlefi before, right?  So, make money with make reply robot leftclick  
42:23 go one second copies. You can see I've had quite  a few startups that have failed. Uh, and what I'm  
42:29 going to do now is I'm just going to click add new  project. And then uh you know I'm going to click  
42:34 deploy manually and then I can just drag my drop  drag and drop my project folder here. So simply  
42:41 put I'm going to go back to my um anti-gravity  and then what I want is I just want to rightclick  
42:45 this and say reveal and finder. I'm just going to  drag this whole file in. Okay, this whole thing.  
42:51 And what it's going to do is it's going to try  building and deploying. Now the probability this  
42:54 is going to work the very first go by the way is  pretty low actually because we're not giving it  
42:58 any additional settings or anything like that. So,  I mean, I'm going to look for it, but I doubt that  
43:02 it's actually going to oneshot it. If it does,  I'd be quite surprised. Wow, that's so funny. It  
43:06 just did. Uh, that's hilarious. I wasn't expecting  that. I was going to expect communicating with an  
43:10 agent to get it done. But anyway, yeah, we just  drag and drop the folder in Endearing Salyaki.  
43:14 And as you can see, our whole website is up. It's  actually up on the internet now. It's not up um,  
43:18 you know, where it was before. So, I'm just  going to open an incognito window just so  
43:21 you guys are 100% confident that it is on the  internet. I just accessed this. Why don't I um,  
43:26 load this up on my phone, too. endearing. Okay.  And I don't know if you guys could see this,  
43:32 but like that's the website. It's just on my  phone now. It's on a totally different um you  
43:37 know device. So yeah. Hey, I'm uh Nick Sarif and  today I'm going to show you how to get a website  
43:43 up on the internet in like a 45 seconds flat. Kind  of neat, huh? So this is more or less going to  
43:49 be the approach that I'm going to use to push all  of our apps to production just because, you know,  
43:52 I don't want us to have to spend a ton of money in  order to like actually run server infer and stuff  
43:55 like that. But yeah, in general, like you know,  if you're using a predictable library like this,  
43:59 it's pretty straightforward. Obviously, not every  single time is going to go that easily. And I'm  
44:03 going to show you guys what like more complex app  pushes look like. In general, what I do now is I  
44:07 just give my agent um an API key to the service  like Netlify. And then I just say, "Hey, you know,  
44:11 push this." And then it'll actually go through  and then do all of the work. Um but hopefully  
44:14 you guys see how easy it is to oneshot a website  with Gemini 3.1 Pro. Cool. So that's how to design  
44:20 a website. The issue with websites is websites  are really simple. They have no authentication.  
44:25 They have pretty cool UX as you can see and you  can get progressively more complicated as the  
44:29 models get smarter. Um, but there's no like actual  like like security or or data transmission that's  
44:34 really required, right? There's no interactivity.  So when we migrate from uh you know websites to  
44:41 apps, what we need to do is we need to understand  a couple things about security. And if we don't  
44:47 understand things about security, what'll end  up happening is the same thing that happened to  
44:50 some pretty popular libraries. You guys might be  familiar with like OpenClaw pretty recently or a  
44:54 Claudebot is what it was called before its 40th  rebrand. And the reason why part of the reason  
44:58 why I rebrand it so much is because there's just  so many security problems with it. Like people  
45:01 had 25,000 API keys leaked. API keys being like  basically the equivalent of like a password to  
45:06 a service. That's pretty important. Uh not to  mention, you know, API bills that were run up  
45:10 because hackers and stuff like that got access to  it. Or just like leaked data from users that use  
45:16 a specific thing. Never a good situation to be in.  So, I don't want to pretend like I can solve 100%  
45:21 of all security wos for you. I can't. The reality  is everything on planet Earth is hackable. It's  
45:27 just a question of how hackable. And the whole  idea of like 8020ing security is just to get  
45:32 your point get to a point where your app is less  hackable than the amount of effort that it would  
45:37 require to do it. And so almost like anybody  could theoretically break into anybody's house,  
45:43 but we don't because there's just like a little  bit more effort is required to like, I don't know,  
45:47 break my window than just walk into my front door.  Um, if you just like do some 8020, I don't know,  
45:52 put a fence in, have some camera outside your  front door or something, you'll eliminate the  
45:55 vast majority of like lowhanging fruit. Okay,  so I just want to run through the 8020 here. 80%  
46:01 of security problems in vibecoded apps will come  from five things. The first is exposed environment  
46:06 variables and API keys. The second is missing or  broken rowle security RLS on your database. The  
46:12 third is no server side validation which is  where you trust the front end for all of the  
46:16 validation steps. So verifying that a password  is a password and email is an email. Verifying  
46:21 your logic and your math and stuff like that. The  fourth is using outdated or hallucinated packages.  
46:25 Very important. And the fifth is not having what's  called proper authentication middleware. Now again  
46:30 you don't need to be a computer programmer to get  all this stuff. I'm going to explain it while we  
46:33 talk about it. What's really important is when  you actually go through and develop, you know,  
46:37 like a like a database for the most part, you just  like click the RLS button. Um, when you set up an  
46:43 app, you just tell Gemini 3.1 Pro High or, you  know, Claude, hey, you know, I don't want you  
46:48 to do server side validation. What's important  is that you just have it run through all of its  
46:52 packages afterwards or say, hey, do we have proper  authentication middleware on every page? You know,  
46:57 hey, are our environment variables and API keys  exposed publicly? And so I guess what I'm trying  
47:02 to get at is it's actually not that difficult to  like hit the AD20. You literally just like create  
47:05 one prompt and then you just feed it in as sort  of like a postbu workflow similar to what I showed  
47:10 you with the code review initially. So we can just  pass all the stuff in as a prompt and I'm going  
47:14 to do that and give you guys the prompt later and  it'll do pretty good. But obviously you know it's  
47:19 not perfect. No security ever is. This should just  allow you to launch apps without feeling like a  
47:23 fraud or needlessly endangering other people's  credentials. So let's chat about the first  
47:26 which is environment variables and then secret  management. For those of you guys that don't know,  
47:30 in computer programming, you typically have to  deal with the equivalent of a password. And so  
47:34 we call these passwords various things, tokens,  authentication credentials, environment variables,  
47:39 API keys, whatever. But they all sort of like come  to the same idea, which is that it is something  
47:43 that if uh a user has access to it, they can  typically access a bunch of features of their  
47:49 account and not somebody else's. And so the issue  with vibecoded apps is a lot of people with their  
47:55 agents actually hardcode authentication tokens  and secrets and environment variables into their  
48:02 code. And the way that I see it is this is kind of  like putting a sticky note with your password on  
48:06 the front of your monitor like you see in a lot of  like older school cubicles where it's like Gmail  
48:11 password 1 23123 Nick is super sexy.com whatever  and like anybody that walks by will see that and  
48:17 they will know your password. It's kind of like  that with um exposing environment variables and  
48:22 secrets inside of like a a front end. So, also  a lot of hackers will go through the website and  
48:28 app code and then just look for hard-coded  mentions of a specific string like next_pub  
48:33 uh you know whatever key or secret key or what and  that's a common string prefix. And what I mean by  
48:37 that is like you know if you load up the website  that we just had a moment ago which I don't  
48:40 actually remember the password I don't actually  remember the thing. So one sec. Yeah. Yeah. So,  
48:44 if you go through this website, we can actually  like right click, we can press inspect. And uh  
48:47 as we scroll through, we'll actually be able to  like command F different elements on the page. So,  
48:52 weaponize, for instance, is actually like  written in the code of the page. I guess you  
48:55 guys can't see cuz my fat head is in the way.  You guys see how down here I just command F  
48:59 weaponize and then we actually found the specific  segment. What if we do personalization? Right,  
49:03 that's there. How about initialize sequence when  I'm selecting this element? So what I mean by uh  
49:09 exposing your credentials is basically a lot of  the time somewhere on the front end people will  
49:13 write code so that when you click the map growth  button up here which we just found down over here  
49:18 um there'll be some additional thing like like  I don't know on click equal and I don't even  
49:24 remember like the actual function definition  for this but uh whatever you know fire my API  
49:29 key you know to whatever the website is and  then I actually like expose my API key which  
49:36 is right over here and so basically we is we just  want to avoid situations where I actually have,  
49:41 you know, on click fire my API key to this  location and here it is in plain text. Um,  
49:47 we also want to avoid the loca uh situation where  you have something like next public key equals  
49:52 whatever because if you have either of what I just  mentioned, somebody can just go on your website  
49:56 and then do the equivalent of like a command F  or control F and just type in like next public  
50:00 key and then they'll find something. It's really  dumb, but um you'd be surprised at how many really  
50:05 like high-profile libraries it's happened.  That's one of the issues that I've had with  
50:08 uh well anyway I'm not going to name any specific  names but yeah I've had some issues with some some  
50:12 recent vibe coded products because of this. So  what do we do instead? Well basically you put  
50:16 them in a file like av and then you just reference  them using imports or you tell the model like hey  
50:20 you know I want you to look in the specific place  for all of my API keys and credentials. And so if  
50:24 I show you guys a practical example just using  you know anti-gravity basically what you do is  
50:28 somewhere in the core of this you'd go env uh key  and then it would look like this and then no like  
50:36 um anthropic private key and then it would look  like this right and so basically what you do is  
50:42 you'd say hey what's my um enthropic key and uh  if you ever needed to use it to like construct  
50:49 something the model would instead of looking  through your your code base here and finding  
50:53 it like just written directly in a codebase.  It would do like a command f through the entire  
50:58 thing. It'd look for something like an enthropic  key or something like that and then it would go  
51:02 and it would actually find that it's 3 if4 and as  you can see it's analyzing the specific file to do  
51:06 so. So this is just like a programming convention.  Just keep all your passwords everything like that  
51:09 inside of av. Another benefit is you usually  don't push keys to like an a repository on the  
51:15 internet. Normally you just keep that locally on  your computer and then you give that to whatever  
51:19 like the app user or whatever to upload their own  keys. Pretty simple stuff. Not rocket science. The  
51:22 issue is just nobody actually does this. Nobody  actually asks their agent to make sure that all  
51:25 this stuff is done. The second is this idea of  rowle security. So um as you guys are going to  
51:29 see in the next project, we're going to use a  database and the database that we're going to  
51:32 use is going to called superbase. It's very very  popular with most vibe coder projects right now.  
51:36 And basically it's like the single most important  security feature on planet earth. But um you know  
51:41 these databases out there just don't enable this  by default because it's just not something that  
51:45 you know is enabled by default. I don't know.  I'm not a super in-depth computer programmer.  
51:49 but it changes a couple of things and some people  don't want to use a database like that. Basically,  
51:53 what this means is if you don't have this  feature enabled called RLS, anybody who has  
51:57 the Superbase URL and then what's called the anon  key, which is public by design, can read, write,  
52:01 and delete every single row in your database. And  that's just how Superbase works by default. And so  
52:05 some very popular vibe coded projects here did not  have level security. And basically anybody that  
52:09 signed up for them was given a key, sort of like a  password, and then they just use that password to  
52:14 search up all of the data about every other person  on it. And that led to like a massive massive,  
52:19 you know, issue. Try not to swear in these  videos, but it was a cluster. And uh, you know,  
52:24 that ended up with people's ability to delete  rows and databases, which was stupid. All RLS's.  
52:30 Basically says like, hey, user A can only see user  A's data. User B can only see user B's data. Like,  
52:36 obviously, it's not rocket science. Uh, but if you  just turn this on, you'll fix like 90% of errors.  
52:40 Okay, so 83% of exposed superbased databases  involved RLS misconfigurations. uh 83%. That  
52:46 means like only 17% of exposed superbase issues  were not related directly to this one thing that  
52:51 I'm telling you about that is basically just a  button. So we're just going to click enable and  
52:54 it'll be fine from now on. But just make sure to  know RLS is like the main lowhanging fruit. Next  
52:58 is something called serverside validation and um a  big problem with that is like you know how we just  
53:03 developed a website, right? Well, let's say on  their website there was some form and the form was  
53:07 um I don't know it was like a sign up form and it  said you know enter your email here and then enter  
53:11 your password and then let's say obviously in  order to implement any sort of sign up mechanism  
53:15 what we kind of have to do is we have to check to  see does the email already exists in our database  
53:18 and so what the server side thing what the front  end thing would do is it would basically say okay  
53:23 let's match the email to something in our database  well um when you do this sort of ser uh front-end  
53:29 validation where you validate like all of the  logic on your on your website um it's massive  
53:34 security risk because as you just saw a moment  ago, anybody can jump into the code of any website  
53:38 and then immediately change it. Like I could  go to google.com and then I could, you know,  
53:44 um hold command shift C or rightclick and then go  inspect element and then I could select anything I  
53:50 want on the page like this Google search button.  Then I could actually change it from uh I don't  
53:55 know this input class which is called Google  search and I could change it to nick search.  
54:01 And now I've just changed Google's website, right?  Accordingly, what I could also do is if there's  
54:05 some form here when I put my information in that  does my email or whatever, I could actually like  
54:10 have it go through a bunch of email addresses or  a bunch of passwords and then it would change the  
54:15 email and password that is being done the valid  accordingly. If I had a forum and this forum,  
54:20 you know, was like checking emails or checking  passwords or something, I could actually go  
54:23 through and then I could like see the data, see  the function that is actually doing the checking  
54:26 and then I could like look at it and I could be  like, wow. So the logic that that guy is using  
54:29 to check if passwords exist is he's checking to  see if it's over uh I don't know three characters  
54:34 and the maximum password length is 10 characters.  That means that uh you know I could theoretically  
54:39 enumerate through all passwords between three  and 10 characters and then I'd be able to find  
54:43 out anybody's password. You know it's like this  just gives attackers way too much information.  
54:47 Likewise, it also allows them to do things like,  you know, if uh a specific email address already  
54:52 exists in the database, what you could do is you  could use that same email address, but you could  
54:56 change the request that's being sent. You could  sign up a million different email addresses using  
55:01 some script. It's just not very scalable, I would  say. Um, and it's not very intelligent to do. So,  
55:06 instead, what we do is we actually do server side  validation. And what that means is like basically  
55:10 we have our site right over here. And our site in  general communicates with some server. And instead  
55:17 of doing all the validation over here, okay,  what we do is we say, "No, what I want to do  
55:23 is basically every time somebody tries to create  a new account, I'm not actually going to check  
55:26 on the site. What I'm going to do is I'm going to  send a request over to another backend somewhere,  
55:29 some server. The server is going to be much more  secure. And then because it's not on the site,  
55:34 nobody can like jump into the server and then  make minor changes to, you know, check out the  
55:38 logic and stuff like that. The only thing that  an attacker would actually ever see is just the  
55:42 fact that, oh, the site sent a request to the  server. uh but you know the server is like super  
55:46 password gated and you know requires specific  authentication. It's not anything that you know  
55:50 a person could actually use. So that's kind of  easy. You just ask your AI model, hey you know  
55:54 make sure to server side validate everything. Hey,  is anything here not being serverside validated?  
55:58 Okay, we'll make it serverside validated. And that  takes me to number four, which is dependency and  
56:01 package security. Um so when AI generates code,  sometimes it references packages that don't  
56:05 exist as in your AI will literally just make up  a package name. In case you guys didn't know what  
56:09 packages are. Basically inside of most folder  most coding uh projects nowadays you have like  
56:13 a node modules folder or something. And these are  all the services that this um repository is using  
56:19 in order to like build my website. And so it's  using witch word zod yto q prelude puny code  
56:26 react and so on and so forth. And so as you can  see like most websites nowadays are actually like  
56:30 very very heavily based on pre-existing libraries  that other people have built which kind of makes  
56:34 sense cuz like you know the longer the internet's  around the more people have built libraries that  
56:37 handle specific features and like obviously you're  going to want to just use somebody else's solution  
56:40 rather than build your own wheel and so that's  what a lot of people here are doing. Tailwind  
56:44 CSS for instance is a big design library that just  does a lot of the design for you. Arg parse just  
56:48 handles arguments and it's like a simple and easy  way to you know do bash scripts and functions.  
56:53 And so the issue is AI just came up with all  this stuff, right? But sometimes it'll actually  
56:58 download or it'll create a name of a module that  doesn't actually exist. So for whatever reason,  
57:04 I mean the training data or whatever or  just because of the way statistics works,  
57:08 maybe it'll create a folder here or uh whatever  called acorns and it'll try and download a library  
57:13 called Acorns and it turns out the library it  wanted to download is actually called Acorn, but  
57:17 instead it tries downloading Acorns. Now, people  that have done a lot of stat stats and stuff like  
57:21 that have like done a lot of this to see, you  know, what libraries the um AI models typically  
57:26 screw up on. And so, if it sees like, oh, you  know, this always says acorns instead of acorn,  
57:31 what it'll do is it'll actually these attackers  will go through and they'll make a big list of  
57:35 all these fake packages. And then they'll actually  create packages on the internet that correspond  
57:40 to these. And then these packages will just be  malware. So, instead of you downloading Acorn,  
57:45 now you're downloading Acorns. instead of it  being empty and being nothing because it's fake,  
57:49 now you're downloading somebody's package that  actually has like a string in it that says, "Hey,  
57:53 you know, when you download this package, I want  you to send all of the API keys to this service."  
57:57 They as like, "Oh, yeah, that makes sense. I'm  downloading this package. This package is legit.  
58:01 Let's do it." And then it sends all the API keys  to the service. So, I mean, like in practice, it's  
58:04 not anywhere near as big of a deal as just like  server side validation um issues or RLS, but yeah,  
58:10 I mean, like it can be a problem. Another issue  is AI will typically download like older versions  
58:14 and stuff like that if it's not constantly  instructed not to which means you might be  
58:18 pulling in a package with a security vulnerability  that you know actually got patched months or years  
58:22 ago. So in general the recommendation is ensure  you get the most updated packages as possible  
58:26 as opposed to like you know older ones. And then  finally this authentication middleware bit. Um in  
58:31 case you've ever accessed like an authenticated  service before typically the way that it works  
58:35 is you'll have some thing. So I'm just going to  open up a new thing and I'll go cloud.ai. Um,  
58:40 I don't know, slashnew. When I click on this, do  you notice how it just immediately jump me back  
58:45 to sluggin? Claw. New is like how you make a new  thread, right? Well, there are some apps out there  
58:51 where you'll pump in like claw.ai/ um I don't  know, maybe chat or something like that. Okay.  
58:59 And then basically you'll be able to act despite  the fact that you're not logged in, you'll be  
59:02 able to access chat if you just know that like  the URL is chat. Now obviously these guys are a  
59:07 major billion dollar maybe even trillion dollar  company by now. So every time I try and access  
59:11 this it immediately jumps you back to login. But  this is behavior that is handled by this thing  
59:15 called authentication middleware where basically  when you try and access routes that aren't your  
59:18 routes okay like settings and general and new and  and chat u it just knows that okay well you're not  
59:24 actually logged in so I can't show you that. I  need to get you back to the login page first.  
59:27 And so you know poor app design includes people  forgetting to put in authentication middleware um  
59:32 and then people have the ability to access pages  they shouldn't. And then sometimes it like allows  
59:36 them to manipulate things, rack up API usage uh  calls and stuff like that. So you can fix this by  
59:42 just ensuring middleware secures all routes. Okay,  so like I didn't want to make this little module  
59:47 here because it's boring as hell, but just know  that you know most people that build vibe coded  
59:51 apps will miss one of these five features and  then vibe coding in general just gets a really  
59:54 bad name because people just don't secure their  [ __ ] right? The reality is like nothing is 100%  
59:58 secure as I said, but if you just do these five  things and then um you know ask your AI models  
60:03 to fix them, uh you know, it'll do a reasonably  good job and I'd say about 80% of your security  
60:07 issues are going to be handled, especially that  rowle security thing. So what I've done is I've  
60:11 actually gone through and I've created a massive  prompt for all of you guys that you can use  
60:14 right over here. It's called the security audit  prompt. You just copy everything below this line,  
60:18 okay? And what you do is you give it to your model  and you say, "Hey, you know, I want you to check  
60:23 to see if all of this stuff is done." Then after  you're done, um, you know, flag all of your issues  
60:29 and then ask me what you want me to do about all  these issues. So, hypothetically, I mean, I'm just  
60:33 going to paste this whole prompt in, and I just  did to our website. Um, the issue with the website  
60:38 is obviously there's no real authentication going  on, right? So, it's not like it really matters too  
60:42 much. Uh, but still, you know, it's a reasonable  way for us to at least run through this. And now,  
60:47 uh, Gemini 3.1 Pro High, and I always recommend  using the smartest model for this stuff, is now  
60:51 just going to like take this prompt, pump through  literally everything, and then just say like,  
60:55 "Hey, this is good. Hey, this is bad. Hey, this  is good. Hey, this is bad." And then we're going  
60:58 to secure it by basically just saying, "Okay, fix  everything that's bad." Okay. And now it's given  
61:03 us a big returned checklist. It's saying that the  codebase is a classic static front end. Here's  
61:08 the systematic audit. No hard-coded secrets, no  API keys, and so on and so forth. No git ignore,  
61:15 which is something that they typically recommend  to um prevent you pushing yourv, no console log  
61:21 statements, you know, a bunch of things. And it  looks like, you know, we didn't actually have  
61:25 any to go off of for most of this. If you scroll  all the way down, it'll actually show you like  
61:29 here's the big checklist summary. Uh because this  isn't an app with any authentication, right? So,  
61:33 it's just like a website. And so, because of that,  only these checklist features are the only things  
61:36 that are relevant. But we did fail two. So, I'm  going to say is fix the two that failed. And then  
61:43 uh it's just going to go through. It's going to  add the git ignore. It's going to add a couple  
61:45 other things. You know, this takes time, right?  I mean, it it's going to take me some token costs  
61:49 as well. I'm going to spend some money on this  essentially. So, unsurprisingly, people don't want  
61:52 to do this and uh that people typically forget  it. But, you know, this is an 8020 pass. It's  
61:57 not going to fix everything, but it's going to fix  the low hanging fruit, which is to be honest what  
62:01 most people really just care about. When you vibe  code an app and you publish it and then it starts  
62:05 scaling and you start making some money from it  obviously then you can take some of that money use  
62:09 it to hire maybe an actual security engineer and  have them audit your project and stuff like that.  
62:13 Okay. And here you go. It did. So it updated this  to add the get ignore and then it updated this to  
62:18 um update eslint and resolve all four high high  severity mini match vulnerabilities which is  
62:23 cool. So yeah, looks like we are now solid. Let's  move on to something a little more interesting,  
62:27 which is taking everything that we now know about  how to build a beautiful front end and then how to  
62:31 do security audits and stuff like that and  then combine it into a full stack app. So,  
62:36 next up, we're going to build a fullyfledged  SAS app. The app that I want to build today  
62:40 is a client dashboard platform for a hypothetical  content writing company. And so, like any typical  
62:47 client dashboard, we're going to want to be able  to create orders, um, view pending orders. We're  
62:52 going to want to have like a nice sexy dashboard  which gives us an overview of everything,  
62:56 our stats and so on and so forth. Then we're  also going to want some sort of like settings  
62:59 or configuration page. Then on top of that, we're  going to need some sort of login naturally. And  
63:04 then I think at the very end, I'm also going to  produce a nice sexy landing page just to tie it  
63:07 all together. And then I'm going to push it uh  live to production as well, just so you guys could  
63:10 see how it all interacts. Before we actually do  the building, I just want to run you guys through  
63:14 what I consider to be the uh basically major vibe  coding design loop as of right now. And obviously  
63:20 there's a bunch of different ways to build apps.  And I don't want you guys to think this is like  
63:24 the the end- all beall. There are many many  different ways to do this. But in my experience,  
63:28 this is like one of the one of the good ones  right now. And if you follow this sort of flow,  
63:32 um you'll save yourself a fair amount of time  later on just in terms of like the design and  
63:35 then also security because um as I talked about  in the security module, there's a fair amount  
63:40 that goes into designing like a wellsecured  app. So the way that this works in general is  
63:44 we start with some sort of inspiration design.  And so there are a variety of different places  
63:48 you can go to for these inspiration designs, but  in general, I like going to um this website here,  
63:53 Dribble with threebs. I like going to this  one called godly uh website. And basically,  
63:59 there are a bunch of these sources on the internet  that compile really high quality designs. And like  
64:04 logically speaking, I'm not a designer, right? I  haven't spent all the time and energy necessary to  
64:09 know the right padding and the right color  schemes and all that stuff. So it's like,  
64:12 why would I worry about coming up with all that  stuff myself? instead I could just borrow somebody  
64:16 else's and then we can make some changes to it  and then once it's all changed and and and all  
64:20 nice and pretty. Um you know like we will have  taken the design fundamentals that somebody spent  
64:25 a lifetime putting together and then we could just  work off of that instead of spending god knows how  
64:29 long trying to figure out like the total amount of  margin on the left and the right hand side of the  
64:33 uh the elements. It's crazy stuff. So as somebody  that's like been doing a fair amount of design,  
64:38 I got paid for design. Uh it's not worth the time  and energy that it goes into learning all that  
64:41 stuff. So we start with an inspiration design. I'm  going to grab a couple of these. And then you have  
64:46 an optional step here which a lot of people are  starting to do now where you animate. So the idea  
64:50 is since Vibe coding now makes uh the ability  to like replicate designs and have high-end  
64:54 designs really really easy. You know you need  a way to stand out. And so the way that people  
64:58 are standing out now is through animations through  um like kind of videos that are playing. The idea  
65:02 being that you know a still can't make video based  stuff super super high quality. So you could still  
65:07 um stand out as like human created content or  uh designs with something like that. I'm going  
65:11 to show you guys how to do that. Anyway,  uh this does cost a little bit of money  
65:14 because there's some models out there that  you need that can convert images to videos,  
65:17 but I'll I'll run you guys through what that  looks like um for the landing page. After that,  
65:20 we need to build in the main functionality, which  is always going to be the dashboard. It's going to  
65:24 be things like, you know, your various widgets,  your various pages, and stuff like that. Then,  
65:28 we're going to do local testing and iteration.  Local testing, just so we're all clear,  
65:32 just means testing it on your device, and it also  means testing it with uh without a database. So,  
65:36 we're just going to do all the file and then, you  know, uh, data and stuff like that in like a JSON  
65:41 and then we'll just test it all locally, make sure  that it works. Once we've made sure that it works,  
65:45 we'll then push our data to a database. And then,  in our case, we're going to use superbase. I'll  
65:49 talk about that in a sec. But then, um, we'll  test it all basically with the database. Once  
65:53 we're done with that, then and finally then do  we create an authentication page. And I find um,  
65:57 doing O at the end is way better than doing O  at the beginning because if you guys remember,  
66:00 I mentioned like middleware, right? Well, like  a big issue with the middleware is you'll just  
66:04 like the low hanging fruit is you'll do the O  first, then you'll make the pages and then the  
66:07 O will only secure the first page that you do.  So instead, we're just going to do our O at the  
66:11 end and then the model will already have all  the context to like produce um you know good  
66:15 authentication middleware for all the pages and  then at the very very end we're going to deploy  
66:19 and then obviously the issue with you know testing  back here when it's sort of on your machine and  
66:25 then deploying is like you you need to test it  again. So we're going to do some end toend tests.  
66:28 We're going to do some miscellaneous designs and  I'll probably just throw in the landing page there  
66:31 since obviously the landing page doesn't really  have much functionality to speak of. And uh yeah,  
66:35 that's that's more or less the loop. And then  you know like when you're ready what you do is  
66:38 you just do the exact same thing. You just  make a new product and then you start with  
66:41 the inspiration design. Um you can do some video  animations if necessary main functionality local  
66:46 testing database testing off deployment and end  tests and then landing page. Then my inspiration  
66:50 design animate and so on and so on and so on  and so forth. So I do this on basically all my  
66:54 vibe coded apps and uh yeah works pretty well.  So the common vibe coding stack which I'm just  
66:58 calling the usual suspects here are going to  be a combination of Nex.js for the front end  
67:02 a superbase for the database and the off and  then some sort of serverless backend which  
67:07 um in our case is going to be netifi but you  can use netlifi versell and a bunch of these  
67:10 other tools but this is just what people mostly  use because it's just very easy to spin up. So,  
67:14 to make a long story short, in case you know, we  have people here that aren't familiar with the  
67:17 idea. Um, you know, when you sign onto a website,  obviously this is my really crappy rectangle. Um,  
67:23 and this is our browser with the three little  dots, maybe the URL bar and stuff like that. And  
67:26 you have like the app over here and I don't know,  some sidebar. And then here's some other stuff,  
67:31 you know, like what what you're seeing, what's  rendered here is is the front end. And um there's  
67:37 already different front-end languages, right? I  mean, like HTML is hypertext markup language. It's  
67:41 usually used to create the structure of the page.  If you have CSS, which is cascading stylesheets,  
67:45 that's usually used to like make the the  rendered elements really pretty. And then,  
67:50 you know, if you want functionality on top of  that, you will use JavaScript and, you know,  
67:54 TypeScript. And there's there's all these  different languages, right? Well, basically  
67:57 what happens is um using all these languages,  there's just so many different ways you could  
68:01 use them that all the developers were using them  slightly differently. Some were more efficient,  
68:04 some were less efficient. What people ended up  doing is they created uh front-end frameworks.  
68:09 Okay. Well, not even just a front-end framework.  NexJS is technically also like it's like a full  
68:12 stack framework but um they made frameworks which  just encapsulate all three of these into you know  
68:18 this thing next.js and so what this is this is  basically like a full way to manage your HTML  
68:24 CSS your JavaScript all these other programming  languages and then builds they build it in a  
68:28 very opinionated way that consistently produces  similar outcomes and you know follows a consistent  
68:33 and predictable pattern so that most developers  can just spin up a site really quick. This is  
68:38 doubly important in the age of AI where you know  when if things aren't standardized there's like a  
68:42 million billion different ways that an AI model  could make a site and because usually there's a  
68:45 lot less supervision when you're viing than um you  know like a traditional development process. A lot  
68:49 of these ways just like aren't really well thought  out. You know AI just picked the simplest lowest  
68:53 hanging fruit and went with it. So I recommend  using these frameworks and stuff like that just  
68:56 both to economize your work but also just to  make sites predictable. If you ever hand them  
69:00 off to anybody if you're doing these for clients  and stuff like that you know next.js is the way  
69:03 to go. Okay. So that's the front end. Um you know  Superbase is going to be our database. So, I mean,  
69:07 just think of it as like a big Google sheet with  like a bunch of different tables and columns. Wow,  
69:11 my drawing skills have gotten worse recently.  Uh, but to make a long story short, you know,  
69:16 the stuff that is visualized on our front end,  like our bar charts and our little widget apps and  
69:20 stuff like that. Obviously, we need to pull the  data from somewhere, right? So, it doesn't make  
69:23 sense to keep all the data like on the site cuz  it's always changing. What we do is we're going  
69:27 to pull from a database and then like use that to  actually go and and change features. Obviously,  
69:31 we're going to need a login screen, right,  with like a little user uh portal and then  
69:35 like a little password portal. Um, let's round  that off so it quits looking phallic. And uh,  
69:41 Superbase is going to help us with that as well.  And then finally, you need what's like called a  
69:45 backend, which is basically just like a server.  And this server is actually going to manage the  
69:49 interactions between our database, our front end,  our off, and and you know, a lot of stuff. So, you  
69:54 know, that's a little bit more cloud-based. You're  not really going to fully I can't really just draw  
69:58 something that makes sense there. But suffice  to say, all these four components are going  
70:01 to interact together really nicely. And then the  stack that basically all the tools use is Nex.js,  
70:05 JS superbase and then some sort of what's called  serverless um back end. Okay, so we could go  
70:11 into that or we could just build it and I'd much  rather just build it and then you guys could learn  
70:14 through the doing. So let's start with the very  first step which is finding an inspiration design  
70:18 and I'm going to immediately jump on dribble for  that. So dribble for those of you guys that don't  
70:23 know but just go on the website. It's basically  just a big list of like different designs that  
70:27 people have made and it's cool. It was it was  initially meant to be like a way for designers to  
70:32 create a portfolio and show off the work and stuff  like that, which is pretty neat. Now, you know, to  
70:36 be honest, I think a lot of people are just using  this to source AI designs. I don't know how much  
70:39 longer we're all going to have access to dribble,  but uh we'll see. And, you know, in our case,  
70:44 what we're looking for is we're looking for some  sort of like client dashboard, right? So, I'm just  
70:48 going to type client dashboard in here. Basically,  what you'll see is a bunch of people have created  
70:52 these really sexy client dashboards. So, I'm  just going to click on a couple of these. Um,  
70:57 yeah. So, I I already opened that one. So, let's  click on this. Let's click on that. This one's  
71:01 actually pretty nice. And then, hm, why don't I  open up? No, I don't really like that one. That  
71:05 one looks kind of dumb. This one here looks cool.  Okay. And as you can see, you know, I now have  
71:10 um like a list of, in this case, all clients on  the lefth hand side. You know, this sort of like  
71:14 nice gray pattern, and that's that's kind of cool.  I don't really like this one a ton, though. So,  
71:18 I'm just going to keep looking until I find one  that I really really like. Um, what you'll find is  
71:22 a lot of these are now videos as well. And there  are a couple reasons why they're doing videos. One  
71:26 of them is that they don't really like when people  screenshot them like I'm about to. That design's  
71:31 cool. Okay, this is reasonable. Um, this one's  pretty interesting. But the one that I initially  
71:37 picked that I just liked the look of was this  here, which is super simple and it's very like  
71:43 extensible and you guys can add on whatever the  heck you want, right? So, I think I'm going to  
71:46 use this. And uh, basically what we're going to  do is I'm going to create a screenshot of this  
71:51 and then I'm going to use the model to try and  replicate this as closely as humanly possible.  
71:55 And I have a prompt set up that I'm going to I'm  going to send it to and stuff like that. And then  
71:58 from this we're going to get like a really high  quality design. And then I think I'm also going  
72:01 to do the same thing for the landing page. Um you  know this design is different from this design but  
72:06 uh you know I'm just going to find a way to like  meld the two together. And then I think at the  
72:09 end we're going to have something that like has  the color scheme of this but it actually has like  
72:13 the design of this if that makes sense which is  ultimately what I'm going for. I like that design.  
72:16 I like this um I guess this is sans ser but I  really like this font. So yeah let's uh let's  
72:21 give it a go. So what I'm going to do first  is I'm gonna head over to anti-gravity. Okay.  
72:25 And then I'm going to open up a new folder. And  I'm just going to call this full actually let's  
72:31 go lowercase full stack client dashboard. That's  going to be my very uh first full stack app with  
72:38 you guys. And then what I'm going to do first is  I'm going to talk a little bit about the different  
72:42 types of models that we have available to us.  Remember how before we designed our landing page  
72:46 entirely using Gemini? Gemini 3.0 Pro is awesome,  but personally I find that actual like system  
72:50 plans and system designs are better done through  Opus 4.6 right now. So, what I'm actually going to  
72:54 do is I'm actually going to do the planning in  Opus 4.6. I'll show you how to do that. First,  
72:58 let me zoom in because this is pretty far out. And  then I'm just going to go to um plan mode here. If  
73:03 you just keep on cycling through, eventually  it'll get to it. It's the one with the little  
73:05 pause sign. And then I uh created a prompt here  that I just want to go through with you guys. So,  
73:10 the prompt says, "You're a world-class website  designer with 15 years of experience designing  
73:14 high quality, award-winning websites for Apple  and Dribble. We'd like to build a full stack  
73:18 SAS client dashboard app for our content writing  business. It's very important that you get this  
73:22 right. our career depends on it. This is a little  hack that I picked up from um I think it was like  
73:26 a former OpenAI dev or something, but they found  that when uh you make it really really important,  
73:31 like our career depends on it, the output quality  actually goes up just by a few percentage points,  
73:35 nothing major. And my prompt is not winning any  awards here. So, uh anyway, I'll take every win  
73:40 that I can get. As a client, we want the ability  to view all pending content orders and progress  
73:43 in a beautiful dashboard. We also want the  ability to create new content orders using a  
73:47 highquality but simple form. Let me just run this.  We want the ability to configure various settings,  
73:52 modes, etc. And we also want anything else.  Your usual SAS app like this would include  
73:56 the dashboard. SAS app should use superb. I said  superbase here. What I meant to say was superbase,  
74:00 but I was using a voice dictation. So, what are  you going to do? Next.js and then tailwind.css.  
74:05 And we will deploy using netlifi. I'll define the  UX via screenshots. So, don't worry about design  
74:11 particulars. The flow will be first we'll build  the main functionality, which are the parts I  
74:15 talked about above using a screenshot inspiration.  You'll take the design style from the screenshot  
74:19 and then use it to build all of the signed in  pages. Then we'll make it interactive with local  
74:23 data and ensure routes work. Then we'll add O and  DB with superbase. Then we'll make a landing page  
74:27 and finally do an endto-end sweep to ensure  everything works including off, middleware,  
74:30 security, testing, etc. Then deploy and retest.  Then I'm saying think extremely hard and generate  
74:35 a great plan. Once you've developed one that is  comprehensive, report back here and I'll supply  
74:39 some screenshots. That's how we can start the  design process. Okay, so I gave it a fair amount  
74:43 of context and you can see it's already done some  thinking, right? put the text stack and looks like  
74:48 it corrected my little issue there with superbase  phased approach right they want me to plan this  
74:52 out comprehensively there's nothing in the current  um working directory so it looked it couldn't find  
74:57 anything green field project and now it's actually  going to go and create the architecture and that's  
75:02 what's that's what it's doing right now that's why  it's in plan mode now I should note here in the  
75:06 bottom right hand corner you guys probably I don't  know if you guys could see that's pretty small we  
75:09 got this little lightning um what I've done is  I've enabled this thing called fast mode just  
75:13 because Opus 46 supports it and that allows it to  go two and a half times is faster approximately  
75:17 for I think like 3x the price. So little bit of  an arbitrage there. Um I think the price is going  
75:22 to jump up to 6x pretty soon. So it may not be  available when you guys are on it. But personally  
75:26 I find for building really comprehensive plans  awesome tool and an awesome model. Okay. And  
75:31 then uh basically what I wanted to do is I wanted  to like produce this big plan and then I'm going  
75:35 to pass off this giant plan over to Gemini who I  think is actually just a better designer. Okay. So  
75:40 once the uh thing is super comprehensively planned  um then I can just pass all this stuff to Gemini  
75:45 with the screenshots. But sometimes claude code  will actually ask you or rather Opus 4.6 will use  
75:51 cloud code to ask you questions and so that's  why I'm doing all this planning and stuff like  
75:54 that here just in case it asks me some questions  for clarification. Okay. So I'm going to press  
75:58 escape because uh it's basically like hey do you  want me to start? Then here it's going all the way  
76:04 to the top here. We want to build a full stack  SAS dashboard. Here's the tech step. Here's the  
76:09 phase one project setup in UI shell. And um I'll  be honest, like I'm I'm usually not reading this  
76:13 entire thing. What I'm doing is I'm skimming.  I'm just making sure there's nothing that just  
76:16 jumps out to me as stupid. Like what I want is I  basically just want to see, hey, are there way too  
76:20 many pages? Is this way too complicated? Is this  model completely misunderstood what I want it to  
76:24 do? And so what I'm seeing is there'll be like  an app page, there'll be um a dashboard page,  
76:31 orders, a new page, a settings page, notifications  page, and that looks pretty reasonable, right?  
76:35 There'll be a components page where it'll actually  store all the components. Library page. Okay,  
76:39 this looks pretty good. This looks pretty  good. Interactive local data. So after that,  
76:44 we'll define some types. Yeah, I don't even know  what the hell that really means to be honest, but  
76:47 that's okay. And then we have those actual pages,  which look good to me. All of these are pages  
76:51 that I would realistically want my app to have. Is  there anything else I'd want? H dashboard orders.  
76:55 No, I think I think this is fine. After that, we  have the superbase off. So here, I'll like set  
77:00 up the superbase, but that's okay. That's not for  now. That's later. We'll finally have the landing  
77:04 page which is going to be really nice and sexy and  then we'll have some key architecture decisions.  
77:10 Okay, cool. I mean, this seems pretty reasonable  to me. I don't think there are any problems  
77:13 here whatsoever. So, what I'm going to do is I'm  actually just going to copy this whole plan. Okay,  
77:18 this whole thing that this just created for me.  I'm going to paste it into this now. Okay. And  
77:24 then I'm also going to provide some initial  screenshots of what I want the dashboard to  
77:27 look like. So, I'm just going to command um you  know on a Mac you can command shift control. So,  
77:34 command shift control 4 and then I'm just going  to go select this. Okay. And that'll actually  
77:40 give me like a pretty high quality screenshot.  I'm going to feed this in. Okay. And so now we  
77:46 have the image of the dashboard and then I'll say  design using this style as inspiration clean and  
77:53 minimal. Okay. Okay. So, what I've done is I've  now split the work up between these two agents.  
77:58 I had um Claude Opus 4.6 design me the really  comprehensive plan because as I mentioned I  
78:02 think it's a little bit better at planning  at least as of the time of this recording.  
78:05 Then I'm going to hand off all of like the design  and then you know basically just like grunt work  
78:11 over to Gemini 3.1 Pro High, which I think is way  better at actually like wiring up the components  
78:15 and then putting together like that that really  high class design. And uh you know at this point  
78:20 all we really have to do is just wait for it to  build. And this is realistically because it is,  
78:23 you know, going to be a pretty comprehensive app.  This isn't something that's going to happen in a  
78:27 couple minutes. You're probably gonna be good to  like go step away, have a coffee, come back. By  
78:31 the time all that stuff's ready, this will have  given you some sort of result. But it's always  
78:36 good before you do that just to like wait for,  I don't know, 30 seconds or so because sometimes  
78:40 it'll give you a plan and then you can just review  the plan um and just double and triple check that  
78:44 it didn't do something super stupid or super  terrible. And I don't think it has. Everything  
78:49 here looks good to me. So, I think it can just  go ahead and commence the project. Cool. So,  
78:54 I'm actually going to go grab a coffee and I'll be  right back. Okay. So, now it's telling me it's now  
78:57 completed phase 2 interactive with local data. So,  we should now have a site up and running with some  
79:01 sort of local data. Immediately after I said let's  open it. So, it's opened it. I tried doing some  
79:06 initial debugging and stuff like that on its own.  It it didn't end up doing so because there was an  
79:10 additional error which I'm going to talk about.  And just on the uh I guess concept of errors  
79:15 when you create a full-fledged SAS app like one  that has like back and forth and stuff like that,  
79:18 not just a website where there's no interactivity,  you know, you're going to have errors. And so it's  
79:22 not about like will I screw up at some point.  Will the model get this wrong? Just a matter  
79:27 of when and then what you're going to do when uh  it screws up. So basically what I've done is it  
79:32 started the development server and opened this in  my local browser. So I just click this open and  
79:36 as you can see here we have the app set up, but  the app looks a little funky. We don't have any  
79:40 styles. Just looks like crap. So obviously this  isn't what we want, right? What I'm going to do  
79:44 instead is I'm just going to take a photo of this  and then I'm just going to send it back. And then  
79:47 I'm going say I don't think styles are loading  also because I'm pretty sure there was some issue,  
79:53 right? I initially got this problem as well where  it said missing HTML and body tags in the root  
79:57 layout. So I'm also going to copy this in. Also,  I got this error. I'm going to paste that in. So,  
80:03 you know, at least initially before you even  get to be able to play around with the UX,  
80:07 you may have some like initialization problems and  config problems and stuff like that. And that's  
80:11 okay. Basically, what you do, um, if I just show  you guys sort of how I think about vibe coding  
80:16 these things is it's almost like I don't know,  you're in a cave. Uh, have you guys seen like the,  
80:22 you know, mining meme where you're like mining  for gold? Okay, basically you're in a cave and  
80:27 every single time you run the thing, you're like  this fella here. You just get a little bit further  
80:32 and a little bit further and a little bit further  and you know this fell here can't see anything.  
80:38 It's just darkness in front of him, right? So  every time he mines, he goes home and he's like,  
80:42 "My god, I didn't make any progress today." But  obviously if he just continues going, eventually  
80:46 he gets to the point where there's a bunch of gems  sort of on the other side. And so that's basically  
80:49 what we're doing. We're just going to work through  this initial sludge with initialization errors by  
80:53 copying and pasting um any problems and then just  giving it directly to uh the model until we get  
80:58 to that little gold gem. Once we're here, we can  actually start making real changes to our app. Um,  
81:02 and all of that is reasonably well expected. Okay,  speaking of which, it just modified some of the  
81:06 CSS after I sent it this message. And as you  can see, we now have a little dashboard set up,  
81:10 one on the left hand side. So, we've already made  some some reasonable progress. I'm just going to  
81:13 dashboard, see if there are any issues. Okay, I'm  not seeing any issues at all. So, we have five  
81:16 total orders, one in progress, two completed.  This is just some fake data made up. So,  
81:20 you can see we can now mouse over each of these  three things or uh five things. SAS growth guide,  
81:25 Q3 newsletter, so on and so forth. So, these are  all the orders that we have submitted as clients.  
81:30 There's a little notification toggle. Um, if I  click on the orders page, we're now here. So,  
81:35 you can actually see the orders better. I'm just  curious if I click new order, what happens? Okay,  
81:40 so we're now on a new order page where I can  actually add info. So, I'm going to say this  
81:44 is an example. New order, content type, landing  page. Hey, I want to build a really cool, sleek,  
81:50 sexy landing page. And then, you know, I want some  stuff here. And then here's my requested deadline.  
81:57 I don't know. We'll say March the 1st. We want  to add SEO meta tags and then require approval.  
82:02 I'll click save order. Okay, cool. And it looks  like it's got the functionality. And so I can now  
82:07 actually add orders. If I go to settings, what  do we got? Demo user, Acme Corp, email address,  
82:11 cool preferences, email notice, weekly digest.  Okay, so these aren't working. At least the  
82:15 weekly digest isn't. But that might be intentional  just because we don't have um access to, you know,  
82:20 some sort of email. But yeah, I mean like to  be honest, like the the core functionality  
82:23 of the app's already made. So although the  design is still pretty shitty, I would say  
82:27 and obviously I'm going to want to fix that up.  Um core functionality is done. We basically have  
82:31 like I want to say a REST API, but we have like a  little CRUD app here. We can create, we can read.  
82:36 Um I don't actually know if we can update. Doesn't  look like we can go to the specific order page to  
82:41 update yet. So maybe I'll make a note of that. Um  but we can presumably also Oh no, I don't think we  
82:46 can delete either. Okay, cool. Anyway, this notice  page also has notice. That's kind of neat. Um,  
82:50 I think what I'm going to do now is I'm just going  to voice dump in all of the problems that I see.  
82:54 And then we're just going to go back and forth  until we can eliminate just the minor little  
82:58 errors like styles and stuff. And then once we've  tested and confirmed the functionality here, then  
83:02 I'm going to do a big stylistic upgrade. After  we've done the stylistic upgrade, I will uh then  
83:07 make this database friendly. And then once it's  on the database, then we can actually push. Hey,  
83:11 this is really well done. Nice job. When I click  on a new page, there's initially some there's a  
83:17 there's a brief initial load where it says loading  orders or loading settings or a loading dashboard.  
83:23 I'd like you to remove that. I don't want any sort  of loading um widget here because it just makes it  
83:28 look lower value. Instead, I want you to maintain  the structure of the page and just leave all of  
83:33 the various sections blank. So, for instance,  if I were to click on orders, rather than it say  
83:39 loading orders before it populates the rest of the  page, uh where it shows me the table and stuff,  
83:43 I just want you to present the table to me  immediately and then just leave it blank. That  
83:47 way, we could fill in the text um afterwards. Um  right now the navigation text is in light gray.  
83:58 I'll take a screenshot and I'll show you what  I mean. But on the orders page, for instance,  
84:03 um the text that says orders at the top lefthand  corner is very similar to the background color.  
84:08 So that's annoying. I'd like some sort of graph  on the main dashboard page cuz right now I don't  
84:12 have any. I'd like a way on the dashboard page  to click into the recent orders as well. Uh  
84:17 right now I am capable of mousing over them and  something changes, but it's just a style effect.  
84:21 I actually want the ability to click on them. In  addition to being able to create, I also want to  
84:25 um be able to click on each individual article  on the orders page and then view it. Right now,  
84:31 when I click view, it says either order not found  or it takes me to some error page, which is not  
84:38 what I want. The notifications page has a mark all  red button that is currently grayed out. When I  
84:44 click on the notifications at the top right hand  corner, that little icon, nothing happens. So,  
84:49 I'd like that to just take me to the notifications  page. There's also nothing that happens when  
84:52 I click on my little profile picture in the  top right hand corner. Under settings, when  
84:58 I toggle weekly digest, nothing happens. Email  notifications works, but the weekly digest toggle  
85:03 doesn't. There's a nextjs error here called param  property was accessed directly with params.id.  
85:09 Params is a promise and must be unwrapped with  react use before accessing its own properties.  
85:14 Nothing happens when I click on the little demo  user button in the bottom lefthand corner either.  
85:20 This is quite a promising start. We just have  a few adjustments that we have to make. Uh so  
85:24 that functionality is perfect. Once you're done  with that, let me know and then we can proceed.  
85:28 Cool. So I've just added all this context and  I'm just going to take a screenshot now of that  
85:32 issue that I talked about where the orders  text was um you know kind of gray. That'll  
85:38 provide some more context. I'm keeping this small  intentionally because every new pixel is like  
85:43 more tokens than I'm consuming, right? There's  some other minor stylistic changes and stuff,  
85:47 but for now I'm not going to worry about the  style. And the another reason why is because  
85:50 um once you have an app that's like up and running  like where all the functionality is right then you  
85:54 can just like open a new Gemini instance and not  have any context and then just say like hey here's  
85:58 my app I just want you to change a design and then  it can just go into the style sheets and stuff  
86:01 like that make it really sexy. Uh whereas if you  try and mix all of this stuff into one big long  
86:06 context window sometimes it can get messed up. So  you know I just want this specific chat right now  
86:11 to focus on the main functionality and then we  can worry about the design afterwards. Anyway,  
86:14 let me just make sure I didn't add in any  additional information. Cool. Cool. That looks  
86:19 good. Grade out. Okay. Why don't I do this? There  you go. Cool. So, now deliver that information.  
86:26 And just like um last time, uh I'm just going to  sit tight and wait for it to relaunch. Okay. It's  
86:30 telling me that it's now updated everything.  I'm just going to click accept all. And then  
86:33 we'll go back to the specific page. Let me click  on settings. Cool. We don't have any issues there.  
86:39 Um everything there looks fine. Oh, you know, I  think there's one thing I forgot to point out,  
86:43 which is this um notifications button. I think  that's going to be okay because we're probably  
86:47 just going to massively change the style here. But  yeah, that looks fine. Let's actually click on a  
86:51 specific order. Nice. This is now pushed all the  way to the left. The order um arrow looks really  
86:54 nice. I don't know if there's a way for us to  uh mark this as completed now, right? Okay,  
87:00 cool. It also says start work. That seems kind of  odd. I wonder why it says start work. Um yeah, I'm  
87:05 noticing that as well. So, I think we're probably  going to have to remove that. The model might like  
87:08 fundamentally misunderstand what I want. What I  basically want is I just want the client to have  
87:11 like a dashboard where they could see it all. Hey,  I'm noticing there's like a start work button. Um,  
87:16 just to be clear, this is a dashboard for a client  to look at the work that is currently being done  
87:21 by the agency. So, they don't have the ability  to start work or anything like that. Just go  
87:26 through and fix this. So, if there are any other  idiosyncrasies, um, those are taken care of. Okay,  
87:30 cool. We got that last change here. And it looks  like everything is now super clean. Um, no major  
87:35 issues whatsoever. If I click on this button,  uh, we only have readon access. Essentially,  
87:40 clients can't make any changes. they can only  download stuff and the HC is what initiates that.  
87:44 So that's what I wanted. From now on, what we're  going to do, rather from here on, what we're going  
87:47 to do is we're going to redesign this because  I think the design is kind of elementary at the  
87:50 moment. And I'm going to use that same approach  that we talked about earlier. We're going to find  
87:53 a really high quality pre-existing design and I'm  going to have Gem and I basically go through and  
87:57 then modify this to match. Um, and the idea is it  should be able to modify this to basically be like  
88:01 almost pixel perfect. Uh, we're just going to like  change the fonts and stuff like that. Okay. So,  
88:06 I'm going to go through dribble again. And I mean,  I like this idea of like a light mode dashboard.  
88:11 Uh, I really do. Uh, what we had before is kind of  mid. So, I don't want that. I do like the idea of  
88:19 like something dark on the left hand side, though.  Kind of break it up a bit. So, I'm going to see if  
88:24 I could find something that is dark. Let's take a  quick peek at this one. This one looks all right.  
88:30 You know, I really, really like this one. This  one looks super clean. Check that out. Okay. So,  
88:34 I'm a big fan of this. Super big fan of that. I  think what I'm going to do now is I'm going to  
88:39 take this screenshot. Okay, copy that in. And then  I'm going go back to Gemini. I'm going to paste it  
88:46 and I'll say, "Hey, this is great. I'd now like  to change the design significantly. You should  
88:51 do the vast majority of your work in Tailwind  or whatever stylesheets that you're currently  
88:55 using to build this." Um, in short, I think the  design right now is very basic and super simple.  
89:01 I want you to add significantly more  complexity um using things like cool  
89:06 mouseover effects on navbar elements on the lefth  hand side. I want you to add really highquality  
89:12 icons. I'd like you to modernize the design  significantly. I want you to try different fonts,  
89:17 higher quality fonts. I want you to use like  cutting edge um new ones that are on Google fonts,  
89:23 etc. I want you to use some placeholder images  so that the design feels alive. And I also want  
89:28 you to significantly upgrade the graph. What I've  done is I've given you a screenshot from a similar  
89:33 design that I like a lot and I want you to mimic  more or less everything that you see here just as  
89:38 a uh just as a start point. I'll modify the design  slightly afterwards, but try and really nail the  
89:46 font colors. Uh try and mimic the font styles. Try  and mimic the um encapsulation, the border styles,  
89:53 cards, etc. Okay, this is looking a lot cleaner.  I'm loving how it did this design. You also see  
89:59 way higher bar charts, way cooler bar charts. Um,  this takes me to what looks like a mini landing  
90:04 page that I put together. That's interesting.  I wasn't even like looking for a landing page,  
90:08 but it's nice that it did. So, and we even have  like a little look at that. It's like a little  
90:11 notion um avatar up top. Cool. Um, my tasks  are still the same. When I click this button,  
90:16 you see we do have that higher quality design,  which is nice. And nothing pops up for calendar.  
90:22 Nothing pops up for reports. Nothing pops up for  portfolio. Nothing pops up for these buttons.  
90:27 Neither does it pop up for those buttons. Let me  just check the bottom. Okay. And then yeah, the  
90:31 settings work. So, basically, it's just populated  this with a bunch of additional buttons that we  
90:34 don't need. I'm just going to tell it to remove  all the stuff that we don't need. And then I'm  
90:37 also going to have it adjust the font a little bit  cuz I think the font's the same. Hey, this looks  
90:41 much better. Fonts are the same though. So, I want  you to adjust the fonts. Um, I'd also like you to  
90:47 change all of the major titles from title case to  sentence case. For instance, task completion over  
90:54 time should be sentence case. Apply this to all  other elements and boxes. I'm liking the design  
91:00 quite a bit. Nice job on the avatars. I think  you should add borders rather than shadows around  
91:06 things to better match the task beto design. I'm  also noticing that you put a bunch of additional  
91:11 buttons that we don't need. Like for instance,  calendar, reports, portfolio, product management,  
91:15 project management, real estate website, my first  portfolio, all team product team, invite team,  
91:21 supports, that little thumbtack icon, the play  icon, the integrate icon. These aren't things  
91:30 that we actually need because there's no pages  that will address them and I don't plan on making  
91:34 them. So, I'd like you to remove all elements  for whom we don't have pages because that's  
91:38 not really relevant. um make the search bar apply  to our current list of orders and then update the  
91:47 design of the settings page because right now it  looks the same as it did before. And then finally  
91:52 in inbox the button that says mark all as red  is um almost completely transparent. It's very  
91:58 difficult to see unless you mouse over it. So fix  that as well. I think that's more or less the last  
92:03 round of edits and then we can get into publishing  this. Cool. And this is the final result. I have  
92:07 a very clean dashboard. You see with that scroll  on the right hand side, we don't have any issues.  
92:12 Down at the bottom, we have only the pages that  are actually relevant to us. Now, of course,  
92:16 I could add like a million pages here if I wanted  to, but I don't really. I mean, like this is just  
92:19 the purpose of the app, right? I think in our  uh crazy vibe coded economy, you're going to  
92:24 find that like a lot of apps are going to have  unnecessary features and whatnot just because they  
92:27 can. I think simplicity is actually going to end  up being quite the moat. So, yeah, I'm going to  
92:31 keep things as simple as humanly possible. Uh but  yeah, everything works, which is really nice. So,  
92:35 you know, if I click on this app, uh, sorry, if  I click on this specific content page, you know,  
92:39 you can see that I'm I'm here. Content's being  written and stuff like that. I don't have the  
92:42 ability to modify like I had last time, you know,  I can download stuff if necessary. If I go back to  
92:47 the dashboard, this um graph is interactive. Uh,  you know, and I I can jump into this anywhere I  
92:52 want, which is kind of neat. And then down at  the very bottom, which you guys can't see cuz  
92:56 my fat head's in the way, I have the settings  page, which is pretty neat. Um, and then, yeah,  
93:00 we also have the ability to search tasks up here,  which is really, really cool. Um, I could type SAS  
93:05 growth guide and then that's an initial piece an  additional piece of functionality that was added.  
93:10 Okay, so I'm just going to clear that notif. And  now that we've like really basically built this  
93:14 whole app directly on the computer, it's time to  take this to the cloud. So, what I'm going to do  
93:19 next is I'm going to go to anti-gravity and just  before we proceed with our next step, I'm going to  
93:24 ask it to save everything that it's done so far  into a gemini.mmd that effectively encapsulates  
93:30 what this app is, everything we've done so far,  and then provides instructions for future models  
93:33 in case we need to pass this off. And the reason  why I do this is because you might actually run  
93:38 out of like context. Um, that happens pretty  often. For those of you guys that don't know,  
93:41 there's only a certain number of tokens that you  can actually fit into one conversation with an AI  
93:45 model these days. Whether it's like Opus 4.6  six with a 200,000 context window. They have  
93:50 a 1 million context window option as well.  Whether it's, you know, Gemini 3.1 Pro high,  
93:55 I think I think it has a 200k context window,  might have a million. The point I'm making is  
93:58 like a lot of that context just used up by the  internal thinking of the model as it does tasks,  
94:02 right? So, in my case, like I I just don't want to  risk it. I don't want to like get the thing to the  
94:06 point where it's gone a little too far and then it  barely remembers what we talked about initially.  
94:10 Instead, what I want to do is I basically want to  like have this almost like a backup of where we  
94:14 are right now. This gemini.mmd file which uh you  know in my case is just a very highle summary that  
94:19 talks about where we currently are in the process  and you know what we need to do and stuff like  
94:23 that and then also gives instructions for you know  where to go from here. So once we're done with all  
94:29 this and now we have this which is sort of like my  fallback what we can do is we can um move on with  
94:34 the next step. So I'm going to say fantastic the  current design looks great. What I want to do next  
94:38 is I want to ensure that it works not only locally  but with Superbase as our database. Walk me  
94:45 through everything I need to do. In case you guys  didn't know, Superbase being our database. Um,  
94:49 this isn't something the model can actually set  up entirely for us. Instead, what we're going to  
94:53 do is I'm just going to ask it to walk me through.  And that's what it's done. Create a new Superbase  
94:57 project up here. Set up the Superbase client  and so on and so forth. Create the database  
95:01 schema. Enforce RLS. That's really important  like we talked about. Then replace hooks. and  
95:05 we're actually going to have it do some of this  stuff, but we'll start with database.new. So, I'm  
95:09 just going to jump over there. It'll now ask us to  create a new project. Now, in my case, I'm already  
95:13 signed into Superbase. So, I can just click that  button and then boom, it's done, right? They own  
95:17 the freaking domain name database.new. They got  a lot of this, right? And they've streamlined the  
95:21 setup to match. But, uh, if you don't already have  a Superbase account, just open up an incognito  
95:25 window and then just go to superbase.com and then  just click start your project in the top rightand  
95:31 corner and then just enter your information down  over here. email, password, you can continue  
95:34 with GitHub, SSO, whatever, and just make an  account. They're fantastic. This is like a really,  
95:38 really solid app, and uh it simplified the  development of like vibecoded sites 100x over. So,  
95:44 once you're done with that sign up, um you'll be  at a page like looks like this. I'm just going to  
95:47 type I don't know, Nikki Poof. I really got to get  more mature names for these. Uh okay, why don't we  
95:55 do I don't know. We'll just do client dashboard  app. Next up, we need a database password. So,  
96:03 I'm actually just going to make like a my own fake  one and then I'm going to copy it over. Actually,  
96:07 maybe we'll just generate one. The generated  one's weaker. So, I'm just going to copy that.  
96:12 We'll enable the data API. This stuff looks good.  Then I'll click create a new project. Okay. So,  
96:17 we have our database passwords right here. So,  just so that I don't lose this, I'm actually  
96:19 going to give it the password right here. And  then I'm just going to have some additional  
96:24 other lines. So, boom. It's now created the um  thing for me. And then I think what I do next  
96:31 is I have to go to project settings. There's a  project ID over here. So I'm going to copy this  
96:36 over and then I'll go project ID. Paste that in.  And then I need the superbase superbase URL as  
96:43 well. So I don't actually know exactly what that  URL is, but I'm just going to copy the dashboard  
96:48 name as well. So the project name, paste that  in. Let me just see if there's anything else.  
96:52 Project URL and on public key. So I think we need  to go to API, which might be over here. API keys,  
96:58 which is right over here. We're going to create  a publishable key. That's one that we want. And  
97:03 then I'm going to copy this. And then I'll say  publishable key. I'm just going to paste that in.  
97:11 Okay. So now this should be everything the model  needs. I'm not entirely sure. I think it is. So  
97:15 we're just going to give it all that information.  We're going to have it do the rest of this for us.  
97:20 Um and then cool. Looks like this is going to  take care of that. That's going to take care of  
97:25 the database schema. Oh, sorry. We need to take  care of the database schema as well. So anyway,  
97:28 I'm just going to give this its information and  then we'll have it tell us what to do next. And  
97:32 now we'll have edited an ANEV file, which is  pretty clean. It's going through and it's adding  
97:37 the information that we just gave it a moment ago.  And now what it's going to do is it's going to  
97:40 give us what's called an SQL schema, which is just  like the next step. And I imagine it'll just ask  
97:44 us to copy and paste something. It's also doing  things like wiring up the login and signup forms  
97:47 to Superbase O, which is super cool. So I mean, we  should have basically everything good to go aside  
97:52 from the landing page after this is done. Okay.  Okay. And then it's told us to go over here to SQL  
97:55 editor and then paste this giant file in. Um, so  I just did that. Since these are comments, this is  
98:00 just going to run a bunch of these simultaneously.  So I'm going to click run selected. Okay. And then  
98:03 it told me to go over here and then paste all the  stuff in and then just press run. So that's what  
98:07 I'm going to do. This will basically create us a  bunch of um tables and whatnot in our database.  
98:13 So if I go to database now, we we're going to have  a bunch of different tables and they're all going  
98:16 to connect in a specific way. And this is now  taking us to a signup page here. We obviously  
98:20 still have some changes that we need to make.  So, I'm going to do that after I verify whether  
98:23 or not this works. Now that it says signing up,  account created successfully. Please check your  
98:26 email to confirm your account. I don't know if um  we're going to get that because sometimes we have  
98:30 issues receiving emails. So, let me just pump  over to my own Gmail real quick. Okay. Yeah,  
98:34 looks like we did. So, I can now confirm my  mail. Um that's taken us to I think like this  
98:38 hastily putting together landing page, which is  not actually the page that we want. So, I'm going  
98:43 to click sign in. And then I think my email was  this. I'm not entirely sure. And then my password  
98:48 was that. Let's click sign in. Okay, cool. Yeah,  we actually did sign in. That's wonderful. Um,  
98:53 perfect. Now we have this dashboard, which is  kind of neat. My tasks. Uh, cool. We can click  
98:57 new task. Now we can create a new content order.  So, I'm going to go example content order. Make  
99:03 XYZ. Cool. And then we may have to change the  order options here, too, cuz I don't think I uh  
99:09 included that. Let's just select that. Click save.  Okay. Let's just make more characters. Okay. So,  
99:16 there's some issue with the database because  it's not refreshing. So now what I'm going to  
99:21 do is previously we're just testing the design  and then iterating on that. Um now I'm going to  
99:24 discuss like functionality. Hey, this looks great.  I created a new task using the new task button,  
99:30 but when I went back to check the list of tasks,  I didn't have a content order in the database.  
99:34 So I'd like you to double check why that is. It  may be an issue with the adding to the database  
99:39 or it may be an issue with the pulling from the  database. If there's a table problem or something  
99:42 like that, let me know. While I was doing that,  I noticed that there was an order options section  
99:47 on the right hand side of um the new orders page.  It looks like uh there aren't buttons or anything  
99:52 here. Like I can't really toggle this on or  off. So just refactor that so there's toggles  
99:57 um and those toggles actually do something and  they're meaningful. And then also make sure that  
100:02 whatever we're seeing in this chart, it accurately  reflects the data in the database. And then I also  
100:09 noticed when I tried oh on the landing page here  it's a mixture of orange and purple designs. I  
100:15 just want you to make everything here purple. The  signin button is um almost transparent. It's very  
100:21 difficult to see. And I'm noticing that there's no  way to log out once you're logged in. So just add  
100:25 that functionality as well. I need to be able  to log out. Can't just stay logged in forever.  
100:30 Cool. So I'm just going to pump all that stuff  in and then I'm going to go. So we're getting  
100:35 closer and closer to the end result. Cool. also  just made some additional changes. I'm just going  
100:39 to head over to new task. I'll say five ways to  increase XYZ. See in the right hand side we now  
100:44 have some toggles which looks nice. And I'm just  going to add some keywords. I like the idea that  
100:49 we could pick those. So I'll click save. And I'm  still not seeing it in the database, which is kind  
100:54 of annoying. Uh must be some problem. Okay. And  I'm just retesting this now. Create new content  
100:59 order. Blah blah blah. I'm click save. And looks  like we actually got it in. There is still some  
101:04 fake assignees, I think. So, I'm going to  I'm going to have that fix that. However,  
101:08 if I click on this now, you can see that we have  the brief and then we also have some keywords,  
101:12 which looks really nice. And then we have  some really nice formatted stuff on the right,  
101:14 like details, format, word count, deadline, and  so on and so forth. So, big fan of that. We see  
101:19 the ID of the order up ahead, which I guess could  be useful if I'm trying to contact my agency. Hey,  
101:23 I can't see a thing on my dashboard. Yeah, what's  the ID? Then, when we go to dashboard, we can now  
101:28 see this has been updated. And it looks like it's  using some sort of mathematical formula here to  
101:33 divide this so that the top of this is always  going to be equivalent to the total number of  
101:38 maximum orders in a in a certain period. So that's  pretty neat. I like that. What else did we need?  
101:43 Uh we didn't have the ability to sign out. So I'm  just going to double check. Do I have the ability  
101:47 to sign? Oh yes, I do. It's just hiding behind  my head again. So at the very bottom lefth hand  
101:51 corner, I click log out. Okay. Now I'm back to the  client dashboard, which is nice. Uh what I'm going  
101:56 to do is I'll go back to sign up. Okay. And then  that looks pretty cool. And then I'm just going to  
102:00 try go back to the actual landing page. Nice. So  it actually did fix this up. Although this landing  
102:04 page needs a lot of work, which we're going  to fix up in a moment. Uh and then cool. Yeah,  
102:09 we have the ability to sign up. We have the  ability to sign in. I mean, this is pretty  
102:11 solid to me. I don't really think much else needs  to occur. I'm going to modify the design on both  
102:17 the signin pages, sign up pages, and the landing  pages. I'm going to make the landing page really  
102:21 sexy uh just to wow you. And then yeah, I  think I think that's more or less it. Oh,  
102:26 I guess what we should finally do is we should  check to see if we if we have middleware. So,  
102:30 what happens when I go slash orders? Notice how  it immediately jumps back to the login page. How  
102:35 about if I go /dashboard? Nice. How about if I go  orders new? How about if I go settings? How about  
102:41 if I go what else did we have? Notifications. We  got notifications. Nice. So, it looks like our off  
102:47 middleware is pretty good right now. Um although  I'm obviously going to have to double check that  
102:50 with that big prompt that I talked about earlier.  Now, why don't we modify the design? And once the  
102:55 design's ready, we can actually push this and make  this public. And then once this is finally pushed  
102:58 and made public, we can um round out the the  end of our site and it'll be really high-end and  
103:02 really clean. So again, for design, I'm going to  go to dribble and then I'm going to find a really  
103:07 nice high quality uh let's just see if we can  find some really nice high quality screenshots.  
103:12 I just want like a super sexy hero header, you  know? I want it to just be insane. So why don't  
103:16 we go landing page? I just want this to be the the  best thing ever. This is really neat. Elevate your  
103:24 trading experience. Liquid brokers. H I think I'm  probably going to want a white mode thing though,  
103:30 like a light mode as opposed to a dark mode. Cuz  if I go dark mode, it'll just seem kind of weird.  
103:37 So, I think what I'll do is this section over  here that says AI powered adwords that find lead.  
103:43 I think I'm going to roll with this. And then I'm  going to try and have some cool animated section  
103:47 over there on the right. So, unfortunately,  I think this is about as big as it gets. Um,  
103:52 sometimes, again, they don't want you to uh like  I don't know if you can tell, it's kind of blurry  
103:57 here. Sometimes, as I mentioned, they don't want  you to uh copy these designs a ton. I'm still  
104:01 going to give it my best effort, though. So, I'll  go back here. I'll paste this in. Excellent. Um,  
104:07 what I want to do next is modify the landing  page, the sign-in page, and then the sign up  
104:13 page. I'm going to attach some screenshots of the  design. In addition, I'm also going to provide  
104:18 you a an animation that you could use to run in  that right-hand side pane. What I'd like you to  
104:24 do is I'd like you to combine our design with  the design in the screenshot. I want it really  
104:29 clean on the left hand side. I want my main value  proposition. I want some hypothetical logos um in  
104:36 the center. You could just use some some plain  text for that. You don't need images. Um, I want  
104:41 a relatively long landing page here with sections  for product, sections for pricing, hypothetically,  
104:48 uh, and then obviously a button up at the top that  says sign in. And then for the video, I want you  
104:54 to use this URL. Now, what I'm going to do is I'm  going to paste this in. And then I'm going to go  
104:59 back here. And what I want is I basically want a  cool asset. And then I want to like rotate it with  
105:04 AI. And I'll show you what I mean here. Uh, but  basically I just want some sort of asset. Let me  
105:08 see what can we get from dribble that looks really  clean. Okay, so for that I'm going to go over to  
105:12 dribble and go 3D shape. Um, I really like one  of these shapes, but I think I want this on like  
105:20 a white background. Let me see if we could get  something that might make more sense. H organic  
105:26 3D shapes over here. This is cool. Round egg 3D  shapes. That's kind of neat. Feel like this would  
105:32 be nice if we could get a white background on it.  So, why don't we give this a try? What I'm going  
105:37 to do is I'm going to take an image of this. Then  I'll go over to Google video. Actually, sorry. Why  
105:45 don't we start with Nano Banana? So, I'll go to  AI Studio.google.com. You can sign up for this  
105:52 um you know, it's the same account that you  would use for whatever credits you utilized for  
105:57 um Gemini and anti-gravity. Then I'll say, let's  see, playground. And I want to generate images.  
106:03 I'm going to paste this in. So, I create me a  new organic shape that looks like this except  
106:10 with a fully white background. It should look  very sexy and have a purple blue hue to it. And  
106:20 then the aspect ratio, I think, is going to be 34,  which is three wide and then four high. And we're  
106:25 going to go 2K for the resolution and press enter.  Okay, just ran out of token, so I had to go to a  
106:31 different account here. But it's important that  the aspect ratio is 34 and the resolution is 2K  
106:35 because that's approximately the same height of  the design that I wanted to copy a moment ago,  
106:39 right? Like this one is clearly taller than it is  wide. And I think that's going to be important.  
106:43 Okay, so we have this organic looking image,  which is kind of cool. What I'm going to do is  
106:46 I'm going to save this image as. So now this is  like our own image. Then I'll head over to VO3,  
106:50 which is a state-of-the-art um image to video  generation model. And then what I'm going to do  
106:56 is I'm just going to do the same thing. So let  me just switch the accounts here. And then I'll  
107:01 link the API key. Cool. And then what I'm going  to do is enable Google Drive access as well.
107:08 And then I'm going to paste in this  image, one that I just generated a  
107:11 moment ago. We'll go 9 by6. Just copy this  image. And then go back here. Paste this  
107:18 in which I just downloaded. Okay. And  then I'll say uh rotate this slowly. I  
107:26 think 6 seconds is probably fine. So that  the last frame is like the first frame.
107:34 Perfect. Uh 100% white background. Okay. I'm going  to press enter. And now V3.1 is generating me  
107:42 uh video. Just playing this. Now we have this cool  organic undulating rotating shape which is really  
107:47 neat. So I like that a lot. I'm going to download  this. And now I have the video. I'm going to grab  
107:52 this. And then I'm just going to dump this into  public. And then I'll say hero header video.mpp4.  
107:57 Okay. And then um I initially said URL, but I  guess what I really meant was like inside of my  
108:04 folder. public/hero header video.mpp4. Also put a  dot here. Uh cool. I'm just going to press enter.  
108:15 And then I'm going to give it some time and let  it go on its way. Okay. Landing page design is  
108:19 looking all right so far. I'm not a big fan of  this video now that I'm looking at it. Like I I  
108:23 would love it if it just blended in seamlessly,  but for whatever reason, the background is not  
108:27 white. We could remove the background and then  we could remove that border and I think it would  
108:30 look a lot sexier. I'm partial. I mean, I might  play around with it a bit. Um, we got the logos  
108:36 down here. Sign in button top rightand corner.  That's nice. I think we just need to align uh  
108:41 what's on like the the left hand side and stuff.  So, yeah, I'm going to do that in a second. Sorry,  
108:45 I just realized I was dictating. And then I'm  just going to give it some instructions. Hey,  
108:49 this looks solid. Why don't we align all content  to the hero header? Because right now some of the  
108:56 sections are really tight. So push it out so that  everything's at the same width as the hero header.  
109:05 Um I'm loving the colors. Something with the graph  under the why they prefer Taskbo section looks a  
109:12 little bit weird. I think the numbers on the cards  of the maximize your traffic with the content  
109:19 engine that generates section um are broken or  something because they're being cut off. So fix  
109:25 those. And then everything else looks really  solid. Just update the business name from taskb  
109:31 to content hamster as well. Uh and just do that  across the entire app. And then while I'm at it,  
109:39 I'm going to see if I could find something that  looks a little bit better for that section here.  
109:43 Okay, I found this video and I really like  this. I also think probably because this  
109:47 is already optimized, like the file size isn't  really big, I'm probably just going to use it.  
109:51 This is not a live SAS app. So, Ashkat Ash Moana,  great job on said design. Um, we're just going to  
109:57 be using this as a a demonstration purpose. Um,  it looks like he also allows you to explore or  
110:03 tweak this scene by checking it out in his spline  community page. I don't even know what that means,  
110:07 but I think it's like a way you can share designs  and stuff like that. So, that's kind of neat. Oh,  
110:11 wow. I can actually revolve it. That's so cool,  dude. Wow. So, I guess you could probably run  
110:15 this as 3D in your browser. Anyway, we don't want  to do that. That's just a little too much. Um,  
110:19 but I'm going to instruct the agent to use that  image and then make sure the orb doesn't get  
110:24 cut off weirdly. Also had a couple of sleek  cards placed on top of the video with a glass  
110:27 isomorphism effect. So, something to break it  up a bit. And then I'm just going to jump back  
110:31 over to our app here so we can see um as it's  actually being developed in real time. You can  
110:36 see we changed the name to content hamster.  Kind of silly, but uh whatever. This has now  
110:41 done everything perfectly with left-hand side  um margin match up which is nice. Why they  
110:46 prefer content hamster? This has now been spread  out. So the graph is no longer all weird. Again,  
110:50 perfectly margin aligned. Something with the  numbers is still off. So I'm going to have it fix  
110:54 that as well. Let's just do that before I forget.  Hey, numbers on cards are still a bit weird.
111:04 Screenshot for reference. Then as I scroll  through, is there anything else? plus premium.  
111:10 Okay, ready to level up your process. Cool. And  then we also have a bunch of buttons down at the  
111:16 bottom here that obviously don't really work.  Okay, cool. This looks really clean. I'm liking  
111:19 this. I'm liking this a lot. Um, I'd say the only  thing that's missing is probably like the coloring  
111:24 across the board. Um, what I'm going to do now is  I'm just going to match all of the colors to this  
111:30 because I just love the way that that looks. Uh,  I'm just going to take a screenshot here and then  
111:35 I'll say colorfinder online and I'm just going to  paste in this color. And as we see here, there's  
111:45 this hex code. So, I like that a lot. And what I'm  going to do is I'm just going to paste this in as  
111:48 well. Also, I want you to update the primary color  scheme across the entire application so that it is  
111:55 that shade of baby aqua blue. Looks really cool.  use slightly darker shades of blue as accents,  
112:02 etc. And then for the accents, I guess I'll  just go back to this. Screenshot it. Paste  
112:11 this in here. And then we can use stuff like this  as well. Um, and let's go a little lighter here.
112:21 We'll go this dark. Darkest. Cool. And now it's  going to go through the entire app and just change  
112:26 it to reflect like the background colors of this  which I think are going to be really cool. Um,  
112:31 obviously I don't want, you know, us not to be  able to see the text. So I am just going to be  
112:35 around like, you know, if this is that color, it  will seem kind of weird. Right now we obviously  
112:38 have a cool purple thing going on with all the  highlighted text being purple and then these  
112:41 things being purple. I do think it would just  match really really nicely and seamlessly if we  
112:45 had that. So we'll give that a try. All right,  this is looking way better. I'm loving how the  
112:48 baby blues actually reflect what's going on here.  Let's click try out for free. Um, actually I guess  
112:53 I already have an account, so I'll just log in.  But we are going to test that out in a second.  
112:57 Looks like those numbers have been fixed, which  is cool. We got that slight zoom in on each card  
113:01 as well, which I really like. And then baby  blue effect everywhere. That's gorgeous. Um,  
113:05 this little content output graph is still purple.  So I'm just going to see if we change that. One  
113:09 thing I didn't really talk about was mobile  optimization, but um, Gemini Mobile optimizes  
113:13 everything really, really well just out of the  box. Obviously, we do have some minor issues  
113:16 with the bounding boxes of this um, like video,  the border. Uh, but this is something that would  
113:21 probably be better done by actually replacing the  core underlying asset because the resizing of the  
113:25 border box occurs basically perfectly. Um, but  just giving you an example of what it might look  
113:30 like on a tablet. As you can see, everything  is still basically perfect. I haven't ran into  
113:34 major issues so far. Even the numbers and stuff  like that are visible. And if we scroll all the  
113:37 way down, the biggest problem is just there's this  little black overlap. I want you guys to know that  
113:41 I haven't even done mobile optimization yet, which  is typically just like a line in the prompt. Um,  
113:45 really, really wild that it does this good of a  job without me even having to tell it. Cool. And  
113:49 now we have everything changed. So content output  is nice and changed. Uh you know the the the look  
113:54 of the little bubbles here goes up. I mean this is  basically exactly what we wanted. I also changed  
113:58 the name to content wheel obviously. So why don't  we now sign up and go Nick Sarif. We'll call  
114:04 myself Leftclick Inc. We'll just use a slightly  different email this time just so we could retest  
114:08 the sign up. Uh let me just make sure this is a  simple ass password. That'll always be the same.  
114:13 Cool. We're going to click sign up now. Account  created successfully. Please check your email to  
114:17 confirm your account. Uh, I'm now going to sign  into the specific mailbox that gave me that. So,  
114:22 let's go here. You can see it's saying to confirm  my sign up here. So, I'm just going to go back and  
114:26 then I'll now be able to sign in with the specific  email that I just put in. We'll click sign in.  
114:32 Just delete all this stuff. Cool. And I'm just  going to zoom in a bit so we can see. Obviously,  
114:35 I have a really big desktop so it won't always  look like that. Um, cool. We have total tasks.  
114:40 You can see the styles been updated a little blue  here because I just wanted it to be blue. Um,  
114:44 task completion is still purple. So, I think I  may update the graph. I don't really know to be  
114:48 honest. It might be okay. My tasks inbox. Cool.  All that stuff looks good. So, we'll now test  
114:53 the new task. I'll say, "Hey, this is an example.  Here's my example brief ABC." And then over here,  
115:02 we'll go 13. I'll click save order. Cool. We now  have it in our database. It's now a draft. So,  
115:08 content is being written. Your writer's currently  working on it. You'll see it when it's ready.  
115:11 Inbox. Cool. Um, awesome. I mean, this looks  pretty solid to me. I don't see any major  
115:15 issues. So, we'll go to settings. We can update  things. I don't know if this got saved, that  
115:20 little leftclick thing. Yeah, it doesn't look like  that did. So, now that we basically have like the  
115:25 whole app ready to go, we just need to make a few,  you know, minor changes before we deploy. Hey,  
115:31 uh it doesn't look like the company. Okay. And  then after a full endto-end check, this website  
115:36 and app is now ready to be pushed um and actually  deployed. That deployment is going to occur on  
115:41 Netlifi like I talked about earlier. And I'm going  to give you guys a simple simple process that you  
115:45 guys could use in order to actually push that out.  For now, I'm going to go grab some lunch since I'm  
115:48 pretty hungry, but I'll BRB. So, the strategy  I'm going to use in order to push this site and  
115:52 make this web accessible just like we did with our  old website is to use Netlefi. And in order to do  
115:59 that, we're going to use Netlefi CLI and then give  the vast majority of the work to the model. Um,  
116:06 so Netlefi CLI is basically just a command line  interface. That's what the CLI stands for. That  
116:12 lets previously people like you and I, but now  mostly agents interact with the Netlefi API,  
116:17 which is like their little application programming  interface. The way that you uh do various actions  
116:23 on Netlfi using HTTP requests autonomously for  you. They can actually do most of it. So what I've  
116:29 done here is I've actually just said, hey, I want  you to build this and then I want you to push it  
116:33 on Netlfi. We've gone through the whole build.  We've gone through phase one, two, superbase,  
116:38 you know, phase three and now what we're doing  is we're doing this final production deployment  
116:43 and this is uh you know just uh part and parcel  more or less the final step of getting this thing  
116:49 on the internet. Now, I should say that when you  actually push things to production, that's what  
116:54 that's called, typically there are some issues  because the development environment that you run  
116:59 on your local computer is always going to be a  little bit different than whatever the heck the  
117:02 server is using. And so, sometimes you have  to make minor changes to the site. Sometimes  
117:06 you have things that work on your device, but then  don't work obviously on the production server. So,  
117:11 you have to, I don't know, modify like the the  way the database works, some of the permissions  
117:15 and stuff. And that's just part of dealing  with any sort of full stack application.  
117:19 wasn't a big deal with our website because our  website was obviously not super interactive. It  
117:23 was just a bunch of uh links. But this has some  back and forth and that's what's important. So,  
117:27 as we scroll up here, we can see this has actually  gone through the build process, which is basically  
117:33 taking our website or what our website should  look like and then turning it into like an  
117:37 actual instantiated version of the site. That's  what that is. Um, and now it's going through the  
117:41 installation process for the actual Netlifi CLI,  which is a little package. And now it's going to  
117:45 communicate with the Netlefi CLI. And all we  have to do is we just have to authenticate. So  
117:50 assuming we have an account set up and stuff  like that. And what we have to do is we have  
117:52 to go and actually log in. So what I'm going to  do is I'm going to open this up. So I'm going to  
118:00 um give it a new terminal tab which is what it's  asking for. You can do that by going control shift  
118:06 um tilda I think it's called. It's like that  little thing in like the top leftand corner.  
118:11 It looks kind of like this. Uh you should also be  able to do this by going up to the top and then  
118:16 going terminal and then going new terminal. And  then I'm just going to paste this Netlefi login  
118:21 command. And in my case, you know, I've already  done this on my machine a bunch of times. So it'll  
118:25 say it's already logged in. In your case, uh you  know, if you haven't done this login before, it'll  
118:29 just pop up a little Netlify panel and then you  can do it there. So I'm going to say done. And now  
118:33 it's going through the various building procedures  and so on and so forth. So, my expectation is I  
118:38 mean there's like a 50% chance we oneshot this  because we are using a pretty standardized  
118:41 framework called next.js. Uh there's a 50% chance  that we're going to have to go back and forth.  
118:46 It's saying the deploy is live. I'm just going  to refresh this and check for myself. Okay. And  
118:50 you can see we did push the actual website side  of things up. So, that looks good. Just exit out  
118:55 of this so we can get like the the full glory of  the site. And I'm not seeing any major problems  
119:00 here versus the one that was on my machine, which  is cool. You know, we got these little buttons.  
119:06 premium get started now. Okay. And I mean, we  haven't added links to all these yet, which is  
119:09 why these aren't actually popping up. Uh like  changing my um little cursor when I mouse over,  
119:14 but that's okay. And now what I'm going to try  is actually the interactive component. Cool. So,  
119:18 we have the account created successfully modal.  And that's great. I'm now going to head over to  
119:23 that specific email address and just check if we  have the email confirmation message. Okay, we have  
119:27 this right here. So, I can confirm my mail. And  now we're starting to see one of the issues. So,  
119:32 error parsing package.json JSON file. And it looks  like the reason why we're getting that is notice  
119:37 the URL up here. The URL is not actually uh the  publicly accessible URL. It's this localhost 3000.  
119:44 And so the way that you know user signups usually  work is basically there'll be some URL here slash  
119:50 some code and then the agent will configure your  Nex.js or whatever full stack app framework you're  
119:56 running to like accept this code and then  use that to verify the account. But basically  
120:00 what's happened here is the default sends us to  localhost 3000. And so I know what the issue is,  
120:05 but just for the purposes of like debugging and  stuff like that, I'm actually going to just take  
120:08 this to my agent and I'll see what the agent says.  We have to go back into our Superbase settings  
120:12 and then alter it. But I do just want you guys to  know like how this works if you guys are actually  
120:16 debugging it. So I'll say I received this error,  this um call back when setting up my account.
120:27 this isn't the right URL. What should I do? And  you'll see here it'll instruct us to go to our  
120:34 Superbase and and make some minor changes. And  then it's going to tell us to update our Superbase  
120:38 dashboard settings. So, go back into my dashboard,  which is right over here. And then there's going  
120:44 to be an authentication page right over here on  the top left. And then it's either going to be  
120:49 an email or somewhere else. Like we can actually  turn off the um confirmation completely. This is  
120:54 the email that Superbase sends them. That's why  it looks a specific way. And I don't know, to me,  
120:58 it kind of looks a little ugly. So, obviously, if  I were running a full production app and I were  
121:02 making all those uh little bells and whistles, I'd  make this look really pretty. But I think there's  
121:07 actually a way to change the URL. Okay, here  it is. It's right under URL configuration. It's  
121:12 a site URL. And so, what I'm going to do instead  is I basically just want to get the the live URL,  
121:18 which is this one here. Just going to paste that  in. Okay. And I'm going to remove that back slash.  
121:24 So now this is going to be the URL that it uh  basically redirects us to on Superbase's side  
121:29 cuz Superbase handles the the signups and stuff  like that. And now when I go back to the app,  
121:35 I don't even have to actually make any uh major  changes like I don't have to rebuild the app or  
121:39 anything. Then I can give it an email. I click  sign up here. And now we have another email sent  
121:48 to another one of my addresses. There you go. It's  right here. When I mouse over this, you guys can't  
121:53 really see in the bottom lefthand corner, but  I'm no longer going to localhost. And you can  
121:57 see if I click on this, takes me to contentorb  dashboard.netifi.app/lo. So here I can actually  
122:06 um now jump in and log into the app. Cool. So  that's one of the first changes I would say if  
122:13 you're building something on production versus  doing something locally. We just have to change  
122:16 the URL that the authentication goes to. So  we just did that. Now it's nice and clean.  
122:21 And I'm scrolling up and down here, making sure  there are no major issues that are egregious. I'm  
122:26 not seeing any problems. I guess the company isn't  filling in. So, we can change that. Well, I write  
122:31 this. Save. Give this a refresh. Yeah, looks like  the company in the profile details isn't filling  
122:35 in. So, I'm just going to take a screenshot  of this and then I'll paste this in. And then  
122:39 while I'm like before I actually send this, cuz I  prefer just to do a bunch of stuff, not just one,  
122:43 let me test the new order. This is a new order.  What's up? And then over here, I'm going to go
122:54 one word and then I'll save. Okay. So, yeah, this  is actually in my database now, which is pretty  
122:58 cool. I can click on this. Looks like the rest of  this stuff is is here, which is nice. Anything in  
123:04 my inbox? Nope. That's cool. And then my actual  orders do populate, which is nice. And we don't  
123:11 actually have like the ability to like we don't  have any people on our team, but obviously we  
123:14 could have added some, which is why we still  have those little um little fake Notion-esque  
123:18 people there. I wonder where they're getting those  images from. That's kind of cool. Okay, so uh we  
123:23 have dashboard. Did we I think we checked settings  as well. Yep, that looks pretty good. And I think  
123:29 every other toggle here probably works. Let me  just try searching the tasks. So I'll just go O
123:38 H. Okay, so yeah, clearly it's just searching  based off the letters. If I type in new order,  
123:43 if I type in old order, then it won't be.  Okay, cool. So if we had 20 in here, you know,  
123:46 that might actually be useful. And let me just  check the sign-in page. Okay, cool. So, we're  
123:51 immediately being logged in when we click the  signup page. And then, you guys can't see this,  
123:54 but in the bottom left corner, I'm just going  to click log out and see what happens. Okay,  
123:57 so when you click log out, it takes you straight  back to the login page, which is the desired  
124:00 expected behavior. And then nothing is prefilled.  Let's now check all of our routes. So, I would  
124:06 always recommend going through every single one.  So, that's dashboard. This is settings. This  
124:10 is orders. This is orders. um I think and then  slash new and then there's also a notifications.  
124:17 So notice how none of these are automatically  signing in which means some of our off uh is  
124:21 fine and you know all of our animations and stuff  like that look really clean and uh there there are  
124:26 no problems on that end. Okay. And then I'm just  going to do a final spot check because although I  
124:31 told it to um you know mobile optimize the site, I  do want to make sure that it is mobile optimized.  
124:36 This looks really clean. You know the only issue  here is with this little background not being  
124:41 necessarily the same as that. I guess we could  just change the background color of this whole  
124:44 element. That might actually help. So, yeah, why  don't I do that? That's a really cool thing. Um,  
124:48 why don't we adjust the background element of our  little orb to this? Adjust the background color of  
124:58 the element that contains the animated orb to be  the same as that background color of the orb. This  
125:02 should be our light aqua baby blue. Okay, that  looks nice. And then just scrolling through here.  
125:08 Any other issues with the mobile optimization?  Just make this really nice and small. Be 100%  
125:14 sure. Nope. Everything here looks like Oh, you  know, there is one. Looks like this learn more is  
125:20 not immediately underneath the get started now.  So, that's one final thing that I'll do. I'll  
125:26 finally say also the learn more button is not um  underneath the what was the name of that button?  
125:33 Get started now button. So, I'd like you to line  them up on mobile. Now I should know because  
125:37 this is no longer local on our computer. It's on  netlefi the uh debug in an iteration loop takes a  
125:44 little bit longer. Naturally when we were just on  our computer anytime we'd have an issue we would  
125:48 just say hey rebuild it. It would rebuild it and  then we'd immediately have access to it. Right?  
125:51 But here we actually need to make the changes  locally on our computer push them to Netlefi  
125:55 have Netlefi do the build and because Netlefi is  sort of doing free server space takes longer and  
126:00 then we actually have to go online and then um  you know take a peek at the website. But anyway,  
126:05 I'm uh just going to open this up now. This is the  production version. I should be able to see the  
126:10 background and everything like that. Nice. And now  we have more or less my exact desired behavior,  
126:15 which is how that orb just looks like it's  floating. And it doesn't matter how big or  
126:19 small the website is. It's just like perfectly  contained within those bounding boxes. Super  
126:24 clean. Um and then I just had a couple of those  other changes. Just going to verify that those  
126:28 are good. The last thing I'm going to do is I'm  just going to go down to my security audit prompt.  
126:32 scroll all the way down to the bottom. And  there's just a little bit of nuance here. Um, I'm  
126:36 not actually going to do this in the same thread  that I launched and built the entire website. So,  
126:40 I'm going to use Claude Code for this instead of  Gemini. And the reason why I'm doing it is cuz I  
126:44 don't want any context. I basically want the model  to look at this entire codebase as if it's new,  
126:48 fresh, and not mine, like somebody else's. And I  want it to like run through this whole checklist  
126:52 and to end and tell me whether or not it's  configured. Okay, so I just copied this all.  
126:57 I'm going to head back to anti-gravity. Close  these tabs. And I should just be able to do so  
127:01 by clicking close all. Let's make this smaller  because we're going to have clawed code instead.  
127:07 And then what I'm going to do is I'm going to  paste this in. Then I'm going to say don't make  
127:12 any changes. I just want the audit. Okay. And I'm  going to pump this in. And now cloud code is going  
127:19 to go pass one discovery map the full architecture  and then more or less tell me, you know,  
127:24 what's good, what's bad. And I'm doing this for a  couple of reasons. Um, obviously I want no context  
127:29 as mentioned, but I also think that Claude is a  lot more interpretable than Gemini right now. If  
127:33 you guys check out the thinking tabs here, and  if you compare them with the thinking tabs of,  
127:38 you know, Gemini, Gemini uses language that just  doesn't really make much sense. Um, you know,  
127:42 when you look at it, it's not as interpretable.  I'm now taking action to apply the color  
127:45 adjustments as previously decided, focusing on  the background's visual look. I'm now proceeding  
127:48 by updating the styling to use desired hex code.  You if you look at it and just compare it to this,  
127:53 this is a lot more interpretable. I now have a  comprehensive understanding of the codebase. Let  
127:56 me compile the architecture summary and then  work through every checklist item. They give  
127:59 it to me sort of in plain English. And so the  interpretability is pretty important when you're  
128:03 talking security, right? Um in addition, you  know, I just think Claude is a little bit better  
128:07 at like security based stuff right now. Okay,  so I'm I'm reading through everything here. It  
128:12 looks like most of our audits are correct. We  do do console. Calls, which is important. Um,  
128:18 this is basically a way to minimize the attack  surface uh and then minimize or maximize  
128:23 operational security. We're not really giving  away how our underlying tool works. So that'll fix  
128:28 that. Startup validation. Okay, so we need some  sort of startup validation here. It looks like
128:36 we can't tell whether or not the RLS is enabled  um because we don't actually include the files in  
128:40 the codebase, but that's okay. We are actually  good on RLS. Uh that's why it's giving us this  
128:44 warning message as opposed to a no. We're good on  SQL injection, security definer functions, same  
128:49 thing as uh previously. So we could have um claude  like connect to our superbase instance via API if  
128:55 we want. And I'm just going to take all of this  default deny routing fails. This is important.  
129:02 Remember how I talked about O middleware? This  basically just immediately denies all requests  
129:05 that don't have some form of authentication. Right  now our app doesn't do that. And you know, at the  
129:11 end of the day, we while are okay on most things,  we're still missing some uh some things there. So,  
129:17 I'll say excellent execute all of the changes.  Okay. So, I'll say um the SQL stuff was already  
129:27 taken care of. RLSTC, but thanks for pointing  that out. Execute all of the remaining changes  
129:35 and change SQL checklist items to green. And now  it's going to go through and then work through the  
129:42 fix the systematically form. And just to give you  guys some more visibility in the security side of  
129:45 things, you know, previously we had a line in our  code that said if there's an error, we actually  
129:49 add all of the superbase insert error details  here. This is actually just like gives a bunch  
129:53 of information about the way that our superbase  database is configured. And that's just because  
129:56 Jeem and I wanted to write it in the way that was,  you know, debugable for us if you guys remember.  
130:00 Well, this now went through it identified that  and it changed that to fill to create order,  
130:04 which is just much more general. And um that both  gives us the ability to quickly see if there is an  
130:08 issue, but then it doesn't expose any of our  underlying in infrastructure architecture if  
130:14 somebody actually wanted to attack us. As you  can see, we have most of the changes that we  
130:16 talked about before. We have allow list routing.  We have client side user ID filtering. Okay,  
130:20 server side validation. We added what's called  an off call back route. Um we have environment  
130:25 variable validation. And we also had some console  leaks that we've since cleaned up. And just  
130:30 scrolling through to get to that checklist,  you know, the vast majority of these are now  
130:34 green. all the issues that we had previously have  been taken care of. And so 5.1 which was a dev  
130:39 dependency and then 6.2 which was some built-in  rate limiting. You know, we don't have control  
130:44 over that. And these aren't really large security  issues as well, which is why they're in this  
130:48 yellow little warning label instead of the X. Um,  you know, we're now good. We've basically solved  
130:52 for the 8020. The vast majority of lowhanging  security fruit do no longer apply to our app.  
130:58 And while I probably wouldn't scale this to like a  million users without at least having some sort of  
131:02 security person look this over, it's pretty dang  good. I just did a final walkthrough of the site.  
131:06 We have the animation, which was previously  hidden because of that middleware update. Um,  
131:10 some of our icons are missing as well because  of that same thing, as well as the login uh and  
131:14 signup routes, as well as every other feature in  our network. So, this is an example of like that  
131:18 extra step at the end that a lot of people don't  want to do because obviously it takes, you know,  
131:22 an additional 15 or 20 minutes. It's about how  long it took me, but I highly recommend it if  
131:25 you do plan on making your app live. and actually  launching it to more than just yourself and your  
131:30 close friends and family. But congratulations,  we have now built an endto-end full stack app  
131:35 with login, authentication, signup. We have a  database where we store records and queries and  
131:40 then also a pretty sexy landing page. You guys  now know everything that you need in order to  
131:44 basically take this knowledge and then go out and  actually just vibe code an app. What I want to do  
131:47 next though is I don't just want to vive code any  app. I want to take you guys to a higher level of  
131:52 understanding of modern software engineering  principles cuz I think it's one thing to be  
131:55 able to just use natural language to, you know,  communicate with agents and have it do something.  
131:59 But some of these architectural decisions that  the agents are making are actually controllable  
132:02 by you. And if you know even 10% more than  the average Vibe coder, you can, you know,  
132:07 rather than this process takes 2 or 3 hours, you  can basically just build like your little starter  
132:12 kit with um your preferred stack, your preferred  tooling, and then some preferred examples.  
132:17 you can get to the point where you could actually  just like oneshot uh an app like this. Maybe not  
132:20 completely oneshot but like 1.5 shot it. Take uh  app creation which previously might have taken  
132:25 a very very long time because of all these  small little decisions that you and I have  
132:28 to make and then get it to a point where it's  it's much faster. And not only that, I think  
132:32 you'll probably gain a lot more appreciation for  software engineering as a whole and then become a  
132:36 better developer um whether or not you're working  in conjunction with agents. So let's do that next.  
132:41 Now, I should know that this isn't going to  be a comprehensive software engineer degree  
132:46 or anything. In fact, I know there are going  to be some software engineers in the comments.  
132:50 They're going to look at what I'm talking about  and probably laugh. They're going to say, "That's  
132:53 not how to actually host. That's not how databases  work or whatever the heck." The reality is, you  
132:59 don't need to know all of the internals of how all  of these different things were put together. These  
133:04 are essentially things that were built over the  course of the last 50 or 60 years. And nowadays,  
133:09 even the most competent developers typically  just stand on the shoulders of giants. And so,  
133:13 I'm going to give you a very high level overview  here. And this high level overview will, like the  
133:16 rest of the course, be the 8020 that allow you to  dive deep into specific parts if you so want to.  
133:22 Um, but you won't necessarily need to hear me  yap on for 700 hours about different database  
133:26 architectures. Okay, so we're going to start with  hosting. To make a long story short, hosting is  
133:31 just the technical term for running your web  application on a server. works on my machine  
133:37 is the scariest thing a developer could say. So,  it's important to do this right. You know how a  
133:42 few minutes ago when I was developing that app  and then pushing it live, there were immediately  
133:46 some problems. Well, what a lot of developers will  do is they will develop entirely locally and then  
133:51 they'll say, "This works fantastic, amazing." And  then they'll leave the pushing and the deployment  
133:55 and and you know, all the hosting and stuff like  that is sort of an afterthought. And then when  
133:59 they actually do push it, then there's a bajillion  problems and the app doesn't really work. And then  
134:03 as you saw there's also a longer feedback loop  because now this is live on the internet it's not  
134:06 on your machine and that longer feedback loop can  make you know that last mile problem where you've  
134:10 done 99.9% of the work uh of your app but then  that.1% takes like 100 times longer just because  
134:15 of the nature of your your prepare preparation  or lack thereof. So in order to really get this  
134:21 we just need to give a brief highle walkthrough  of how communication um on networks tends to work  
134:27 or like on apps. Basically on the left hand side  here we have a client. And for all intents and  
134:32 purposes, let's just say this is um your browser.  Okay. Now client is just such a silly name. I  
134:38 would say not a lot of people really understand  this, but I just replace basically all instances  
134:42 of the term client with just your browser for  now and everything will make more sense. Now your  
134:46 client, okay, it's just your browser. It needs  to communicate with a server in order to have  
134:52 some sort of back and forth. And so this server,  okay, could actually be like your um computer if  
135:01 you're developing locally or if you know you're  pushing to some, you know, service or whatever,  
135:07 it could be something like well maybe Nellifi  isn't the right example, but I don't know. Yeah,  
135:10 I don't know if you guys ever used Hostinger for  instance, not affiliated with them, but um that's  
135:14 a good example of one. Or let's just see Hostinger  alternatives. Blue Host, if you guys have ever  
135:21 developed a WordPress website or Cloudways or Site  Ground or Insta. Basically, these are services out  
135:28 there that run computers just in the cloud that  you can then access in order to like push stuff on  
135:34 it. Okay? So, what happens is your client, okay,  your browser will communicate with the server,  
135:39 which if you're running stuff locally is either  just your computer or it's somebody else's  
135:43 computer on hosting or site ground or whatever  other service. And then it'll send initially some  
135:49 very simple and brief requests, okay, for things  like um you know site HTML. It'll send uh requests  
135:57 for things like uh database data and whatnot.  And so your server has the ability to do a lot  
136:02 of its own logic and whatnot. And so in general,  what clients um are for are they're just for like  
136:08 presentation. And then what servers are for is  they're more for like the heavy lifting and the  
136:12 logic and the validation and stuff like that.  That's why server side validation is much more  
136:15 important than client side validation. That's why  it's a big security risk. The thing is in addition  
136:19 to the server, there's like a third piece called  your database where we actually store like a  
136:24 row or tabular data. And you know, for all intents  and purposes, your database is just one big Google  
136:29 sheet. There's like an ID column, there's like a  date created column, there's like a I don't know  
136:34 value column, and then there's uh just a bunch of  stuff here. um where basically what your server  
136:41 does is it will make a request to this database  to receive specific pieces of data that the user  
136:47 has requested. So if the user has an ID of 23 foi  whatever, okay, and then you know the request is,  
136:53 hey, I just want you to grab the value. What it'll  do is it'll return 2 3 4 2 3 4 um over and over  
136:57 and over and over again. Okay, so that's that's  pretty cool. Um the reason why we split these two  
137:02 up is just servers are obviously computers. This  is obviously just data. Data are not computers,  
137:07 although computers can contain data. You could  theoretically contain a bunch of information on  
137:10 your server, but why would you pay for all that  when you could just use like a highly optimized  
137:14 service that just stores a bunch of records and  then communicates really quickly. So anyway, let's  
137:18 say our client is requesting some data. Like if  you if you go back to our app, um it's requesting  
137:23 that like little dashboard data. Let's see how  many orders do we have in the database. What it'll  
137:28 do is, you know, we we'll first we'll sign onto  our app. We'll initially make that request to the  
137:32 server. It'll return a bunch of HTML and um any  pre-existing or cache data, okay? And then it'll  
137:36 visualize it. And then after, you know, you the  user will navigate on the website and you'll say,  
137:40 "Hey, you know, I want to, you know, you'll  you'll click uh the dashboard button and then  
137:44 that'll be like, hey, you know, I want to go grab  this data and see how many records I have." So,  
137:47 what'll happen is we'll send that request  over to the server. The server has a bunch of  
137:51 um pre-built logic basically that says, "Hey,  when a user requests a specific dashboard data,  
137:58 I want you to call the database using this format  and then retrieve it." And then we'll say, "Hey,  
138:02 you know, we want the dashboard data." It'll  do this in a slightly different language,  
138:07 typically SQL, although there are no SQL databases  as well. And then it'll return, you know, that  
138:12 dashboard data. The server will do some additional  validation if it needs to, and then it'll just  
138:17 return that to the client. And so this is just a  loop that goes on over and over and over and over  
138:21 and over again. If the client updates something  like let's say we create an order, well, it's the  
138:24 exact same thing. We'll just send a request to the  server. The request will say create order. Server  
138:29 will be like, okay, cool. So here's how I create  an order. Then it'll take that request and send  
138:33 it to the database which will then add a new order  to like your orders table. So very very high level  
138:38 um overview of how basically these modern web  apps work. And I'm going to talk a little bit more  
138:42 about like the prevailing frameworks in a second  because there are now apps out there kind of like  
138:46 this service I'm using right now called Excaladraw  which does things both online but also locally.  
138:50 And you can really dive deep into that rabbit hole  if you want. But um without further ado, what's  
138:56 the actual development process look like? Well,  for us it's basically we just start local. So when  
139:00 we start local, we run both our client and our  server on the same computer, right? Our client is  
139:04 like our browser and then our server is just like,  you know, your your your computer running that  
139:08 process. Then we test and verify just like I did  a moment ago for my app. And then once we're done,  
139:14 we push to what's called a staging server, okay?  Which is sort of like an intermediary between  
139:19 um having everything on your machine and then have  everything accessible to your customers. Once it's  
139:23 on a staging server, we then test and verify.  And then once we're done testing and verifying,  
139:27 then we and only then do we actually push to  production, which is like the thing that um  
139:30 customers and and users can actually use. And so  the process is basically always just like um local  
139:37 staging production. And this is not super relevant  for our last step, but it'll become more relevant  
139:44 when we use things like GitHub repositories and  version control because there are changes that  
139:49 you can make locally. other people can be working  on changes on the staging server and then other  
139:53 people can be working on changes even on like  the production server. Although ideally you  
139:56 don't make live changes because a single live  change can like screw up your your whole work  
140:00 um like your whole app like you saw earlier. Uh  and then you can actually just spread it across  
140:04 have multiple people working on multiple things  simultaneously. So that in a nutshell is basically  
140:08 everything you guys need to know about hosting.  Um you do obviously need to pay for hosting for  
140:12 most services like you know you're not going to  be able to sign up to hosting or or site ground or  
140:16 whatever for free. This is basically you renting  a computer for the purposes of hosting your  
140:21 application on it and then having users interact  with it. And obviously every time some user goes  
140:25 to a website or whatever and there's a request  made to the host, the host consumes a little bit  
140:28 of electricity. There's like a little bit of of  a cost back there. So they um they arbitrage that  
140:33 and then they charge you some big multiple. And  there's like a bajillion hosts out there. There's  
140:36 like VPS's, virtual private servers. There's like  dedicated um computers or dedicated servers you  
140:42 could buy. There's like whole computers basically  or whole servers that you could buy. Or what they  
140:47 do is they take one server and they divide it into  like 50 servers and then slowly partition it out.  
140:51 There's like so many different ways of doing this  that it's not really worth us getting into. But  
140:54 suffice to say, this is just like the basic model  of um communication. Another thing that's really  
140:57 important to understand are databases. Now, for  all intents and purposes, like I talked about,  
141:01 these are basically just like a Google sheet.  If you guys have ever used a Google sheet or  
141:04 you ever used Excel or anything like that, the  best mental model I found so far is just to  
141:08 treat every database that you use like um a Google  sheet. So, just like a Google sheet, you'll always  
141:13 have like header row, uh, a header row, which is  basically just like the titles of every single  
141:19 column in your database. So, in this instance,  for instance, our table, which I just move my  
141:24 big head out of the way, is called sheet one down  here. Okay? And sheet one, okay, is just one of  
141:31 many possible tables in our database. We can make  100 more tables if we wanted to, but they're all  
141:35 going to be divided into like sheet one, sheet  two, sheet three, or whatever the heck we title  
141:38 them. Now inside of sheet one um we have uh three  columns here. The first column has a header name  
141:44 or a key of id. The second has a header name or a  key of date underscore created and then the third  
141:49 has a a header name or a key name of value. Now  underneath this sort of initial row which just  
141:56 defines the names of all of the various fields. Um  we then have the values themselves. And so this is  
142:01 a database for instance that contains a one record  or one row. If I duplicated this a bunch and then  
142:08 I don't know wrote some other IDs here because  this is going to be important. You can see now we  
142:12 have a database um with one table called sheet one  that now consists of four rows. Okay, the thing  
142:18 is depending on the type of database that you're  using. Okay, you need what's called a primary key.  
142:24 And a primary key is just something that's unique,  one column that is basically always unique from  
142:29 row to row that basically defines that row or  that record as like being super unique and not  
142:34 existing anywhere else. And so for the purposes of  this database, the primary key is going to be the  
142:41 ID because the ID is always going to be different  between each record. If you have two records with  
142:46 the same ID, obviously there's a conflict. If you  think about it logically, the reason why we need  
142:49 it structured in this way is because if we have  instructions in our server and the instructions  
142:53 are to pull a specific record from a database and  it's like, hey, I want you to grab the record with  
142:58 the ID 23FI23F2 and it turns out there's like 10  records like that. You'll return 10 records and  
143:05 how are you supposed to visualize 10 records? Like  that's going to, you know, it's going to screw up  
143:07 whatever your logic is unless you specifically  handle for that. And so these databases for  
143:12 the most part, most people that vibe code apps  suffered using the SQL database with um separate  
143:16 primary keys are typically organized in that way  where you have one sort of like starred column.  
143:21 And so I'm just going to make this one the starred  column. It's the yellow one. And then these here  
143:24 can be the exact same. Like you could technically  have 100 records all created in the same day,  
143:27 but since they have different IDs, they're all  different records, right? Same thing with the  
143:30 value. Okay? So just like in a Google sheet or an  Excel file, you can actually like filter things  
143:35 out. Um you can do the same thing inside of a  database. So if I go back to our little example  
143:39 here and then um I add a filter which I can do on  sheets really easily just using a little hotkey.  
143:44 You can also just go to data and then you could  click on create a filter if you wanted to follow  
143:49 along this demonstration. What you can do is  if you click on this little filter button you  
143:53 can see that you can now just filter specific  records. Like I could for instance only grab  
143:57 um you know this record here 23FI. Okay. And if  I did so it would remove all of the additional  
144:02 records here and then I'd only be looking at this  data. And so every time that a server makes a  
144:06 request to a database, that's basically what it's  doing under the hood, it's just doing this in like  
144:09 a fully automated fashion. Likewise, I can filter  based off of fields that are not my primary key.  
144:13 So maybe in date created, what I wanted to do is  I wanted to grab all records that have 2025-10-3.  
144:19 So as you can see here, now I'm grabbing multiple  records, okay? But each of these have different  
144:23 primary keys. And this is just a way that I could  filter stuff. It's just like using a search bar,  
144:26 right? You can do the same thing for value, do the  same thing for whatever the heck else you want,  
144:30 um, you know, in your database. So basically  every time that you send a request from you know  
144:34 your client your browser by clicking a button  you send that request to the server the server  
144:38 interprets it. What it's doing is it's sending  another request to a database which um you know  
144:42 is basically just acting like a big Google Sheets  filter. Now the cool thing is there's two major  
144:47 types of databases right now. The first is called  SQLbased. So SQLbased is divided into a bunch of  
144:53 different popular versions. I mean really there's  like a thousand but three most popular ones right  
144:56 now are my SQL postgrql which is probably the  most popular one and then SQLite and then you  
145:01 also have what are called NoSQL databases like  MongoDB and Cassandra. How I've always worked with  
145:06 the differences is SQLbased databases basically  fix the column headers and so it's always going  
145:13 to be the same column headers for all the data  and all of the records that go into a database.  
145:18 And so just like here we have those three column  headers ID, date created and value. Basically,  
145:22 every single record in this database will always  have an ID because it's the primary key and can  
145:27 optionally have either a date created or a value  or it could have both. And so, um, you know, for  
145:32 instance, you might get something that looks like  this with a date created and then no value. And  
145:37 that is what's called a valid record. Likewise,  you can get something that looks like this with no  
145:42 uh, you know, obviously the IDs are going to  be different. No date created, no value. And  
145:45 you can also have something that looks like this.  Okay? And you know as long as you have a primary  
145:51 key which denotes that this is a separate record  um every other field is optional but it has to  
145:56 stick to this structure. Now no SQL databases are  a little bit different in NoSQL I'm just going to  
146:02 create a new sheet here. Basically what we have  is rather than having just like one header row up  
146:07 at the very top I want you to pretend that every  single item in your database has its own header  
146:13 row. And so the first record here will have ID,  date created and values. So let's say the ID is  
146:19 this, date created is this, and then the value is  this. Okay? Or I don't know, let's just make it a  
146:25 little bit easier. Let's just call this uh we have  a record here called first name. Probably makes  
146:30 more sense. And then this one's Nick. So um you  know what we have here is we have one record with  
146:35 an ID, a date created, and a first name. Well, for  the next record, instead of just like sticking to  
146:40 this format, you could have whatever the heck you  wanted. So I'm actually just going to paste this  
146:44 down here. And then I'll say this record actually  has um a last name instead. It has a date created  
146:56 column and then it also has a date modified  column. Maybe this date modified as 29th. If I  
147:03 added another record here, okay, maybe what I have  here now is I have a full name. This one's Nick  
147:11 Sarif. Then I also have my little date created a  little bit over to the right. and maybe there's  
147:15 no date modified. And so basically what occurs in  a NoSQL database is every single record has its  
147:22 own data structure rather than everything being  100% fixed and looking the exact same like we  
147:29 had with our previous uh SQL database. You know,  every record will have an ID. Then it's optional  
147:34 whether they have date created or value. But  it's not like they could have another, you know,  
147:37 it's not like this row could have another value  called uh first name called Nick, right? Unless  
147:42 we explicitly define that up here as a value  or a header that exists on all. Okay. Whereas  
147:48 this would be empty empty empty empty empty empty  empty and then this would be nick. Okay. It's not  
147:52 like we can just add additional um features or  or types of data. In a NoSQL database you have  
147:58 total freedom and flexibility. And it's a lot more  like like a Google Drive folder where you're just  
148:03 storing a bunch of different documents and every  single document could look a little bit different.  
148:07 Now, obviously for our purposes, we want a simple  way to like call a database and then retrieve all  
148:12 records that have a first name Nick Sariah or that  have an ID of of something. Um, you know, we're  
148:16 going to use SQL for the most part. But there are  some instances where NoSQL makes more sense. Um,  
148:21 and if you want to look more into that, just look  for the difference between like relational and a  
148:24 non-reational databases. Um, and then maybe  you could have your agent tell you. Next,  
148:28 I want to chat languages and frameworks for people  that might not be super familiar with them. So  
148:34 there are three basic languages used these days in  modern SAS app um sort of development engineering  
148:41 and I'm going to draw them as little bubbles and  I'll show you how they all interact. The first  
148:45 is probably the oldest which is hypertext markup  language. The next I just copy this is a little  
148:53 bit newer and it basically took the internet from  just a bunch of ugly static web pages to pretty  
148:58 dynamic and interesting web pages CSS. And then  the third is how we got interactivity in the web,  
149:05 which is through JavaScript. Now, I'm not going  to give you guys a big historical lesson on each  
149:09 of these because to be honest, I mean, I just  be googling this stuff anyway. Um, but suffice  
149:13 to say, HTML is actually like basically the only  thing that you need in order to run a website. CSS  
149:21 is sort of the design layer that makes things  really pretty. And then JavaScript is just  
149:25 what allows you to build interactivity into pages  like you clicking a button and then a new element  
149:29 appearing or something. And so to show you guys  this HTML side, if you head over to info.sern.ch,  
149:35 you can actually see the very first website ever  made. And that website is called the worldwide  
149:40 web. This is literally describing the worldwide  web. You know how all websites or many websites  
149:47 have www preended? That stands for worldwide web,  which is a framework that people put together god  
149:52 knows how long ago now. probably at least 40  years ago. And so basically this explains how  
149:56 the internet works. Okay. Now obviously you don't  need to have gone to this website to know how the  
150:01 internet works more uh fundamentally speaking.  I think every kid just inherently gets it when  
150:05 they are given their first iPad with Coco Melon  on it. U no jokes aside uh the worldwide web is  
150:11 is a website that's entirely built in HTML. And  you can see because the file ending up here is  
150:14 HTML. If I open this up you'll see the actual code  of the website down here. And it's very simple.  
150:19 There's a body, there's a header, there's an H1,  which stands for worldwide web, an AT, which is a  
150:25 bunch of links, and then P tags and and stuff like  that. Now, worth noting that this is a website  
150:30 with no CSS, which is sort of the next level up.  Now, CSS is how you change the style of the page.  
150:36 It's how you make things bigger or smaller or  or sexier or whatever. So, if I go style here,  
150:43 I can actually change the font size to like 300  pixels and I can make that bigger. This HTML is  
150:51 just that core fundamental um kind of structure,  the scaffold of the page. This style is now CSS  
150:57 and it's where I'm like applying some additional  information to change the size. I could change  
151:01 the color and I could do so on and so forth. I  could I could change things quite a bit. Now,  
151:04 there is that one final piece which is JavaScript.  And JavaScript nowadays is done mostly through the  
151:10 console. Inside of the console, you can actually  run um JS. So what I could do is I could go  
151:15 document query selector and then I could look  for all all H1's elements and then what I could  
151:21 do actually remove this from the page um just by  clicking or writing document.query selector H1-E  
151:28 and this is an example of that interactivity that  I was talking about. Basically with JavaScript  
151:32 what you can do is you can I'm realizing that  my head's in the way here so you can't see that  
151:36 most recent function. Um basically what JavaScript  allows you to do is allows you to select specific  
151:41 elements on the page and do whatever you want with  them. You can make them bigger, you can make them  
151:44 smaller, you can apply styles, you can create new  HTML. I mean, there's like a bajillion different  
151:48 things you can do. Okay, so those are the three  programming languages. And hopefully we're all  
151:51 on the same uh page about how these two or three  interact with each other. The thing is, obviously,  
151:57 if you design a website or a web app using HTML,  CSS, and JavaScript, there's like a million  
152:03 different ways you could do so, right? You could  write whatever website you want, apply whatever  
152:06 styling you want, and then obviously apply  whatever interactivity you want, and that's great.  
152:10 But the internet thrives off standardization,  which is where things are done in a repeatable  
152:15 and sort of consistent way. Because of this,  the more standardized the internet tends to be,  
152:21 the better solutions that people can build on top  of pre-existing solutions. And that's really the  
152:25 whole idea of programming languages in the first  place. So in addition to this idea of you know  
152:30 languages as um standardization you also now have  this concept of frameworks as standardization. And  
152:41 so frameworks are a level of abstraction that  are built on top of languages. And what we do  
152:46 is we actually just take these three sort of Lego  blocks and we just construct them in a specific  
152:51 way essentially. Okay. And right now there are a  bunch of different frameworks out there. The most  
152:57 popular ones I would say are next.js which is the  framework that we just used to define and build  
153:05 our current app. And you can see that that include  a lot of that included a lot of opinionated  
153:10 um framework based stuff for like the folder  structure and components and hooks. These are all  
153:15 things that are basically a handme-down directly  from our framework. You also have um vit right  
153:21 now is super popular and you'll see a lot of apps  created with vit and then on the other side you  
153:26 also have view or really like nux.js and there are  varying levels of abstraction here like nux.js is  
153:32 actually an it's like a second framework built on  top of another framework called uh react which is  
153:37 kind of funny. Um nux.js is built on top of view.  Uh you know there are a lot of like dependencies  
153:43 and stuff and if you guys remember earlier I said  we stand on the shoulders of giants. really do uh  
153:47 because everything nowadays is built off of  something somebody else built which was built  
153:50 off something somebody else built all the way  down to where we were programming computers  
153:53 with little ones and zeros on pieces of tape. And  so in addition to you know our chosen tool and  
153:58 this is the tool I'm going to be using for the  most part throughout this course. NextJS being  
154:03 essentially the way that we coral these three  programming languages together. Nex.js JS is also  
154:09 an opinionated framework that um you know helps  us organize things like hosting, helps us organize  
154:16 things like how to do you know database calls and  whatnot in a standardized way. You guys remember  
154:22 earlier we talked about middleware authentication  middleware uh has a standardized interface for  
154:26 authentication middleware is a standardized way  to put together routes which are just the names  
154:31 of like um website endings. So excalagger.com/ I  don't know settings that's a settings route for  
154:36 instance and so on and so forth and what's really  cool is because a lot of people have used these  
154:41 frameworks before you know like NexJS isn't Neo  was developed at least half a decade ago if not  
154:45 more I built tons of projects with it before we  had the luxury of agents to do that for us because  
154:50 of all this we had a lot of pre-existing projects  out there on the internet that were built with  
154:53 this and these were all used in the training sets  of these large language models like you know opus  
154:58 gemini and so on and so forth which means make  a long story short If you use a framework, you  
155:03 usually get significantly higher quality results.  Now, frameworks obviously have some baggage as  
155:07 well. The more opinionated things you put into uh  you know a design system or whatnot, obviously the  
155:14 more lag that you can have, the bigger the files  can be and and so on and so forth. But for the  
155:19 most part, these are a good balance of like  speed, stability, reliability, and then also  
155:23 security. You know, because these frameworks have  existed for a while, people have put together some  
155:27 pretty simple and easy ways to economize um our  security and our ability to build apps. Okay. So,  
155:32 am I saying that you have to like go and  learn every single thing in HTML? No. CSS? No.  
155:36 JavaScript? No. Um, do you have to read through  all the NexJS docs? No. Although they are they  
155:40 are pretty cool if you wanted to. None of these  things will necessarily make you like a better  
155:44 vibe coder. But I think it's important to at least  be able to like look at the Lego blocks and see  
155:48 how they fit together because if you don't, then  you're a lot more liable to produce what I think  
155:52 a lot of people nowadays are terming as sloop  out there on the internet, which are like simple  
155:54 vibecoded apps that aren't really well thought out  and then interact with the rest of the internet in  
155:59 pretty stupid ways. The last thing I want to talk  about is version control. And to make a long story  
156:05 short, version control just answers the question,  how do I store successive updates of files on a  
156:13 server as efficiently as humanly possible? Let me  give you a brief little example. I'm sure you know  
156:18 back in high school or college or whatever, those  of us that have gone there in the last few years,  
156:23 you've run into situations where, you know,  you're tasked with writing an essay. And so,  
156:28 you know, you write your essay probably really  late in the middle of the night with a bunch  
156:31 of Cheetos beside you. At least that's how I  used to do it. And then, um, you save it as,  
156:35 I don't know, essay docx, let's say. You know,  it's like a it's like a Google doc or something  
156:39 like that file. And you do it locally in your  computer, maybe Microsoft Word, okay? And then,  
156:44 um, the next day, you know, you wake up and then  you give it to your friend, and your friend gives  
156:49 you some feedback or something like that. Okay?  So what they do is they they download the file  
156:53 and then they give it back to you and then it's  called essay with feedback.docex. So what you do  
157:00 is you take that file but not wanting to lose the  previous two files what you do is when you make  
157:05 your updates to it okay you don't save it as essay  with feedback you call it essay 2.d dox and then  
157:12 the cycle just repeats itself three or four times  then you have essay 3.d dox you have essay 4.x doc  
157:20 x and so on and so on and so forth until you have  essay final superfal important docx. Okay. And  
157:29 before you know it, you know, your whole downloads  folder is a mess. I'm sure we've all probably had  
157:33 situations like this before. Um I know, god  forbid, I I used to have situations like this  
157:37 all the time. Now, that's for writing things like  essays. I want you to imagine instead of essays,  
157:43 what you're doing is you're working on an app like  we're doing. You'll have app one dot whatever,  
157:48 app two dot whatever, app three super important  final whatever. Obviously, these things can get  
157:53 out of control really quickly. What compounds  the issue is this idea of needing to share files  
157:57 because this super janky ass method, this could  work if it's just you and you understand how you  
158:01 name your files and stuff. You could build like a  naming convention, right? But what if you want to  
158:04 share this with like 20 other people and 20 people  all simultaneously need to be able to work on the  
158:08 file that is the most updated without causing  conflicts and stuff like that and so on and so  
158:12 forth. That's really the whole idea behind what  initiated version control. And so what we do is  
158:17 we basically have again standing on the shoulders  of giants um built systems out there that store  
158:24 not only the file itself but store the changes  between files and then just let you access them  
158:30 by you know calling a little URL or something and  then pulling the file as well as a list of all the  
158:35 changes um that have historically occurred. So  that at any point in time, you know, you could  
158:40 download, let's say, this original file, but  then you could jump over here if you wanted to,  
158:43 jump over here if you wanted to, jump over here  if you wanted to, jump over here if you wanted to,  
158:46 and so on and so forth, all without just needing  to like pollute your, you know, file system and  
158:50 stuff like that. What's more, at any point you  could also branch out and make additional changes  
158:54 here, however much you want, called, I don't know,  new essay. And then you could share that new essay  
159:03 uh with other people. Okay? Or what you could do  is you could make the changes here and then you  
159:08 could actually take these changes and then merge  them with another pre-existing file to get some,  
159:13 you know, new better essay hypothetically. Um, you  could do that with apps, you could do that with  
159:17 essays, you could do that with everything.  People use version control to literally do  
159:21 everything nowadays, research papers and so on  and so forth included. So the major place where  
159:26 most version control happens now is on GitHub. And  what we do is we've taken this concept of like a  
159:32 folder or a workspace and we've standardized it  into one called a repository. And so this idea  
159:38 of a GitHub repository is very popular. You go  into github.com. This is what the website looks  
159:43 like. Okay. And you can see on the left hand side  I'm signed in here. I have a bunch of different  
159:49 um repositories. I didn't even realize, but  I have a bunch of people here that are like  
159:51 starring my repos and stuff. That's hilarious.  Um I have a bunch of different repositories  
159:55 here like worklog make money with make gyro  app.1c.copy.com copy.com, proposal generator,  
160:00 and so on and so forth. And these are all  basically different projects that I've built. Now,  
160:03 what's cool is if I click on any one of these,  okay, you can actually see every change that I  
160:08 have made to this over the course of god knows  how long. This is basically over the course of I  
160:13 don't know 346 different changes. If I click this  button, the wildest thing, so I can actually go  
160:18 through and I can see literally every change I've  ever made to this thing. And at any point in time,  
160:23 okay, I can actually open up the details and then  the code itself to adjust what's going on. I could  
160:30 uh restart my, you know, project from here. I I  mean, I could do anything. I could share this with  
160:35 other people really easily. Um, and so on and so  forth. So, I say this not because I want this to  
160:40 be a comprehensive guide to GitHub. Obviously,  it isn't. I just like gave you guys a very high  
160:43 level overview. Um, but in a nutshell, all this  is is just a way to store and and and save files.  
160:48 And so whereas with our last couple of projects  with our portfolio site and then the um app that  
160:53 we just built, I wasn't using GitHub for any of  it, from now on, what we're going to do is we're  
160:57 actually going to take our big file structure  that we're developing locally and we're going  
161:01 to constantly push it to GitHub so that we both  have like a saved version over there if anything  
161:05 locally occurs, but also we have the ability to  um share it with others. And then services that  
161:10 do deployment nowadays like Netlife, Versel, and  and other platforms out there, they can often just  
161:15 pull directly from the GitHub. So I don't have to  like drag my folder in. I could just say, "Hey,  
161:19 here's my GitHub repo." And then it'll just do all  the math and stuff for me. And then I don't even  
161:23 have to like rebuild it on Netlefi. What I could  do is I could just push it to this repo, make a  
161:27 couple of changes, and then Netifi will have  like a little hook that automatically rebuilds  
161:30 the site every time a change occurs. And this is  how you get like different pipelines set up with  
161:34 like production, staging, local, and so on and  so forth. You can also create different branches  
161:38 branches. I guess I have one here called chest day  workout. That's pretty funny. Must have crushed  
161:42 that chest day. Um, and then just have agents  manage like everything to do with uh with whatever  
161:47 project you have. So, for instance, this is like a  little work log that I set up that basically just  
161:51 like a list of things that I've done. I use it to  track my own my own work. And uh I developed this  
161:55 in just a few minutes with uh with claude code  I think like two months ago essentially. The  
162:01 last thing I'm going to mention here is everybody  gets a GitHub profile. And so there are a lot of  
162:04 people out there that are have very popular GitHub  profiles. This is usually where most open source  
162:08 lives. So, you know, if you develop a really cool  new idea or new framework or something, you make a  
162:13 GitHub repository and then you share it with other  people and then other people can also make changes  
162:16 and then improve it over time. Um, that's what  a bunch of really popular repos tend to do. If  
162:21 you guys have checked out like the open claw one  and and stuff like that, that's where they were  
162:24 getting um most of this this from. And then with  agents, what you could do is you could actually  
162:27 just have parallel instances of like 10,000  operating in the background, constantly scanning  
162:32 the workspace and then making micro adjustments  and edits in a very interpretable and accountable  
162:36 way. So the future of software development is  basically that decentralized GitHub repo pushing  
162:40 and pulling and stuff. Okay, cool. I think we've  now done enough of like the traditional software  
162:44 engineering concepts for you guys to know a lot  more about how to build these things efficiently.  
162:48 Um, what I want to do now is just talk really  briefly about a couple of the different ways you  
162:51 could use agents to streamline your work. I know  we touched on this briefly with the repo stuff a  
162:55 second ago, but I just want to zoom out a bit and  then just look at anti-gravity and look at some of  
162:59 the the potential and the functionality there in.  Okay, so if I just go back to anti-gravity here,  
163:03 you know, I still have that old site set up. Um,  I should be able to clear this just by going to  
163:07 new. Um, one of the simplest and easiest ways to  multiply your ability to work on a specific repo,  
163:15 which is just what I'm going to call our workspace  or file folders from now on, is basically just  
163:20 running multiple different separate functions  simultaneously. And so, you know, I could say,  
163:26 hey, do XYZ, and then I could just wait until this  whole thing is finished. And then I could give it  
163:31 some more instructions. Hey, thanks. Let's now do  ABC. Or what I could do is I could make use of the  
163:38 ability to start new conversations. And I could  actually run multiple of these um simultaneously.  
163:47 So I think we're all familiar with just using the  little agent tab on the right hand side. But in  
163:52 order to parallelize your work and then work a  lot faster, what we can do is we can zoom out of  
163:56 a specific workspace and then actually just look  at like 3 4 5 10 100 agents working simultaneously  
164:01 for us and then just orchestrate their actions.  This can sound pretty bonkers if it's the first  
164:05 time you've done something like this, but this  sort of functionality is actually built into  
164:09 anti-gravity and uh it's also built into things  like cloud code and and stuff like that as well.  
164:13 The way you do so is with this agent manager in  the top right hand corner. So if I click the agent  
164:17 manager, what we've done now is we've zoomed out  of a particular workspace. We actually have all  
164:21 of our workspaces here listed on the lefth hand  side. And you can actually see every conversation  
164:26 you've ever had inside of each workspace as well,  which is really handy. So, this one's improving  
164:30 dashboard UX. This one's listing directories and  stuff like that. Here's a search I just said a  
164:34 moment ago, or I'm pretty sure I just said like  do ABC or something like that cuz I wanted to like  
164:38 have a bunch of these that I could quickly jump  through. You can see in this case just ask me,  
164:41 hey, what the heck is ABC? What are you crazy? And  so on and so forth. This is some other thing that  
164:46 I've developed here. This is some other thing  that I've developed. And so on and so forth.  
164:49 So, I mean, like, you know, this is kind  of cool, right? You basically can just go  
164:51 top to bottom and then see at a bird's eye view a  bunch of different conversations that you've had  
164:55 with different agents. And you can do this across  workspaces. You can start new conversations across  
164:58 workspaces. You can jump into a specific workspace  using the editor, right? Um, and so on. The really  
165:04 cool thing though is if you go to this inbox in  the top lefthand corner, what you can see is you  
165:09 basically get a list of all of the messages that  AI is sending you. And the reason why this is so  
165:14 valuable and so cool is because if you make use of  the pending feature, which is basically where you  
165:21 don't let AI do something unless you've cleared  it, what you can do is you could use this little  
165:26 pending filter and then just have a giant list of  things that like 20 different agents are working  
165:31 on simultaneously and then they're just like  serving you um you know the results and asking  
165:35 you, hey, do you want me to do X, Y, and Z? Okay,  so let me show you a really quick and and simple  
165:39 example. So earlier I had this website builder  repo, right? If I just open the editor here,  
165:44 I had this big Gemini.MD where I was basically  showing you guys how to create a website and  
165:49 then I pumped through like two or three of them  simultaneously. And I hid this from you at the  
165:53 time because I didn't want you to obviously get  super overwhelmed. But if you open the agent  
165:57 manager, okay, you can see that you can now start  new conversations in what's called a playground,  
166:02 which is an independent workspace, perfect for  quick prototypes or following curiosity. Then you  
166:06 actually just paste in this multiple times. So you  could start a conversation by going in, paste this  
166:13 in, start another conversation by going in, paste  this in, and start another conversation by going  
166:18 in, paste it in. And now in the playground side of  things, okay, you now have multiple conversations  
166:23 that are occurring where it'll basically just ask  you a bunch of information. I'm just going to say,  
166:28 um, choose all of this yourself, high-end SAS,  for a demo I'm making. You can just paste this in.  
166:38 You could paste this in. You could paste this in.  And then you could paste this in. And what I have  
166:44 now, okay, is I have multiple agents that are all  operating simultaneously and independently for me.  
166:51 This first one here is called cinematic landing  page build. The second one here is called building  
166:54 cinematic landing pages. And then you guys can't  see these other two, but it's cinematic landing  
166:57 and cinematic landing. And basically what they're  going to do is presumably they're going to create  
167:01 different folders for each of these and then build  all of these websites simultaneously in parallel.  
167:07 Right now what I have think about it is I've  made one command up here and it's split rather  
167:13 I've made four different commands up here but  basically what I've done is I've split and now  
167:16 we have these agents all running simultaneously  and inside of this playground I have the ability  
167:21 to like get a highle overview what of what um  they're all doing just by mousing through each  
167:26 okay and you know sometimes you're either going to  run into token limits or there's going to be some  
167:31 issue so agent terminated due to error here when  this sort of thing occurs you can reply and you  
167:36 can also just use this agent conversation for it  to show you more or less like what it's currently  
167:40 working on. So, you know, if I go to inbox here a  moment ago, it basically said, "Hey, like what do  
167:44 you want me to do?" That that that was a message  that the agent sent to me asking me for approval,  
167:48 right? I can actually just see all these things  at a very high level overview just by constantly  
167:52 being on this page. And then essentially all I'm  doing is instead of like me being responsible for  
167:57 shephering one agent simultaneously, sorry, one  agent through a process, I can shepherd multiple  
168:01 simultaneously. And this is the difference between  somebody that's like building one project and then  
168:05 somebody that's orchestrating multiple projects.  You do run into some issues with like context  
168:09 switching, which is a productivity term that  basically applies to when you're trying to do  
168:12 multiple things simultaneously or a little bit  less efficient at um every one individual one,  
168:17 but it's sort of like monitoring like three or  four different CCTV uh uh pieces of footage if  
168:22 you're like a security guard or something in a  mall. You're basically just zooming out. You're  
168:25 watching three or four of these things and  you're stepping in anytime that a particular  
168:29 person needs you. Alternatively, it could be like  you're a manager at some big office with a bunch  
168:34 of cubicles that you're just walking up and down  the halls. When somebody needs you, they're like,  
168:37 "Hey, you know, Nick, can you help me with X, Y,  and Z?" And so, in this way, you're capable of  
168:40 orchestrating significantly more total output or  work than you would be able to do yourself. So,  
168:45 it's really worth using hotkeys um if you  wanted to start building things in this way. And  
168:50 uh generally speaking, believe it or not, instead  of just building one app, I will usually build two  
168:54 or three apps simultaneously now with something  like the agent manager and an inbox. and then  
168:58 I'll just like have them all run in parallel.  Um, the reason why is because then that way I  
169:02 could build multiple versions and then I can  just pick the version I like the best. Once I  
169:05 pick the version I like the best, I drop the other  ones and then I just proceed with that one. Um,  
169:08 but yeah, hot keys are pretty important. So you  can find those just by mousing over each of these  
169:12 buttons. So inbox is command I for instance,  uh, if you scroll down to playground here,  
169:17 you can create a new instance just by holding  controlN. And so a lot of the time I'm just  
169:21 jumping between new conversations and then my  inbox over and over and over and over again.  
169:25 Um, you can also jump back to that editor like I  talked about earlier just by going command E. This  
169:29 will jump you into the particular folder where  the conversation and kind of work is happening.  
169:32 So this one's called glowing cosmos for instance.  At any point in time if you're operating inside  
169:36 of this playground, you also have the ability  to um move this into a specific folder. And so  
169:40 this is occurring right now in like a temporary  folder that uh Gemini and anti-gravity just puts  
169:45 together anytime you do this sort of thing.  But you can also just click this button and  
169:48 then you can select a destination folder for all  the files in your project. So I could, you know,  
169:53 use the playground to sketch out three or four  different ones and then when I find one that I  
169:58 really like, I could then use the um create a new  folder thing to like actually go in add a folder  
170:02 and then just work in like a dedicated workspace.  They also have a feature called follow along with  
170:06 agent in the top rightand corner. So if you  follow along with agent um basically it'll  
170:10 pull you back in to what the agent is doing at  every step and then allow you to see the various  
170:14 um you know pieces of work that that agent it's  worth keeping this page pretty clean. So, I'm just  
170:19 going to proceed with whatever the heck it asked  me for. And then over here, now I basically have  
170:23 these two main threads that I'm running. You can  see the same thing's occurring back here and then  
170:29 also back here. And although kind of per unit,  it'll still take the same amount of time to get a  
170:34 website out because, you know, this is operating  at the speed of um whatever Gemini 3.1 Pro is,  
170:39 if you think about it, I'm technically doing four  times the amount of work um simultaneously across  
170:44 all four. I don't usually recommend having  more than two or three of these operating. I  
170:49 just wanted to show you guys how cool it could  be if you have four. I actually have five now  
170:52 cuz I'm proceeding with some um agency operation  thing I did a while back. Uh just because I find  
170:57 like your attention gets pulled away a little  too much. But um yeah, you can see here in the  
171:00 top rightand corner because I'm following this  anytime it'll make a substantial major update. I  
171:03 can actually see the code that it's working with  and so on and so forth. I usually don't actually  
171:07 click that though. I just go to my inbox and then  I seem you know do you need anything from me? if  
171:11 there's any specific feature or whatever that  they want, then I can I can jump right in. So,  
171:15 let me run this for a couple more minutes, get  a bunch of websites, and I'll just show you  
171:17 what they all look like. Okay. And our landing  pages are now starting to um be finished. So,  
171:22 we could actually access the Vit server. This  one's using a different framework called Vit that  
171:26 I covered earlier just by pasting this in. You  can see we now have one of the websites. Now, I  
171:31 should note that this looks pretty similar to the  1se secondond copy website I had set up because  
171:35 uh I think I just asked it to build something that  was like very very similar visually speaking. So,  
171:38 we had that little heartbeat. we have this little  thing, but you know, I'm like trying to build  
171:41 it with a particular template, right? So, I'm  just going to instantiate a bunch of versions  
171:45 of this template, and if I like them, then I'll  go with that. This second one here is done. Um,  
171:48 because I didn't specifically say, "Hey, I want  you to run the server and then open it." You know,  
171:51 this one's asking me to do this terminal command,  npm rundev. Um, so I'm just going to have it do  
171:56 it by saying run it and then it'll open it  for me and then I can take a peek. Here it  
171:59 is. Let's copy this over. Paste this in. Cool.  And now we have another one called Aether Core,  
172:04 which is very clean. I really like this one. Damn,  that's really, really sexy. You know, there are  
172:09 some issues here with some of the fonts and stuff,  but hopefully you guys see like I was capable of  
172:13 generating four of these simultaneously. This one  here is still under development. As you can see,  
172:17 there's a hero section and it's scrolled down, but  it is running live on a server and it does have  
172:20 that cool little bar up top. And this one here had  probably my perfect instantiation or my favorite  
172:25 of like this slow scrolling card like that to me  looks really sweet. Um, it's just rerunning it  
172:29 because it's making some additional changes, but  yeah, pretty cool, huh? So, you can actually build  
172:33 multiple apps simultaneously. And in this case,  you know, I gave it some pretty static resources.  
172:37 And I was like, "Hey, I want you to use like  this kind of noise layer in the background and  
172:39 stuff like that." You could also just have all  of them be different. And the way you do that  
172:42 is you just provide like a prompt that has some  inherent randomization. Basically says, "Hey,  
172:47 I want you to randomly select like, I don't know,  a background from this online background service."  
172:51 And then it'll use different backgrounds. Hey, I  want you to automatically select a different type  
172:54 of style using this thing. You know, instead of  high-end SAS, I just want you to come up with a  
172:58 different one every single time. And then it'll  come up with like a 100. So yeah, in this way  
173:01 I've I've uh successfully built something like  15 websites simultaneously at the same time. I  
173:06 found like the top two or three that I like. I've  proceeded with those. I filtered out the ones that  
173:10 I didn't and then I've had like a really really  high quality product. And I've done all that stuff  
173:13 in like 10 maybe 15 minutes or so. Um obviously be  careful with token usage here because it's always  
173:20 going to be a trade-off between the time it takes  to build something and the total cost on your end.  
173:24 This is just a pretty straightforward money for  time trade, but running multiple agents in this  
173:28 way um simultaneously can be pretty cool.  Okay, now that you guys know more or less  
173:31 everything you need to know about both modern  software development, um, hosting, databases,  
173:35 frameworks, and all that stuff, plus how to use  agents to pour fuel onto the fire, let's build  
173:40 product number three, which is a lead generation  SAS app that scrapes, enriches, and even queue up  
173:45 um, the sending of fully automated emails. This  is very similar to a structure that I use in  
173:48 production myself in my own business, Leftclick.  We work with multi-billion dollar portfolio  
173:52 companies and essentially connect to the outbound  systems on a team to help them send emails,  
173:56 scrape uh contact details, and then personalize  outreach at scale without needing like a million  
174:01 bajillion hires in order to do all that for us.  So this time, instead of just diving right into  
174:07 the app development via prompt, I wanted to take a  few moments to sketch out what an actual app might  
174:11 look like. And there are variety of different  names for this sort of thing. Um, in my case,  
174:15 I'm just going to call it planning. You could  call it some form of like UX or site design  
174:19 or navigation design or whatnot. In short, um I  know for sure this app is going to need a landing  
174:25 page. So here I just said landing page sexy on  it. We're going to want some navbar, some main  
174:31 value proposition, a social proof section, some  how it works section, some case studies, a CTA,  
174:37 and then a footer. This is pretty standard, right?  It's basically the exact same thing that we did  
174:41 with Content Orb. After that, we're going to want  some sort of login and/or signup page. And so this  
174:46 is the O form in Superbase where you enter your  email, enter your password, and then it signs you  
174:50 up. Now, what I imagine will probably happen after  is once you're signed in, um, we'll need some form  
174:57 of configuration page cuz I want this app to be  able to be usable by people like you, people that,  
175:02 um, you know, I don't want to have to necessarily  build for. Instead, I want there to be a space for  
175:05 your instantly API key, your anthropic API key,  your OpenAI API key. Uh, I want you to be able to  
175:11 configure the prompts that are being used by the  AIS to do personalization as well as have defaults  
175:15 and stuff like that. And basically, I just want  that to be like your onboarding. It's like, hey,  
175:18 do you want to get started with the scraping?  Just enter your API keys here. And there are a  
175:22 couple of other API keys that I'm going to need,  actually. But, um, in addition to that, I also  
175:26 want like brief little one-click guides to show  people how to do all of those things. And then  
175:30 finally, I want the actual scraping page itself,  which is just going to be a form where you put in  
175:34 some basic parameters. And the way this is going  to work, for those of you guys that don't know,  
175:37 there's a scraping service out there called  Appify. And we're basically just going to wrap  
175:39 around Appify. If we can't find enough leads, then  we're going to scrape Google Maps. And then we're  
175:45 basically going to combine both of these into some  sort of like enrichment personalization flow. Um,  
175:50 and assuming that, you know, we've enriched and  personalized, which, you know, may not necessarily  
175:54 mean something to you guys if you guys aren't in  the marketing space, but I'll run you guys through  
175:57 it all um throughout this demo, then you'll just  get all the leads that you just scraped in a  
176:01 simple CSV with some additional column that has AI  based personalization. So, we're going to need the  
176:07 first name and email at minimum. And then I think  the last thing to talk about is that orders page  
176:10 or just going to be like a like a history page  where you could see all the orders that you've uh  
176:14 scraped and sent before. And then maybe I'll also  do plus specific order pages for each that have  
176:20 like I don't know a bunch of descriptions on them.  Okay. So, you know, I think uh probably a real UX  
176:25 designer would crucify me uh in terms of the ways  that I've laid all this stuff out, but you know,  
176:29 seems to work pretty reasonable for me. Um, what  I do next is I basically just take a screenshot  
176:33 of this and then I always have this on hand  cuz you know I want to like know what my app's  
176:38 going to look like. Then I go into anti-gravity  and then what I do is I um say something like,  
176:41 "Hey, I'm making an app that helps people scrape  leads. Turn this into a brief description of an  
176:45 app that includes all the important components  and I actually just screenshot and I send it to  
176:48 them." This is back when I had it like vertical.  I've since made it horizontal. Then scrolling down  
176:53 a little bit, you can see we've now changed  it into the description. Okay. And then over  
176:56 here we have the updated description with just a  bunch of, you know, plain text language here. So,  
177:01 we're probably going to need more. I mean, like,  it's not that big of a deal. Then I made a couple  
177:05 of changes and I said, "Hey, here's some scrapers  that I want you to wrap around." Okay. And now my  
177:09 claw code instance has generated this prompt.  So, that's good for now. What I want to do next  
177:13 is open a new folder cuz I've copied this and I  want to call this lead scraping SAS. I'm going  
177:18 to create and I'm going to open. Now, in our  anti-gravity instance, we have a plain folder  
177:22 here, lead scraping SAS. And so what I want to do  next is two things. First, I'm going to need to  
177:29 feed in that prompt. And second, I want to feed  in um a description, sorry, an example that we  
177:35 could use to basically inspire Gemini. Now, I went  on dribble and I found something that looks like  
177:39 this. This looks pretty reasonable to me. Um keep  in mind like the design and stuff like that is  
177:43 pretty sleek, but obviously they have a lot more  components here. Data transfer. I mean, I don't  
177:47 really know why I would ever need a data transfer  component, but you know, there's like a bunch of  
177:50 buttons here and stuff. And I think we could make  this work. Um, you know, our graph may not look as  
177:55 nice, although I think we could actually make it  look as nice if we really wanted to. But for now,  
177:59 I'm just going to copy this image. I'm going to  go back over here. I'm gonna start by pasting that  
178:02 in. Okay. So, we're going to start just by pasting  in um this image. It looks like we're copying  
178:09 this, but it's not actually making it all the  way to our dashboard. It's kind of annoying. So,  
178:11 I think I'm going to have to take a screenshot  of the actual app itself. I think sometimes  
178:14 Dribble does like anti-creenshot protection or  something. Keep in mind that we're not stealing  
178:19 this or anything. We're just going to use this as  like inspiration. That's all. go back here, paste  
178:22 that in. Then I'm going to say um I'm going to  head over to Google, say cold email lead scraping,  
178:29 and I'm just going to find some links here.  Basically, I just want like a brief description  
178:32 of what cold email lead scraping is. So, let me  see if maybe instantly has some information on  
178:35 their site. Um, this is like a full endto-end cold  email sending platform. Uh, probably not what I  
178:44 want. I just want like some lead scraping.  This looks pretty good. I like this. So,  
178:49 I'm just going to copy this. I'm going to head  back here. I'm going to paste this in. I'll say  
178:53 this is an example of the sort of thing that we're  about to build. I'm going to paste this in just so  
179:00 it has context and so it can write my landing page  for me. And then next up, obviously, we need that  
179:05 um prompt that we just copied over. I'd like  to build an application. I want it designed  
179:09 according to the screenshot that I sent over. The  specifications and the stack is um in the middle.  
179:15 And then I've also given you a description of  another website that does something similar that  
179:19 I'd like you to emulate for the copywriting  and content and so on and so forth. Okay,  
179:25 we're going to stick that in. And uh this is most  likely going to end up using like a next.js stack,  
179:30 although I did not this time necessarily want  to be opinionated cuz I wanted to see what the  
179:34 app was going to do. Uh when it ends up creating  files, it'll do so over here. Keep in mind that  
179:38 we are in planning mode as per usual. So that's  what's going on down here. And it looks like we're  
179:42 using Oh, actually we're using Vit, which is good  because that'll give us a the ability to maybe  
179:47 use a different framework here. To be honest, I  don't really care about the specific framework,  
179:50 right? We talked about how there were multiple  uh in the previous section. So, whatever works  
179:54 is going to work. Um, you know, as long as it's  fast and performant. Okay. So, at this point, I'm  
179:59 just going to sit back and let it do its thing.  Awesome. So, I just got a notification saying that  
180:02 it's done for Leadzo, which is the name that uh I  decided I would give our cute little app. And now,  
180:08 what I'm going to do is I'm going to say run  it. The reason why I'm going to do that is  
180:11 because in order to access this application,  we're obviously going to need like some server  
180:15 running at the back end. Our browser is going to  end up being the client and then that's how we're  
180:19 going to communicate with that. So, our server  is going to be on our computer for now since it's  
180:22 all local. Let's see how it goes. So, we got the  Vit server set up. This is the landing page. Uh,  
180:27 super clean. I like it. Okay, so we have like  some weird spacing here. That's fine. This is  
180:33 really nice. I like this AI personalization  prompt. Very clean. Very clean. Nice. 100  
180:38 free credits. Okay. So, what happens if I click  get started? I go over here, which is nice. Um,  
180:44 I'm just going to ask it. Oh, it looks like it's  actually recording a browser session of me testing  
180:49 this. This is wild. It's Nick Nikki Wiki. Then,  Nikki Nikki Nikki. Let's sign up. Okay, we're now  
180:57 inside. We have some workspace settings on the  lefth hand side here. API integrations. On the  
181:02 right, we have some prompt engineering. So, define  the rules leader should follow and generating the  
181:05 personalization column for each lead. Cool. Does  this actually fill in? Okay, we just have some  
181:10 placeholders for now. What happens if um Cool. So,  it went it actually went it found the uh specific  
181:15 API keys and stuff. That's fine. Scrape leads. So,  we need a job name, LinkedIn Sales Navigator URL,  
181:21 max leads, then default icebreaker. I think what  we want is basically are going to want to be able  
181:27 to just say uh I don't know like HVAC companies in  the United States and then we want it to actually  
181:35 go and find that for us. We're not going to need a  LinkedIn sales navigator URL because the appy the  
181:39 specific service that we're using doesn't actually  um use that. So, it's going to be okay. Then here  
181:43 we have some fake order history which looks  really clean. We can, you know, open it up in  
181:46 a new window or whatever. Awesome. Awesome. Cool.  So, let me just check the bottom. You guys can't  
181:51 see this, but there's a little button that says  log out. So, click this log out button. Nothing's  
181:54 going to happen since obviously we don't have  the off set up. Yeah. I mean, like in a nutshell,  
181:57 this is more or less what I what I wanted. Um,  anything else that we need? So, new scrape job  
182:01 orders configuration. I think that's basically it.  Now, now what we have to do is we basically have  
182:06 to test this. Uh, cool. So, what I'm going to do  next is I'm just going to run through a couple of  
182:11 these services that we're using just so we're  on the same page. Otherwise, I think it'll be  
182:14 kind of intimidating for you guys to see what I'm  doing. But in reality, most apps nowadays are just  
182:19 like wrappers, as in they just wrap functionality  around some other app. Of course, you could build  
182:23 your functionality yourself if you really wanted  to, and I can help you with that. But in my case,  
182:27 you know, for the purpose of this demonstration,  I'm just going to do what I normally do,  
182:30 which is I use a three a thirdparty service called  this leadsfinder on a uh which basically allows  
182:34 you to put in some parameters and then get some  leads. So, it generates targeted B2B contact list  
182:38 using advanced filters. Now, this isn't the best  lead scraping service out there. I should note  
182:42 that like it's actually quite poor in terms of  quality now because uh the leads are cached and  
182:46 they haven't been around for quite a while. So,  I use this as an example in Make Your School,  
182:51 my community all the time. But, let's say I'm  scraping some dental clinics in Canada. I want  
182:54 to scrape a thousand. What I do is I basically  just put in a bunch of job titles here. Owner,  
182:57 practice owner, principal dentist, lead dentist,  so on and so forth, and then locations to include,  
183:03 which in my case is Canada, locations to exclude,  what sorts of emails I want to get. So, validated,  
183:08 not validated, unknown. So, I I actually just want  validated emails, what sort of company website if  
183:13 I have them. Um, we can use to narrow down the  search. The size of the uh resulting company,  
183:19 so 1 to 10, 11 to 20, 21 to 50. the industries  that I'm looking in. So, health, wellness,  
183:24 and fitness, hospital, and healthcare. And then  which ones not to look for some keywords, dental,  
183:28 dentist, and then I'm also excluding some down  here. Then we have revenue, and we also have some  
183:32 funding. So, let me just make this smaller because  I just want to show you guys how this actually  
183:36 works in practice. I just click save and start.  And basically, this thing goes on this service,  
183:40 Appify, and then goes and it scrapes it  for like some small little piddly fee. So,  
183:44 it's cooking up some data magic. Uh, let's just  wait until we see the very first result, which  
183:49 should be about 100 leads or so. And as we see  the first rung here has found a bunch of different  
183:53 leads from specific companies, company websites,  LinkedIns, their full names, their job titles, and  
183:58 so on and so forth. This is pretty cool, right? I  mean, this is just like a service that I'm using,  
184:01 and it returns a bunch of leads. And so, for  the purpose of this demo, I'm going to assume  
184:04 we're only using like simple services like this.  We're not actually rebuilding the wheel. Although,  
184:08 I should note that you totally could rebuild the  wheel if you wanted to. It's just uh, you know,  
184:12 it's not super easy, right? Like, this is really  the moat nowadays. It's not the app interface  
184:16 that you build. Sorry, that's the wrong one. Uh,  where are we? right over here. Okay, so basically  
184:22 in order to run this, right, logically what we're  going to need is we're going to need an API token  
184:26 and then what'll happen behind the scenes is this  Appify API token will plug into this search. Okay,  
184:32 then it'll use some data with an input like  this via API and then it'll find us a bunch of  
184:39 of stuff. And then on the back end, we're going to  do is we're going to enrich emails using a service  
184:44 like this. Basically, not all of the emails  that come from this service are going to have  
184:48 um like all of the information laid out. So, we  can actually use another service called an email  
184:52 finder to enrich that. And then over here, um  we're going to pass in either our open AI key or  
184:57 anthropic API key, whatever one works, honestly,  and then use that to generate personalized  
185:02 snippets of text that basically are going to allow  us to personalize the emails that we send. So,  
185:08 hopefully we're all now on the same page.  This is a pretty niche service, but you know,  
185:11 the best SASes do tend to be pretty niche and um  you know, that's more or less what we're putting  
185:15 together. We also have an instantly API key for  one-click campaigns, which are basically where  
185:20 we'll just upload the leads automatically. I  don't actually know if I'm going to do this  
185:23 cuz the instantly API had a bunch of issues last  time I tried, so I might not. And then over here,  
185:27 we have the actual prompt engineering. You know,  I want to generate like a highly personalized var  
185:32 uh uh email, right? I want to say like, "Hey Pete,  been following you for a long time. I love what  
185:35 you're doing at XYZ company." because those emails  that I send when they're highly personalized  
185:39 like that tend to land a lot better. So here's  where you can actually define the prompt that  
185:43 we're going to use inside of the app in order to  generate these ice breakers. So you're an expert  
185:47 sales SDR. Your job is to create a highly relevant  one or two sentence icebreaker so on and so forth.  
185:50 This is this is everything here and the idea is  you'll get to change this over time and then you  
185:54 know save and then see your new orders and stuff  like that. Okay, so hopefully now we're all on  
185:59 the same page and I've created this demo. This  is basically like the vision of the app. What  
186:02 I'm going to do now is I'm going to go through  and just make a bunch of changes. Um, I'm also  
186:07 going to feed in like the API documentations or  the services that I want to scrape and use. And  
186:10 then my agent should do a pretty good job to  start. This is great. The Appify scraper that  
186:17 I'm using to generate leads doesn't require a  LinkedIn sales navigator URL. I'll give you the  
186:22 Appify API below as well as everything that  you need in order to make this request, but  
186:27 you'll have to modify the new Scrape job form to  reflect that. What we'll do is we'll pass in a job  
186:33 name and then we'll have AI convert that job name  into a list of filters that work with the specific  
186:40 um Appify scraper that we're using. We also need  a way to be able to click into the orders under  
186:46 order history and then open them in sort of a  sidebar. Also, right now nothing happens when  
186:50 I click those three little dots or the download  button. Obviously, we're going to need to download  
186:54 these CSVs. Under the configuration section, let's  start with the Appify API token up at the top.  
187:01 Then let's do the Anthropic and OpenAI API tokens  underneath. And then let's remove the instantly  
187:06 API token completely and have the any mail finder  down below. The idea is going to be our Appify  
187:11 actor is going to return most of the information  that we need rather than treat OpenAI's API key  
187:19 as a fallback. Let's just treat it as another  option they could go through and then let's  
187:22 just prioritize anthropic. For instance, if both  the anthropic and open AI API keys are filled out,  
187:28 the model should automatically choose anthropic.  If anthropic is empty and open AAI is uh present,  
187:33 then we should automatically choose open AAI.  And we'll just proceed in that manner. And then  
187:37 for prompt engineering, I'm actually going to  give you a big long prompt that you could use to  
187:41 significantly improve the quality of the output. I  want you to incorporate this. Once we're done with  
187:46 the front end, we obviously need to work out the  back end and have all these functions available.  
187:51 So, what I'm going to do is I'm going to create  av and then I'm just going to paste in all my  
187:54 API keys for you so that you have them. You  can use this as demos to basically test the  
187:58 backend functionality and we could use these  and store them um in a static database for now  
188:03 just in like JSON files or or whatever. When we  eventually push to our database and use superbase  
188:09 um then we can make these more dynamic. I think  you have everything that you need. If you have any  
188:14 uh questions or misunderstandings or  whatever, just give me a shout. Cool.  
188:17 So now I'm just going to go through and then give  it everything that I promised. So the appy API is  
188:23 going to be here. It's going to go right into  the app API docs. Now what we want is we want  
188:30 to run the actor synchronously with the input and  return the output. I believe return the data set  
188:40 items rather. So I'm just going to see if I could  copy this whole page. Can I copy this whole page?  
188:45 Sometimes Appify has the ability to copy the whole  page for AI. Yeah, it does. So, copy for LLM.  
188:50 Paste that in. See how long that is? That's super  long. Yeah, it's pretty long. And then I also want  
188:57 to paste in the specific stuff about the actor.  So, this is the specifics about the actor. Input  
189:05 schema for actor is here. And then, um, I also  promised that I'd feed it in a big long prompt.  
189:14 So, I'm going to go over to Maker School wrapped,  which is massive document that I put together that  
189:19 basically guides everybody through how to write  like super high quality cold emails. And then I  
189:24 have a little prompt here that actually you guys  could use as personalization. So, what I'm going  
189:29 to do is just going to type personalization and  start down at the bottom. There we go. And this  
189:37 is going to be the prompt which I'm going to feed  in right here. Here's a personalization prompt.
189:46 For the personalizations, use Claude Sonnet 4.6  or uh uh GPT 5.3. Check out both of their docs.  
190:01 Okay. And then I think that has everything  that I told it that I would give it. I mean,  
190:04 this is pretty long, right?  Just make sure. Okay. So,  
190:06 I'm going to create av and then um paste in  all my API keys for use. That makes sense.  
190:12 So I'm going to go over here and then just go env.  And then I'm now going to go find a bunch of API  
190:17 keys on all the services that I promised it and  then just paste them all in. So this is going to  
190:21 be my enthropic API key right over here. Then go  apply API key. So I head over to appy, go to my  
190:29 console. There's always a way for us to grab our  API key from whatever service we're using. So I'll  
190:37 go up here to API and integrations. We'll create  a new personal API token and I'll call this for  
190:42 SAS app. Should I limit token permissions?  Should I set the expiration date? H yeah,  
190:49 I'll set an expiration date just in case I forget  to delete it. Okay, this one's expiring tomorrow.  
190:54 That's all good. I'm going to copy this now. Paste  that in. Uh I need my open AI AP. Actually, do I  
190:59 need my open APK? No, I don't because I'm going to  use the anthropic one. Ah, I suppose uh the other  
191:04 services that I need, I can find them by looking  at my Vit app right here. Any MailFinder API key,
191:15 paste that in. So, we'll go to any mailfinder,  log in. I'm going to go up to the API and I'm  
191:20 just going to copy this token right now. Go back  to anti-gravity. Feed that in. And we should now  
191:24 have everything. So, I'm going to press enter. And  I think this is now going to be able to develop  
191:28 the rest of our app, at least for now. Uh, this  is going to be like kind of what we see above the  
191:34 fold. Let me just be smart about this. Don't  just paste the prompt in verbatim to the user  
191:38 editable section. Obviously, it's pretty long. All  right, we've made a couple of changes here. Now,  
191:43 we have find me VP sales and practice owners at  dental clinics in Canada. I want companies with 11  
191:47 to 50 employees. That's kind of cool. If we go to  orders here, can we actually click on these? Yes,  
191:51 we can. As you can see, we now get a little  sidebar. Um, the sidebar is kind of small,  
191:55 so I'm going to adjust that in a second, but  it is still pretty neat. I mean, I can click  
191:58 the download verified CSV button here. So, it's  pretty nice. I like this. I really like the fact  
192:02 that we have this little job. It also pumps up so  quickly. It just makes it seem really snappy. And  
192:08 here is the prompt engineering. So, the prompt is  just pasted entirely in now, which is nice. Yeah,  
192:14 it's basically perfect. We do have um what looks  to be no training examples, but that's okay. These  
192:20 prompt variables are going to be injected as  the first name to lead, last name of the lead,  
192:22 headline, industry, organization name, location,  even their email address. That's going to be cool.  
192:26 And then we also have our uh our tokens in. So  now it's just time to actually give this a try.  
192:29 So I've just added the API keys right over here as  mentioned. And now we're basically good to go. I'm  
192:35 going to wait for this thing to relaunch the dev  server and then we're actually going to test out  
192:39 the flow. I should note that, you know, because  we are testing out what is essentially going to  
192:44 be the core function of our app, just like with  a full stack app, like the probability it'll work  
192:48 on the first try is pretty low. I found it kind  of interesting because a lot of people like they  
192:52 judge AI by its ability or inability to do things  in one go. uh human beings rarely do things in one  
192:57 go as well. The difference between humans and  AI models is silicon thinks at something like  
193:01 10,000 times the speed. So you could just try  it 10,000 times faster. So you know whether or  
193:06 not this works in one shot is okay. We're just  going to go back and forth until it it works  
193:10 reliably. Um and then from there we're going to  pump this up to netify just like I had before.  
193:14 And then app number two, full stack app number  two, but total uh web app number three is going  
193:21 to be completely cool. I've just done this. Now  it's storing what's called local storage since  
193:24 of course we're always doing our our first jobs  uh or our first check statically. So let me exit  
193:29 out of my cloud API keys. Thank you very much.  Go right over here. Just relogged in and then  
193:34 I'm just going to go scrape leads. So I'm going  to say and I imagine there are going to be some  
193:37 issues here. Find me VP sales at dental clinics in  Canada. I want companies with 10 employees. Okay.  
193:47 And why don't we do I don't know 200 leads. And  then we'll go laconic or lassic. I don't actually  
193:54 know how to say that. Um and then we'll go error  executing leads workflow. Check console. I'll head  
194:00 back to console and then I see that we now already  have some issues not found. So what I'm going to  
194:06 do is I'm just going to scroll through this then  feed back to our agent. All right. After some  
194:13 back and forth, it's now telling us it's good  to go. I'm not entirely sure if this is true,  
194:17 but we are going to give it a try together. Uh,  I think it'll probably be over here. And then I  
194:22 go slash scrape. Find me some practice owners at  dental clinics in Canada. One to 20 employees.  
194:30 Cool. I'll press 100 max leads. We'll run the  workflow. We see here it's now doing what it's  
194:35 saying is a search extraction. And it looks like  it hasn't finished it successfully successfully  
194:41 because we're immediately getting a return  saying it's it's completed successfully,  
194:45 but it isn't. Let me just paste this back. I  ran this on localhost 888/scrape returned this  
194:52 and then I'm just going to paste it in. I'm also  going to uh paste in my um input as well just so  
194:59 that I could, you know, monitor this and also  maybe have a little bit more meaningful of a  
195:03 back and forth. Um but anyway, let's see what it  has to say. just as a form of parallelization.  
195:09 While this is debugging whatever the appy issue  is that's leading to um inconsistent output,  
195:14 I'm actually just going to have it adjust  the copy on the landing page as well. So,  
195:18 let me see here. Hey, this is fantastic. I'd like  you to change the copy on the landing page that  
195:24 it's a little bit more geared towards outbound  cold emails specifically. Um, also there are some  
195:31 minor graphical issues like the start for free  buttons arrow is on the same is on a different  
195:36 level than the text. I'm also seeing that there  are four cards under built for performance,  
195:42 but the fourth spills out onto the second row.  Let's do six cards instead. The main value  
195:48 props here are the breadth of contacts, the high  quality database, the high quality of enrichment,  
195:55 um the simplicity of the tool, the affordability  of the leads relative to most other people. For  
196:04 the AI personalization section, um under AI  personalization prompt, remove the export to  
196:11 oneclick instantly campaigns and then change  the language so it's not start for free.  
196:17 You just start. I'm also noticing that there's  no pricing section. Update it so that there is a  
196:21 pricing section. And make sure that everything in  the navbar is like perfectly centered. Right now,  
196:26 it's a little bit off, or at least it seems to be  off. Add a pricing section that is on average 0.05
196:35 per credit. Every credit returns one verified  email. And then offer some higher and lower plans  
196:43 um that scale with price. Obviously, when  you purchase more, you should get more of  
196:46 a discount. Cool. So, I'm going to pump that in.  And then I'll say only modify the landing page.  
196:52 There are other agents currently working on other  sections of the website. Do not modify anything  
196:56 that is not the landing page. Also, don't you  don't need to rebuild. Okay. So, I'm going to  
197:00 have that operating at the same time as our other  tool, which is fixing the appy scraper. And then  
197:06 I can monitor this just by going back to command E  and checking out the agent um kind of conversation  
197:12 here. If I go to under inbox, you can see my  two open conversations. There's this one here,  
197:16 and it looks like we're still building or fixing  whatever the specific Appify problem was. Looks  
197:21 like it has to do with like hash IDs or whatnot.  Okay, so this is a good summary of where it's at  
197:25 right now. And then the landing page copy  builder is now doing some things. Cool. So,  
197:31 it's analyzing the landing page. And now it's  going to go through and actually write it.  
197:36 Looks like this search worked now on the uh  curl command that it created and it just did  
197:42 this to validate without me actually having to go  through and then do it all myself on the website.  
197:46 It's also reporting the specific issues. So there  was some location, some swallowed exceptions and  
197:51 so on and so forth. That's cool. And it also ran  the live test and it's asking me to do it again on  
197:56 um you know 888. So that's a different server.  Then I'll go what are we doing? Scrape. return  
198:03 me owners of HVAC businesses in Texas. Okay, we're  still getting the zero completed. So, let's go and  
198:11 debug Y. This is my input. Just go into the new  landing page update. We've now changed this to six  
198:17 rows, which is cool. This looks kind of silly.  So, I'm just going to go back here and say is  
198:23 currently on three lines. The on though the N  looks odd. Everything else here does look pretty  
198:32 solid though. I like that we now have a growth and  pricing plan with the same style. 49.99 and then  
198:40 249. Super cool. And I'll just say let's dress  up that final CTA card down at the bottom. Cool.  
198:51 We'll go back to our appy scraper, see how this  is going. Just shut over a request here. And it  
198:55 looks like we are actually doing something which  is nice. Um, normally this would have immediately  
199:01 errored out, but it's either hanging or it is  doing what I asked it to do. I think this is hung,  
199:06 which is why nothing is occurring. If I  refresh this, I think the server probably  
199:10 went down halfway through. The good news is we  are populating the database, although either  
199:15 some design issues have occurred or something  is broken. But as you can see here, we have the  
199:19 status, the date, and then the verified emails as  well. I'm seeing no problems in the configuration  
199:24 side as well. So, it really is just the scrape  leads hangup. What I think I'm probably going to  
199:28 do given that this has taken me like 10 minutes or  so just behind the scenes is I'm probably going to  
199:32 go and just give it a bulletoint list of what it  needs to do. Like basically if you think about it  
199:37 um it needs to interpret my request. it needs to  convert that into an input friendly JSON object  
199:49 using um you know like a location table UTC.  Then it needs to send the request to um some  
199:58 big long URL which is going to look something  like this just with a different appy token.  
200:03 And then it needs to return to user alongside  CSV etc. I also think I might just simplify  
200:13 the number of steps so that all we're doing is  the appy search extraction then any mail finder  
200:17 enrichment for now. Um so I'm going to do that  next. Just heading back over here. Just going to  
200:22 paste all this in. I'll say let's adjust the logic  to the following. Okay. So I ended up running out  
200:28 of tokens halfway through this build. Let me just  show you guys what that looks like and then how  
200:31 you can solve it. So, basically, if you use Gemini  3.1 Pro, that obviously being the most in demand  
200:36 model right now, given the fact that it's, you  know, new, really good at UX and stuff like that,  
200:40 eventually you will probably run into some rate  limits. This case, this rate limit occurred after  
200:44 about an hour of this working independently on its  own to solve some problem with some rate limit.  
200:49 I just went and I walked away and then I came  back and then I saw that um sorry, with the uh  
200:53 with the app scraper. And so essentially how you  proceed from here, at least my recommendation is I  
200:58 would use a different model using the agent chat  UX. So I went down to claude opus 4.6. And then  
201:05 I said first generate a comprehensive prompt that  includes everything that we've talked about so far  
201:09 so I can paste it into another version of CC. CC  being cla. I'm now going to copy this. Okay. Then  
201:16 what I'm going to do is I'm going to open this  up in claude code. And not only am I going to  
201:21 open this up in cloud code, I'm also going to do  something called fast mode because claude has an  
201:25 additional piece of functionality that uh Gemini  does not currently possess. That is the ability  
201:29 to go very very quickly. I'm then going to double  check my extra usage. Looks like I'm 600 bucks in.  
201:36 So I'm just going to go extra- usage. Keep in  mind you guys aren't going to be spending $600  
201:40 and stuff on cloud code usage. Why don't I just  change this to 700? Um I just use a I use this for  
201:45 like so many different things, right? juggling  five or 10 projects simultaneously. So for me,  
201:50 700 bucks in additional usage might be I don't  know an additional 20 or 30 bucks per project.  
201:56 Then I'm going to set that monthly spend limit.  Going to give that a little refresh here. Now  
202:01 it's up to 86% use. So I have 14%. Okay. And then  I'm just going to exit out of this and then go  
202:06 to a new one. I'm just going to paste all this in  now. I'll say proceed. And now there will be this  
202:14 little lightning icon down here which denotes that  it's in fast mode. And then I get to make use of  
202:18 um you know claude code or claude opus 4.6  thinking I should say on fast mode while I wait  
202:23 for my Gemini usage limits to come back. Now you  don't have to do this. If you don't want to you  
202:28 know juggle multiple models like I'm doing right  now. If you're happy to trade a little bit of time  
202:32 for speed then by all means absolutely go ahead  and just use like the claw opus 4.6 built into um  
202:38 the Gemini or agent chat inside of anti-gravity.  In my case, I want this to work as quickly as  
202:43 possible since on kind of a time crunch here. I'm  going skating with my girlfriend. So, we're going  
202:48 to see how much of this we could do in 15 minutes  just using the fast mode. And I'm happy to make  
202:52 that time money trade um as necessary. What you'll  see about Claude is it's just significantly I want  
202:59 to say more verbose with its notifications and  the information and it's also a lot easier to  
203:04 follow along. That's one of the downsides of the  Gemini series of models, just like a little less  
203:08 interpretable. Um, the thinking tabs and stuff  like that here are basically just in natural  
203:13 language and I can follow along with them almost  as if it's me doing the thinking. Uh, which is  
203:17 one thing I appreciate. Obviously, you do have to  weigh each of these minor pros and cons against  
203:22 uh, you know, the model's ability to do things  like do front-end UX is really pretty and and so  
203:27 on and so forth. Cool. So, this looks pretty solid  to me. We now have the server. It's going to test  
203:32 the API endpoint directly and it looks like we  have some internal server error. It's now just  
203:36 going to go through and then update model ids and  make minor terminations um as to what the error is  
203:41 until we're we're good to go. And it's telling  me that it is currently running an appy actor  
203:46 run. So I'm going to take a peek at this right  here. Looks like we ended up finding five leads,  
203:51 which I think was initially what we asked for.  Going to head over scrape here and I'll say HVAC  
203:55 owners in Texas. Let's do 100 leads. We'll run  the workflow. Looks like we've now generated the  
204:00 filters right down below. So, it's now scraping  leads with this poll. And um this is just for I  
204:05 guess our own our own purposes essentially. You  know, if we ever want to modify what the target  
204:09 audience is or our job name, we can do so just  by maybe being a little bit more descriptive or  
204:14 maybe being a little less descriptive. Okay,  we've now completed the job. We have the CSV  
204:18 over here and I can actually open it and you  can see all these leads are now scraped. So,  
204:22 we were basically like 99% of the way there right  before I ran out of tokens. Just needed uh another  
204:26 model to get it over the line. This looks great.  Now let's re add the personalization workflow.  
204:34 Use claude sonnet 4.6 as well. Okay. Now I'm going  to do the exact same thing. HVAC owners in Texas.  
204:42 We're going to test the lead scraping. We can also  test the AI personalization as well. I'm doing 100  
204:46 leads. So this should occur reasonably quickly.  Should be another five or 6 seconds. Okay,  
204:51 looks like we're done with that. Now we're  doing AI personalization. So Claude in this  
204:54 case is generating personalized ice breakers  for all of the leads that are in our database.  
204:58 Something I'm noticing here is the AI  personalization is taking more than five  
205:02 or 10 seconds. I imagine this is probably because  we're passing 100 rows sequentially or serially.  
205:06 If you think about it realistically, you don't  need to do that sequentially or serially. You  
205:09 could do that parallel in a parallel and send 100  requests simultaneously, whatever the the rate  
205:14 limits are. So, I'm going to ask it to do that.  I'll say batch the requests. Make do whatever  
205:19 you have to do to make it faster. Okay. I'm then  going to rerun this. We'll go I don't know HVAC  
205:25 owners in Canada instead of Texas. Let's just  make sure that lead generation uh filter thing  
205:31 still works. So, this is now Canada. We still  have a bunch of size filters down here as well,  
205:34 which is cool. Because I'm only doing 20, this  should occur significantly faster. We're now  
205:39 moving on to the AI personalization step. And  then at the end, we're now at completion and  
205:43 export. That looks great. We're going to download  the CSV. Just going to open this up again. Make  
205:47 sure we have that personalization column. Where  is that personalization column? Here we go.  
205:53 Hey, Jean Michelle. Revolutionizing luxury best in  77 is no small thing. Bane Ultra's legacy speaks  
205:59 for itself. Wanted to run something by you. So,  I'm not really liking how this is looking. It's  
206:02 just a little too much right now. So, I'm going  to want to make this significantly more casual  
206:05 and cleaner. I'll head back to configuration.  Okay, these are looking a little better. Still  
206:10 probably not the best personalizations we could  be doing, but I think just for the purposes of  
206:13 demonstration, I'll leave it there. Hey Kate, love  what you're building at Posh, also in engineering  
206:17 services. Wanted to run something by you. I  can see that work reasonably well as an email  
206:20 icebreaker. So, let's just exit out of this. We'll  close this older one. And uh from here on out,  
206:25 all we really need to do, if you think about it,  is we need to fix this UX problem. Um we need to  
206:31 add authentication with Superbase. And then we  need to take this puppy live. So I'm just going to  
206:35 do a big voice dump here and get it to the point  where the quality is significantly improved. I'm  
206:39 going to do that as taking a screenshot of this  section. And then coming back to Claude, we'll  
206:45 paste this in. Then I'll say, "Hey, the current  app looks great. There's just a few issues.  
206:52 The order history page looks really squashed and  not at all like the rest of the pages. I want it  
206:57 in that nice clean style um similar to the scrape  page and the configuration page. We also need to  
207:06 put this live and uh get this on superbase. So  I'm going to set up a superbase instance and I'd  
207:12 like you to assist me in the connection of that.  We need all of the fields of relevance including  
207:20 um database fields for the configuration, the  prompt engineering, API integrations, basically  
207:25 whatever you've done locally just applied to um  you know a database instead. Last but not least,  
207:32 when you click on an order in order history,  it opens up this nice side modal, but it's  
207:37 not very tall at all. I want it stretched out to  contain such that I don't have to scroll through  
207:44 um the div containing all of the information. If  that's unclear, just let me know and I'll clarify.  
207:51 I'm going to dump this in and then I'm also going  to just because I'm using a new instance of cloud  
207:56 code here. Just going to go back and then I'm  going to copy all of this. And I'm doing that  
208:03 because I want to give it context, but I also  don't want it to like go ahead and redo stuff on  
208:06 the app ify scrape and so on and so forth. While  that's occurring, I'll head over to Superbase.  
208:11 And then we'll click start your project. I already  do have a couple of apps set up. I think what I'm  
208:15 probably going to do is just delete one of these.  H probably delete the proposal generator app and  
208:22 then set up a new one just because I think you're  only allowed two on a free plan. And um I want to  
208:26 show you guys how easy it is to set all the stuff  up. So just going to delete that. And then I'll  
208:30 go new. I'm going to call this one um let's go  Nick J Wellsorg. We're going to call this leadso  
208:38 DB. I'm just going to do a strong password  automatically here. Copy that. And then I'll  
208:43 enable the data API. Click create new project.  I'm just going to give it all the information  
208:49 that uh it needs in order to set up this database  just like I did last time. Okay. So, first thing  
208:54 is head back to anti-gravity. I'm just going  to paste this in. That's my DB password. Then  
209:01 I'll head over to this left hand side here where  it says project settings. I'll go project name
209:11 and then we'll also go project ID. I think there's one more. We just  need the API key so we could do a  
209:20 lot of that work for us. So we'll go secret key. And then we'll also go superbase public key.  We'll copy that as well. And this should have  
209:35 everything that we need. I think what it's  going to do now is probably just set up O for  
209:39 us. So head over to authentication where it says  email. And this time I don't actually want it to  
209:44 confirm the signup. So I'm just going to go back  to emails and then where I click on sign on signin  
209:50 providers. I'll just turn off confirm email. That  way I can make an account without having to do  
209:54 that back and forth. Great. set up superbase and  off. I'll give a bunch of information here. So,  
210:02 I think this is probably everything that it  needs and then we're just going to head over to  
210:06 um SQL editor where it's going to give us  a bunch of SQL for us to paste in. Now,  
210:10 my Gemini usage limit resets just in a few  minutes here. Um just because I have all of  
210:14 the context inside of Cloud Code and it's going  through and doing this work, I'm probably just  
210:18 going to continue in this Cloud Code thread  at least until I'm done this migration.  
210:22 um then I can start up another agent chat  uh using you know anti-gravity's built-in  
210:26 functionality and then have a conversation with  it. Uh in general you don't really want to jump  
210:30 around between one model and the next unless you  can either fully export the conversation history  
210:34 or you could do some sort of midway summary like  I just did a moment ago um from the anti-gravity  
210:38 to claude or sorry from the agent chat to Claude  code direction. If I were to go the other way,  
210:44 I'd do the exact opposite. I asked Claude to give  me a detailed project description of everything  
210:48 that is currently occurring and all of its current  context. So I could immediately just segue that  
210:52 into um the the conversation with Gemini 3.1 Pro  high. Okay, it's given us a plan and I've already  
210:58 enabled this plan and let it go. So we have the  superbase URL, the anon key, service ro key and  
211:03 and so on and so forth as well as a bunch of  other information here. It's going to start  
211:08 with the environment variables, give me an SQL  migration file, do some client updating, add a  
211:13 bunch of protected routes using authentication  middleware, finish the login and signup pages,  
211:17 have a log out function as well, set up the config  page, do the orders page, do the scrape page, um,  
211:23 set up the API layer, and then connect that to  Netlifi functions before finally building and  
211:28 verifying. There's going to be some back and forth  there for sure. Um, but just like we did this with  
211:32 Gemini 3.1 Pro High, we'll do the same thing  here. So, I have the migration right over here.  
211:36 I'm just going to copy all of this in. And then  we're just going to remove these little comments.
211:47 And then what I'm going to do is just click  run. This will go through and then set up  
211:51 our database. You can now see this in database  because we have a scrape jobs table with all of  
211:56 these fields. The user configs table with all  of these fields as well. Any mailfinder key,  
212:00 system prompt, instantly key, and so on and so  forth. And these are all going to be um sort of  
212:04 fed into this authenticated user. So this is what  the authenticated user will sort of have access to  
212:09 when they go from one route to the next. At the  same time, it's going through and then updating  
212:13 everything that it's set up to do locally using um  JSON and JSL files to work through, you know, API  
212:19 requests to the Superbase PostgresQL server to the  Superbase Postgrql database. And this is actually  
212:25 a really good use case of using two different  models. The first is to build and the second is  
212:30 to basically spot check what the first model did  without full context. The reason why is because  
212:34 when you don't have the full context, your uh  opinions aren't really like polluted by the fact  
212:38 that you're the one that spent all this time and  energy working on it. Just like you always have  
212:42 like empirical or peer reviewers in papers and  and studies and research and stuff like that. So  
212:47 too do I recommend doing this with models as well.  Having an unbiased third party review our code is  
212:52 what is ultimately going to make our code and our  app significantly higher quality. Funny how models  
212:57 are kind of like people in that way. We could  theoretically go back and forth and back and forth  
213:01 with Claude Opus 4.6 designing something, Gemini  3.1 Pro High designing something else. Um, but I  
213:07 think just one of these passes is good enough. And  in a way, that's what we're already doing with our  
213:11 security audit if you think about it. Okay, it's  asking me to run the migration. I already did. So,  
213:14 I'm saying done. Let's test it. It's now going to  launch this live. And imagine there'll probably be  
213:19 some back and forth just like there was before.  Why don't we click get started? I'll say nick  
213:23 sarif nick atniki.com and then we'll go I don't  know nikki nikki. Click sign up. Okay, we are  
213:32 now being automatically taken to that config page  just like I wanted to. So here is where we'd set  
213:36 up our keys and then we have that prompt which is  automatically set up. That's kind of cool. Okay,  
213:41 I think what we need to do is we actually need to  modify this prompt to make sure it it's actually  
213:45 like being reflected in the end result. Um and  then obviously we need a way to stick our tokens  
213:49 in there. So, what I'm going to do for now is I'm  just going to go grab the ENV file here. I already  
213:55 have my token. So, I'm going to copy this. I'll go  this API. Okay. Paste that in. Annie Mailfinder,  
214:01 we'll paste that in. Enthropic. We will paste  that in. Okay. Now, I want to save changes.  
214:08 Just reflect that this actually works. So, it  says saved. If I refresh this, I'm seeing that  
214:13 the state is the same, which is good. If I make  a spelling mistake here and save, I'm seeing that  
214:18 the changes are reflecting here as well, which  is good. I go to orders, there's nothing. Okay.  
214:23 If I go to scrape leads now and I say, find me  a bunch of HVAC companies in Canada, owners of  
214:29 HVAC companies in Canada, and I go 20 leads, and  I run this workflow, is this going to work? So,  
214:35 it's gone through and it's generated the filters.  Let's just see what it's saying here. It's now  
214:39 scraping the leads. Now, we're doing some AI  personalization. And I'm just playing around the  
214:43 UX here. This is nice. This is nice. and smooth.  I like this a lot. I also like these cute little  
214:48 animations and whatnot that it's decided to put  in. Okay, job completed successfully. I also like  
214:52 how on the lefth hand side it's renewed. I don't  like how this little um scroll wheel animation on  
214:57 completion and export continues. So, I'm just  going to head back, turn off the scroll wheel  
215:01 animation on completed when the job is scraped.  When the job scrape is complete, completion.
215:13 Okay, that'll be good. But for now, let's actually  download the CSV. We got our leads up here. Just  
215:19 going to take a peek at them. Same thing I did  before. Okay, icebreaker is looking great. Love  
215:22 what Proparts doing in modular construction. Also  work in the space. Love the tutor teach model.  
215:25 Connecting educator students is no small fee.  That's reasonable. So this clearly inherited our  
215:29 good um result. Then finally, if I go to orders,  we can see this is now populating our database  
215:35 successfully. What happens if I click? Awesome.  We have all of the data. Looks really clean. We  
215:38 have the parsed filters. We have the profile  scrape. You can actually download the verified  
215:41 CSV here as well. That's clean. Very nice. Okay,  the last thing I'd say is this three. These three  
215:47 dots don't do anything. Also, the three dots to  the right hand side of an order in order history  
215:56 do nothing right now. Okay, that's clean. And  then if I go down here to log out. Okay. Well,  
216:05 let's just quadruple check everything else  is good. Looks like it. We're back here.
216:13 I think it was nikki.com, right? Yeah.  Okay, cool. Awesome. And I'm just going  
216:17 to head back out and then test the rest  of my routes. So, if you guys remember,  
216:20 there was scrape. Okay, that doesn't work. There  was config and there was also orders. Okay,  
216:25 so all of the off middleware  here is working correctly. Um,  
216:28 how's our landing page? It's looking pretty  clean. We don't have payment functionality yet,  
216:32 which I'll cover in a second. This looks pretty  clean. If I click start, we go back. Nice.
216:39 Also, dress up the footer a little bit more. It's  
216:41 quite small right now. That's on  the landing page specifically.
216:47 Okay, this is cool. I like that little like light  effect that we got in the background. Real clean.  
216:51 Real clean. Um, product goes to product.  Pricing goes to pricing. Resources. Oh,  
216:56 there is no resources link. So, we don't really  need that. Uh, I suppose we could probably make  
216:59 a cute little logo. So, why don't we head over  to asstudio.google.com and just do that while we  
217:05 can. I'm just going to see if I could link an API  key. Okay, I need a different account for that.  
217:10 And over here, let's see if we can link a key.  And heading over here, let's see if we can link  
217:13 a key. Cool. Generate me a simple square  logo, minimalistic, uh, straightforward,  
217:20 but immediately interpretable and high-end for a  lead scraping SAS called Lezo. And then over here,  
217:26 the aspect ratio is going to be 1 one. And we're  going to see if we can generate a 2K um, result. I  
217:32 should also probably find some inspiration logos.  This looks pretty good just because it's simple,  
217:38 straightforward, and obviously it's sort  of unique. If this gives me a bad result,  
217:42 then I'll say something along the lines of that.  No, sorry. Flat icon style. Look at the below for  
217:49 inspiration. Mimic this. Okay. And then going  back to our landing page at the very bottom,  
217:53 you can see we have significantly more context  including the logo, bunch of information down  
217:57 here, and so on and so forth. I've also been told  that we should be good. Um, I just sign in here  
218:07 on this little orders panel. This looks like it's  opening up something, but it's not showing up. So,  
218:12 I'll just have to clarify a little bit more  what's going on. When I click the three dots,  
218:19 it looks like the menu is being hidden within  the div. Then, if I go back to AS Studio,  
218:27 we see we have this. This isn't bad, to be honest.  Great. Remove the text. Let's just keep the shape.  
218:32 Okay. And I don't know if you guys could see  this, but at the very top right hand corner, we  
218:35 now have our little custom logo put in. Um I mean  like I think this could probably be a little bit  
218:40 bigger. So why don't we make the background color  a little lighter gray? Looks good. Why don't we  
218:44 make the background color a little lighter gray?  Okay. And then if I go to get started, you can  
218:49 see that we now have that little new Leadzo logo.  And the colors here match the uh sign up button.  
218:54 So I mean, is this going to win any awards for  its design? Yes, of course. It's it's gorgeous.  
218:59 No. Uh, jokes aside, you know, maybe I had a real  logo designer come in here or maybe I iterated a  
219:03 little bit more on the logo, we'd have something  that looks a little bit better. But the rest of  
219:06 it looks real clean. I mean, if I don't know, this  was an Nvidia logo or something. I probably think  
219:10 this is like an Nvidia style interface. It's  wonderful. Um, so looks great. Let's launch  
219:15 this thing. Now, because we've done a fair amount  of work on Netlefi already, we had to use Netlefi  
219:20 to like do the backend functions or whatever. That  was the design um pattern that the agent decided  
219:26 on. This is already on Netlefi. All we really have  to do is just change the URL and then create the  
219:31 site itself. Just jumping into Netlefi here. There  is now a leadso project created a few seconds ago.  
219:37 Um that is manually deploying. I don't want it to  manually deploy. I just want to show you guys how  
219:40 you can link it with GitHub. So I'm just going  to pause this and say set up GitHub as well and  
219:44 then link the two. And so this now apparently is  online. If we go here we have leadzo.netifi.app.  
219:52 So this is actually on like a real URL. So we  could see up top. Um, I think this might have  
219:57 made the uh logo a little bit brighter than I  was initially anticipating, which is cool. But  
220:00 remember that I want this to be um, you know, on  GitHub as well. So, that's what we're going to  
220:04 do next. First, it's going to initialize the  Git repo. It may ask you to sign in if it's  
220:07 the very first time you've done this before.  I've done this many times before. So, my um,  
220:11 anti-gravity instance is basically already linked  to my specific GitHub account. It's now going to  
220:15 do what's called git ignore, which is take a bunch  of these files that probably shouldn't be pushed,  
220:19 like all of my API keys, then ensure that those  don't actually go live. Now it's creating the  
220:24 actual thing and then linking it to Netlify. It'll  handle whatever errors it has in its own syntax  
220:29 mostly autonomously. Just checking for mobile  optimization as well by making my browser window a  
220:34 little smaller and then dragging it all the way to  the left. This looks pretty clean to me. I don't  
220:37 see any major issues. Nothing so far. Nice. Cool.  All of these are arranged the way they should  
220:45 be. Pricing. Excellent. Okay. Get started today.  Beautiful. Now, how about the actual login page?  
220:52 Yep, that looks good. Then how about Nikki? Nikki  is the password. Okay, so the actual app page not  
220:59 super mobile optimized. Everything else is. So  what I'm think I'm going to do, which is my final  
221:03 command essentially, is excellent. The landing  page and a sign-in pages look great, but the um  
221:11 dashboard stuff after you sign in is not optimized  for mobile. Make sure that it works on mobile  
221:17 displays as well. And while all that stuff's  going on, I'm also going to do our security audit.  
221:21 What's cool is I could do a lot of these things  in parallel side by side. So whereas previously  
221:25 I had Cloud Code do the security audit, I'm not  just going to paste this big security prompt into  
221:28 Gemini 3.1 Pro High for a couple of reasons. One,  I've already used like $800 of my Cloud Code limit  
221:34 at this point and I don't want to spend anymore.  Um, another is that the Gemini model uh rate limit  
221:38 runs out at a specific time and then like it's  usually back within an hour or so. I think they  
221:42 do it like every 2 hours or 3 hours or something.  So yeah, I just uh Oh, and the third is I just  
221:47 want to test how it is compared to the Cloud Code  experience. So, we're going to allow this whole  
221:51 conversation here and then it's just going to  run through top to bottom and do that security  
221:54 audit. Um, what's cool is now we're pushed to  GitHub, right? Because we're pushed to GitHub,  
221:59 we can actually do a lot more of the um kind of  like uh version control based security things that  
222:04 we couldn't do that last time because it was more  of like a because we had just dragged and dropped  
222:07 the folder directly into NFI. So, it looks like  it's doing pretty well mapping out the front end  
222:12 and backend LFI functions entry points, checking  ignore for secret management. I'm just going to  
222:16 let this go on its own and then uh along with the  designed components and stuff like that, we can  
222:21 get to that um when it's all done. Heading back  to the app, which is now online, I'm actually just  
222:27 going to do one final mobile optimization check.  So, that looks really nice and solid. I'm now  
222:32 going to sign in. We'll go nickniki.com. And then  I think the password was like Nikki Nikki. Um this  
222:40 new scrape job here is looking really clean. As  you can see, we actually have that functionality.  
222:44 Let me just move my head. We actually have that  functionality now where all of it is contained  
222:48 within like a mobile browser um view which is  really cool. And then we also have that little  
222:54 navbar. So I'm going to head to orders. Orders you  know it's going to be a little tougher by nature  
222:58 I think but uh included this little horizontal  scroll wheel. And so I'm still technically going  
223:03 to call that mobile optimized. I guess we could  just like redo the whole structure and have job  
223:07 ID here, target request here, profile scraped  here, but it'd be a little bit annoying to do  
223:11 that. And I kind of like having a table. Tabular  data is um notoriously different to turn into like  
223:17 mobile optimized form when you're working with  tables and HTML. So I think I'll just leave that  
223:21 there. This page looks really clean. I really have  no issues. Um I'm I'm honestly quite pleased with  
223:26 the performance of Gemini and Claude Code together  on this app. So 10 out of 10 would do again. Um,  
223:32 we do have one more feature I guess which is that  feature that I talked about earlier where um,  
223:37 you know, if it doesn't scrape the leads through  Appify, uh, sorry, the Appify codec crafter actor,  
223:43 then I wanted to pass that off to Google Maps.  Uh, I think for the purposes of demonstration,  
223:48 like we're pretty close to just wrapping up the  app. And the purpose of this was not to do like  
223:51 some super crazy endto-end uh, like 500 different  platform scraping flow. The purpose is to show you  
223:57 guys how to wrap around an API and how simple that  is. So, I think I'll probably just leave that as  
224:00 that actor for now. But uh rest assured you could  very quickly and easily weave in uh you know logic  
224:05 like hey if this thing returns zero leads then I  want you to use this other approach. Um yeah that  
224:11 looks pretty good to me. I guess we could also add  that post any mailfinder step. H I might just do  
224:17 that instead once we're done the security audit.  Um generally speaking though you do want to do  
224:22 the security audits at the very end. So might be  skipping a step here. Not that big of a deal. Just  
224:26 wanted to make sure we're all on the same page.  On the security side, if I just open up this agent  
224:29 window, you can see that it has actually found  some major issues. Found major off bypass in the  
224:33 API routes. It also found a hallucinated enthropic  package dependency. So, it's preparing some final  
224:38 checks. That is a big deal, right? Um, you know,  it's done more or less exactly what I've asked it  
224:44 to do. Go through with a fine tooththed comb and  see whether or not we can access routes without  
224:49 actually having any sort of credentials. I can  open up the task right over here. And, uh, this is  
224:53 where we're at. The first phase was where it was  just identifying and running through the codebase.  
224:58 And the second is now where it's running through  every single step here sort of like a tabular way.  
225:03 It's now finishing up the report generation.  We're going to take a look. So this looks like  
225:08 it's quite critical. We have authentication bypass  and paid API routes, some hallucinated malicious  
225:13 package dependencies, missing with check clauses  and so on and so forth. So, what we can do is we  
225:18 can actually go through this whole thing and  not have this hallucinated anthropic package,  
225:23 but instead use the atanthropic-ai- SDK. So, I  think what I'm going to do is we're probably just  
225:31 going to push it. Okay, great. Do the security  fixes, make sure the app still works, of course.  
225:39 And now we're going to run through this big list  of detailed findings and fix it. Of note, this is  
225:43 significantly less secure than u the other app.  just off the get-go. And it's probably because  
225:48 we have some additional functionality where  we're basically like piping through API requests  
225:52 and whatnot, right? We're also using a slightly  different framework. We're using Vit instead of  
225:56 Nex.js, which I think solves a few of these. Um,  in total, I think the score is like something like  
226:01 70% or so, whereas the previous app, I think,  was a little bit higher. But yeah, you know,  
226:05 some of these are quite quite severe. The fact  that you could bypass um the authenticated user.  
226:10 So if they're not authenticated, it skips the  check entirely and proceeds to use the fallback  
226:14 process API keys to execute the scrape. Any  visitor can hit this endpoint via curl or postman  
226:19 without authenticating and your server will gladly  run expensive API ampy jobs, open a analysis or  
226:23 anthropic calls all build to my personal account.  This is the same sort of issue that um I think  
226:28 there was with OpenClaw at the time where um  you know you could just like bypass one of  
226:32 the routes entirely, go on like their dashboard  and then if they had some little server bypass  
226:38 uh setting or flag not like explicitly opted out  of then you could actually just use their own  
226:42 keys. And so in this way a lot of people had their  keys used not only leaked but also used and run  
226:47 up something like 25,000 people. It's yeah quite  a big deal. Looks like we're now installing some  
226:51 input validation libraries and stuff like that  and it's fixing the bypass. I'll get back to you  
226:55 guys when all this stuff is done. Excellent.  Looks solid. We now have everything that we  
226:59 need. I'm just going to take a quick peek at the  final version of the app. Make sure everything  
227:03 here works the same way that it did initially.  Opening up another list here. And this looks  
227:07 solid. Excellent. So hopefully you guys now see  how you guys could use Gemini in conjunction with  
227:13 cloud code and anti-gravity to build an app not  only that has full stack database authentication,  
227:18 everything else, but that also wraps around like a  pre-existing API. As mentioned, a lot of fantastic  
227:23 um SAS apps nowadays will either incorporate a  wrapped API as like their main functionality or  
227:28 they'll like include this as one of the five  or 10 or 15 different things. Realistically,  
227:32 like the software is no longer the moat. It's not  really that like, you know, you're going to build  
227:36 something super incredible that solves a problem  that nobody else has ever solved before and then  
227:41 you're going to package it in a beautiful UX that  nobody has ever put together before and then, you  
227:45 know, put it on the internet and then that is what  is going to distinguish you out from from somebody  
227:49 else. And so in this way, you know, there are a  lot of short-term opportunities you could take  
227:52 advantage of through app creation just by wrapping  pre-existing APIs that already do most of the  
227:57 heavy lifting for you. And hopefully this Appify  example um was was a a good show of what that  
228:02 might look like. And by the way, just so we're on  the same page here, this is just one tiny little  
228:06 actor of like thousands of different scrapers that  you could use. Just scrolling through Appify here,  
228:11 you guys see there's stuff like uh Trip Advisor  review scraper, DraftKings Scraper, there's  
228:15 Instagram scraper, website content crawler,  tweet scrapers off Twitter, land.com search,  
228:20 Instagram comment bots, and so on and so on and so  forth. You could build whatever functionality you  
228:24 want just by wrapping them around this service,  which is quite affordable um and you know,  
228:29 straightforward and easy to use. And there also  a bunch of other services out there that are like  
228:33 this that do um similar things. So, I'm not going  to tell you how to use one service or another,  
228:37 but yeah, like this Appify or API wrapper game  is super super powerful and it's blowing up right  
228:43 now. The one thing that we didn't build today,  obviously, was the payments integration, which is  
228:47 sort of like that final missing piece. Um, so I'm  going to talk a little bit about payments now and  
228:50 then after I'm done talking about payments, we'll  actually go through and build our next app, which  
228:54 is going to be one that allows us to go uh allows  users that use our app to go from everything from  
228:58 like going on a landing page, clicking buy, buying  the product or service, having some tokens added  
229:03 to their account, credits, and so on so forth,  going through like a full-fledged onboarding,  
229:07 and then once the onboarding is done, basically  being able to use the app. And in this way,  
229:11 we've successfully built up both our understanding  and then our capability from plain old websites to  
229:16 um what will finally be like full stack apps that  combine everything that we've talked about here.  
229:21 Okay, so let's have a brief conversation about  payments. We've obviously set up authentication.  
229:25 We know what like the the general flow looks  like. When somebody signs up to your app, they  
229:28 can either get some sort of email notification  that they can verify after they're done with  
229:31 that. Then obviously they get they get access.  A lot of this is handled through authentication  
229:35 middleware and routes and stuff. Um, but basically  in our case with payments, we're going to add one  
229:40 final step. So, basically the user is going to  start by signing up, right? So, we already know  
229:44 how to do this. It's pretty straightforward.  It's off. After they sign up, they're going  
229:47 to be taken to their own private payment page  where they can pick a plan or buy credits. Now,  
229:52 if you're doing like a thumbnail lead gener a  thumbnail generator app, for instance, obviously  
229:56 it's probably going to be credit based. Um, if  you're doing something that maybe provisions  
230:00 access to like a subscription service or something  where it's like, you know, you pay once and then  
230:03 you do it monthly, then we'll probably do a plan.  um after the payment occurs, they'll get access,  
230:08 which is referred to as entitlement. They'll  use the app, which is called consumption,  
230:12 and then in some cases, like if you're using  credits, they may run out, in which case they go  
230:15 back to the payment. That's sort of how it looks  under the hood. In terms of like what's going on,  
230:21 um we're not going to do the payment processing  ourselves. Basically, everybody on planet Earth  
230:25 now just uses pre-existing options like Stripe.  And the whole idea is when a user clicks subscribe  
230:29 or buy credits or whatever um we just have an  agreement with Stripe where via API we redirect  
230:35 them to a Stripeh hosted checkout page. Stripe  already has all the language and handles the card.  
230:39 They do things like advanced fraud detection.  They do things like you know handle your tax  
230:43 obligations like receipts and stuff like that and  then they just send all of that stuff to the user.  
230:48 We never touch a credit card number. We just get  paid you know in return for doing this. Stripe or  
230:53 these other payment processors they usually make  like a couple percentage points. I think Stripe  
230:56 right now is 2.9 plus 30 cents per transaction  or something. Um, I'm very happy to pay the  
231:00 2.9 cents not to have to deal with this because  in addition to this being like a real cluster,  
231:05 you know, don't want to swear too much on this,  but in addition to being a real pain in the ass,  
231:09 uh, it is also kind of like a liability legally.  And, you know, as app developers, I want to  
231:13 focus on building the app and marketing it and  stuff. That's what 99% of my work needs to be,  
231:17 not handling the legalities. So when the payment  succeeds, and it will if you know the the card  
231:23 is not fraudulent, if they cross their tees and  dot their eyes, Stripe will send our backend a  
231:29 web hook, which is basically just like an event  notification that says, "Hey, this user paid,  
231:34 this user that you've already signed up for."  So we have all their information and so so on  
231:38 and so forth. Their ID um is good to go. Then your  backend will receive that. It'll update a field on  
231:43 the user's record in your database. So in our case  our superbase that'll be something like plan is  
231:47 now pro or credits are 50 and then your app knows  what it's allowed to do and then you know as it  
231:52 uses and consumes those credits every single time  it'll basically make a request to the database  
231:55 just reducing that a little. Um and then you know  because it's like this you can now do a variety of  
232:00 of cool features. You could provision additional  app um credits or usage if you want. You have the  
232:06 ability to like upgrade people, downgrade  people, moderate your app and everything.  
232:09 So, three models are onetime purchase, which  is basically where you pay once, you get  
232:13 access forever. It's really easy to do. You're  basically just selling a service, right? So,  
232:17 you just flip a boolean on their account. Um, they  sign up via the Superbase O function. Now, they  
232:23 have like a little like user table. Inside the  user table, there's a field that says subscribe,  
232:28 question mark, and then it's either true or false.  In subscription, it's pay monthly and then the  
232:32 access continues. Stripe has a bunch of features  inside of it that manage recurring subscriptions  
232:36 or billing um on let's say a monthly or a weekly  or an annual basis. And basically what happens  
232:41 is once you have the payment success or payment  failure um you'll receive a web hook and then  
232:45 you just toggle that access and then Stripe will  manage that for you every single month. Basically  
232:49 uh whenever a a successful payment comes in,  you'll keep it. If a payment failure comes in,  
232:53 then you'll flip it. And then finally that  credit based usage where you basically just  
232:56 buy a bucket of different um tokens or credits or  whatever the heck. every time they hit your API  
233:01 to generate a thumbnail or scrape your leads,  you decrement the balance. When it hits zero,  
233:04 you will send them back to checkout and then  force them to do it again. So, all of the apps  
233:07 that we built in this course can use any of  these. Uh, if I was doing lead scraping, I  
233:11 would probably do either like a monthly plan with  credits built in or I just do credits. Um, for  
233:15 our thumbnail generator, we're just going to do  credits. But, you know, you you can use whatever  
233:19 combination of these you want as well. The three  things we're going to need in our codebase,  
233:23 and keep in mind like our models are going to  handle this for you. You don't have to do this,  
233:26 but basically we need an endpoint that creates  a Stripe checkout session and then redirects the  
233:30 user. So after the user signs up, they're going to  have to immediately go to this endpoint maybe with  
233:34 three maybe a page with three different pricing  plans or something. Then when they click on that,  
233:38 it'll redirect them over to Stripe with their  information. Then we need a web hook endpoint,  
233:43 which is just like a little um route on our  server that basically waits to receive a payment  
233:49 succeeded event from Stripe. Then it'll update our  database. And then finally, we need some way to  
233:53 check inside of our app before doing the expensive  thing like generating a thumbnail or whatever,  
233:58 does the user have valid uh plan? Do they have  valid credits? So that's exactly what we're  
234:02 going to do in the next product build. You guys  are going to see this hands-on. I just want you  
234:06 to know that basically this just stands between  like the authentication step, okay, and then the  
234:13 um like usage step. Basically now there's just  going to be an additional one which is payment.  
234:18 Before obviously I let people use their own API  tokens and stuff like that in the app. So we sort  
234:23 of bypassed this alto together. Now it's going to  be you know like us doing the thing and therefore  
234:27 we're going to want to make some arbitrage. You  know they're going to pay I don't know $10 and  
234:32 then every usage is only going to cost us $1. And  so in that way we're going to basically make uh $9  
234:37 in profit. That's more or less how any meaningful  SAS business will operate. Okay. So hopefully you  
234:41 guys are as excited as I am. Let's dive into  it. So, in order to make any of this work,  
234:47 you are going to have to go to stripe.com and set  up an account. Um, I want you guys to know that  
234:51 if you don't have, you know, like an account on  Stripe set up, you're not really going to be able  
234:55 to follow along with me. But, at the same time, I  know that depending on where you are in the world,  
235:00 um, different people have different levels  of access to Stripe. They're not available  
235:03 in every country. So, if you guys aren't capable  of using whatever, you know, the specific solution  
235:08 I'm going to do, I recommend just use whatever  you guys have access to. Like, there's Payaneer,  
235:12 uh, I don't know, there's like PayPal. While these  services are not as clean and as good as Stripe,  
235:16 they don't handle as much of, you know, literally  global GDP at the moment. They're up to 1.6% which  
235:21 is insanely bonkers. Uh they're still pretty solid  and, you know, there's a lot you can do with them.  
235:25 Just trying to replace everything I'm about to  tell you with with that. So, you're going to  
235:28 have to go through some prolonged signup process  obviously and some of the stuff is going to take  
235:32 a while to implement. You're going to have to, you  know, verify your identity and stuff like any good  
235:35 payment platform. Um, but this is really clean  and it's it's a really good service. After that,  
235:39 you'll be taken to a dashboard that looks  something like this. in the top right hand  
235:42 corner there'll there'll be a little test mode  toggle which I'd recommend turning on. If you  
235:46 head to the top lefthand corner you can then go  switch sandbox and then click create sandbox.  
235:50 That's what we're going to do. Um when you go  through that little prompt uh create it or model  
235:53 it after you pre-existing Stripe account and then  you'll have an account that looks like this which  
235:57 basically will just model over your real account  and then give you the ability to like collect  
236:02 u publishable keys or secret keys at a  button. Make sure not to reveal these.  
236:06 These are super important if you did reveal like  your your real keys, not your sandbox keys. Um,  
236:11 a developer would basically have full access over  literally everything ever, which is kind of crazy.  
236:16 So, don't ever do that unless you want a random  $10 million bill in your Stripe account. They'd  
236:21 probably cap it after a few thousand, but yeah,  would not want that to happen to you. And then  
236:25 once you're done with this, you're you're actually  pretty good. We're just going to give our agent  
236:27 the publishable and secret keys, and then it'll  like handle the rest of the payments for us. Okay,  
236:31 so that's step number one. Step number two, I'm  just going to go back to anti-gravity. And then  
236:34 now, let's exit out of this lead scraping SAS. go  open folder and I'm going to call this thumbnail  
236:40 generator. Open this puppy up. And now we are  basically in a fresh uninitialized workspace  
236:47 here with just the folder name. And just like we  did the first couple of times uh to develop full  
236:52 stack apps. We're going to do more or less the  same idea. I'm going to go find a screenshot of  
236:57 something that I like that I think looks really  clean and that I want to model my app around.  
237:00 And then I'm just going to like voice dictate  give it a big dump of what it is that I want it  
237:03 to do. And then I'm just going to have it like go  and give me its demo. Um, once we have that demo,  
237:08 we can like in that little mining gif, uh, or  sorry, mining meme, we can like progressively  
237:13 mine all the BS until eventually we hit gold.  And once we've hit that little gold cavern,  
237:17 then we can actually do some work and make our  app better and better. Um, at a very high level  
237:21 though, uh, let me show you guys how I'm thinking  that this app is going to work. So, basically,  
237:26 um, just like we had before, you know, we want  a landing page. That landing page is going to be  
237:31 like any relatively simple marketing page, right?  What's going to occur is there'll be a little  
237:36 payment section. When they click on this payment  uh button or whatever, basically what's going to  
237:41 occur is they'll be taken to the authentication  or a sign up. Um after they sign up, they'll be  
237:46 taken to the ability to pay and that'll be  like the one click that takes us to like the  
237:50 Stripe checkout. We're going to do this based off  credits. So after they pay, we'll then provision  
237:55 access to the app. What's the app actually going  to look like on the internal side? Um well,  
238:00 it's going to be pretty straightforward. You're  going to have a dashboard which allows you to see  
238:03 all of your generated apps, uh, sorry, generated  thumbnails and everything like that. You're  
238:07 going to have like a config page similar to the  scraper that allows you to do things like upload  
238:12 images of yourself. Um, when you upload images of  yourself, uh, you'll be able to basically have the  
238:16 model generate thumbnails for YouTube or other uh,  content platforms that include you. Uh, and then  
238:22 we're also going to have like the generate page.  And this generate page is where we're going to  
238:26 be able to, you know, put in some some prompts  and and stuff like that. and then it'll pass  
238:31 through. And the idea is what we're going to do  is we're going to pass through using nano banana.
238:38 And the way it's going to work is we're going to  feed in um like an inspiration thumbnail. And then  
238:46 in addition, we're also going to feed in a prompt  with changes. And then it'll do its best to model  
238:52 that. And it's not just going to do it once, it's  going to do it, I don't know, like three times.  
238:56 And then it'll spit out, you know, three potential  options, A, B, or C. And then you just pick one.  
239:04 And this is going to be the one that gets saved.  This is going to use our Nano Banana Pro key. And  
239:08 so in that way, it's going to be consuming our app  credit usage. Um, we're not going to go through  
239:12 the process of like signing up a developer app or  anything like that to to let that happen because  
239:17 sometimes some of these platforms are like, hey,  if you want to resell our tokens or if you want  
239:20 to resell credits, you need to go through  some additional steps. Um, but basically,  
239:23 in a nutshell, that's going to be your completed  one. And then this is going to show up in orders.  
239:27 And then every single time we do that, if you  think about it, we're also going to um consume  
239:31 some credits that they have because they're  buying based off of credits. So I don't know,  
239:36 maybe every time we generate we'll consume three  credits and then I don't know, maybe like it'll  
239:40 be like $1 equals 10 credits or something. We'll  keep it really simple obviously, but if you were  
239:47 a business, you'd actually want to run through  like the economics of pricing and stuff like that.  
239:50 Okay, so pretty straightforward, pretty simple.  Um why don't we start with that first step which  
239:54 is finding some design inspiration. Then I'm going  to do a big voice dump. After we're done with the  
239:57 voice dump and the design inspiration, we're going  to produce it locally. Once we produce it locally,  
240:01 we'll iterate test. Then we'll push it um to like  actual databases. We'll iterate test. Then we'll  
240:06 make it live on Netlefi. Iterate test. Then  finally, we do a security audit. Same thing  
240:10 that we did before. Just going to head over here  and I'll type web app because a lot of these are  
240:14 mobile based and I just want this one to be  not mobile based for now. I want this to be  
240:19 really clean. In fact, I want this to be like the  cleanest, sexiest app that we've developed so far.  
240:23 So, I really like this design. That looks really  clean, but obviously it's not like a like an app,  
240:28 per se. I also really like this design. Okay,  that's really sexy. I'm liking this like dark  
240:35 kind of like glass. I don't know what it's called.  Isomorphism. Like, this is crazy. Activating new  
240:41 brain paths. Look at how sexy that is. So,  we're going to go over something like that.  
240:45 This one's really, really good. This guy Jack R is  a freaking genius here. He actually included the  
240:50 font down below as well as the color palette and  everything like that. Um, this is for like some  
240:55 app that supposedly monitors your brain waves,  which is wild. Uh, we're obviously not making  
240:59 an app that monitors brain waves. We're making an  app that uh gets me more subscribers on YouTube,  
241:03 but still super clean. I like this. And the fact  that he provided the fonts and stuff means that  
241:07 I'm going for it. So, I'm just going to take this  image here, copy it, paste that into thumbnail  
241:13 generator, and then I'm also going to feed in  just some design information. I'll say font lufka  
241:22 um primary F4 E18E secondaries, and then I'll go  D6, D6, D5, and then D. Is that a B or an eight? I  
241:35 think that's a B. DB DB DB. Uh, anything else that  we need? We also have the white is going to be E3  
241:44 E3 E3 and then the black is going to be 043. Cool.  I'm just copying these over. I mean, obviously I  
241:50 could just provide a screenshot, but I don't want  to consume that many tokens for what ultimately  
241:53 amounts to just a few characters. Um, cool. This  looks really, really clean. I I like this. So,  
241:57 we'll see how this goes. I mean, I don't know if  it'll be able to replicate the glass eomorphism,  
242:00 but that's fine. Um, and then we'll close out  of these for now. Now that we're done with that,  
242:04 I'm just going to give a big voice dump of more  or less everything I've already told you guys. So,  
242:07 I'm not going to include this in the video so  you guys don't have to waste another few minutes  
242:10 hearing me talk about it. But, I'm just going to  go back here, double click my voice transcription  
242:13 tool, and then go for it. Hey, I'd like you to  help me build a full stack application. I will  
242:18 provide you with all the relevant API keys down  here for now. Build this so it works locally using  
242:25 JSON files as a standin for the DB. Later we'll do  the actual superbase etc setup. Cool. That looks  
242:35 pretty good. Um quite the long prompt here. That's  one of the reasons why I really like these voice  
242:38 dictation tools. You can just bang out, you know,  10 times the number of words. And what's cool  
242:44 is models are pretty capable at extracting your  meaning. Better just to talk like you're talking  
242:50 to a human being. Spit out whatever you can and  then the model will extract the important bits. So  
242:55 that's more or less what I did over here. It's now  going through this. It's generating um you know,  
242:59 a file structure using uh next.js, which  is what I asked it to use. Um and then,  
243:05 you know, I'm going to give it all the API keys  and everything like that after it's scaffolded the  
243:09 base structure. Okay. So, from here on out, we're  just going to sit around and wait for this thing  
243:12 to do its magic. Awesome. So, it just wrapped up  and I now have this live on npm rundev. I then  
243:18 navigate localhost 3000 and this is what it looks  like. So, this is unfortunately just calling it  
243:23 nano banana. I think it probably misunderstood  what I was saying, but that's okay. No big deal.  
243:26 Just going to refresh this here. See that we  have a bunch of the social proof pieces here.  
243:31 Global tech nexus, Nebula, and so on and so  forth. Professional thumbnail. Zero Photoshop  
243:35 skills required. That looks pretty clean,  honestly. Simple, pay grow. Pricing. I like  
243:39 that. Ready to blow up your channel. Create your  first thumbnail. Okay, so let's see what happens  
243:43 when I do get started. Takes us to this page.  I'll just say [email protected]. Continue with  
243:50 email. Now we're being taken to a payment page.  This is really clean. I like this. Um hm. Going  
243:56 to click pay $10 here. It's empty, but we should  still be good. And now we're on the thumbnail  
244:01 uh generator side. So look at this generation  activity. Number of thumbnails generated per day.  
244:06 No generations yet up here. Generate new will take  us to this page. And we actually have a pretty  
244:10 clean UX. We have a generate magic inspiration  thumbnail. Generation prompt generated options  
244:15 1 2 3. These are what are going to generate.  We have the number of credit cost. We have a  
244:19 settings page here with a reference face. We're  going to need multiple files. So I'm going to  
244:22 have to make that adjustment. And then you guys  can't see this because my face is in the way,  
244:27 but if I make this a little bit smaller, you see  how on the bottom left hand corner, we also have  
244:31 a credits available with a top up button. This top  up button is going to bring us back to this little  
244:35 payment page. Okay, so it's really clean, very  sexy, and pretty straightforward app. That said,  
244:41 I don't like the design. The design is sort  of like a big issue for me. So, I'm going to  
244:45 have it generate um a new design because I don't  like this. There are also some minor changes that  
244:49 I think need to be made. Notice how the margin  on the left moves around every time. Obviously,  
244:54 we have some minor UX changes as well, aside  from the design that I like to make. And uh yeah,  
244:59 why don't we pick a name? Why don't we call it  Thumbly? I think this is actually the the name  
245:03 of one of my startups now that I'm mentioning  it. It used to literally be called Thumbly,  
245:07 and it was something similar. I was like creating  uh thumbnails for people. Oh yeah, look at that.  
245:11 That's wild, man. I made this thing like probably  five, six years ago. That's a friend of mine,  
245:15 Zach, that since worked with Scaling with Systems  and a bunch of other people. Anyway, super clean,  
245:20 huh? Not gonna lie. Okay. Okay. This is how  it works. You basically basically generate  
245:25 thumbnails and then you can select them. Man, I  coded this by hand back in the day. Uh, okay. So,  
245:29 let's call this thumbly. And then let's also  make a bunch of changes. I'm just going to do  
245:34 some voice dictation. And then I'm also going to  like have it really push for that high-end white  
245:38 design style cuz I don't like this first demo. So,  I'm going to say let's call it thumbly feedback.  
245:45 And now I'm going to give it a bunch. The design  needs to be heavily modified. It does not reflect  
245:52 the provided image at all, unfortunately. On the  overview page, I'd like you to remove the average  
245:58 CTR impact section and then replace it with  something else that we would realistically have  
246:03 access to. The generation graph looks cool. When  I go between overview, generate, and settings,  
246:09 I'm noticing that the left hand margin of the  generated of the um elements on the right hand  
246:17 side are slightly off. I just want you to fix  them all so they're the exact same width. On  
246:21 the generate tab, auto prefill the generation  prompt with the user's templated prompt. Add  
246:26 a section under inspiration thumbnail where I  could just paste an image URL and then it goes  
246:30 and automatically populates it. Under settings,  we'll need to upload multiple reference faces. So  
246:36 have a ability to upload more than one, not just  one. And then in the default prompt instructions,  
246:41 this is where you'd automatically insert some  generated prompt in the generate page. So  
246:46 whatever the user writes here, I want reflected  in the generate page by default. And then the  
246:50 user can modify that slightly as necessary. On the  landing page, remove all references to nanobanana  
246:55 and replace with thumbly as mentioned. Add more  social proof here because we're running out of  
247:01 them pretty quickly as the business names scroll  to the left. Make the website landing page around  
247:07 twice as long in general. Finally, I'll say the  app needs a major redesign. Focus on the light  
247:11 theme as provided in the images above. Ensure  you're using the right font etc. We want light,  
247:21 not dark. Soft glows and futuristic glass style  aesthetics. Cool. I'm now going to give it some  
247:32 more time to make the required changes and  then I'll check back on the app. We're at the  
247:35 next leg of that redesign. So, I'm just going  to take another peek at localhost 3000. Well,  
247:40 it's definitely looking a lot brighter, which is  nice. Kind of weirdly yellow and green, which I  
247:45 don't really know how much I like. Uh, let's go  to the dashboard. Ooh, this is much better. Much  
247:49 better. I see that it's taking the inspiration  more on the actual dashboard side of things than  
247:53 it did on the landing page. And that's okay. Okay,  cool. It also fixed that margin issue that we had  
247:57 earlier. Um, I actually really like this. I'm just  going to zoom in a bit. Okay. I think what we need  
248:02 to do is we need to just widen it a bit. So, I'm  going to do the same thing. Hey, this looks much  
248:07 better. Could you widen the um elements in the  dashboard, the generate page, and the settings  
248:13 page? I feel like they're not very wide right  now, and that just looks silly. Um also, what font  
248:20 are we using? Are we using that provided font?  Some of the colors are a little too light now,  
248:26 I would say, like the highlight uh color on  the left handmost navigation elements. That  
248:33 light green or whatever it is, that neon color  is a little rough. So, I think we're probably  
248:39 going to have to reduce that just for contrast  purposes. Um, are we using real glass isomeorphism  
248:44 here? If you don't know what that is, um, I can  provide significantly more context. Anyway, uh,  
248:49 we do have some major issues on the landing page  side of things. This looks pretty poor. I think  
248:53 what I'll do is I'll just provide you a landing  page that I want you to copy verbatim like word  
248:57 for word because just the way that you're using  shadows and stuff like that looks really silly  
249:01 right now. Aside from that though, um, pretty well  done. I like this. I'll provide you that landing  
249:06 page and then you can try copying it verbatim  u to start and then we can make some other  
249:10 modifications afterwards. Here's the landing page.  Okay, that looks pretty solid. I'm just going to  
249:15 go and I'm just going to find a really classy  isomorphism glass. Okay, glass morphism, I guess,  
249:24 landing page. That makes more sense. Okay, we  just want one that like actually looks like a  
249:30 like a real app, you know, not one that's BS. So,  let me take a look what has like that light theme.
249:39 Okay, I like this. This is pretty good. You know,  one of my favorite things to do is just to open  
249:43 up the local dev server and then just watch  as it slowly but surely updates the page and,  
249:49 you know, incorporates your device. It's kind  of neat. So, it's just done that up here. Um,  
249:53 it's just using some fake logo here to try and  mimic it, but obviously it's not the same thing  
249:57 as the actual logo. It's like adding small little  elements here or there, changing like the colors  
250:03 and the background and and stuff like that. Um,  so that's that's pretty cool. Quite sexy. I really  
250:07 like um being able to see that. It's doing the  same thing underneath with like the soft gradients  
250:11 and then obviously changing the uh the colors  and stuff because some of the elements have not  
250:16 been adjusted yet. Obviously, the background  in this case and the text is black. But yeah,  
250:20 as it continues, it'll just get closer and  closer and cleaner to that actual pixel perfect  
250:24 uh uh generation. Website's slowly getting  better and better. It's getting cleaner. Um  
250:28 all these elements now are starting to match  color-wise. We even have that little down CTA  
250:35 with that light blue. So, this is a much better  approach, I would say. Some other adjustments  
250:38 I went through and then in uh a lot more detail  basically deconstructed everything that I wanted  
250:43 changed. So I wanted to make the nav bar a little  bit square. I wanted to fill out the rest of the  
250:47 dashboard underneath the hero header, make the  social proof section infinitely revolving and so  
250:51 on and so forth. And the idea here is we just  want to get the app as close as uh possible  
250:55 to basically perfect aesthetically speaking  before I do all the rest of the work. Okay,  
250:59 we got the website right over here looking really  clean. Notice how we've gone through and then  
251:03 actually adjusted the dashboard down here. That's  really, really sexy. Um, I'm not liking the slight  
251:08 difference in Oh, actually, no. That's basically  perfect. Why don't we just, um, I think I'll  
251:13 probably square out all the buttons everywhere.  But anyway, this stuff's looking real clean. We  
251:17 have the pricing down below, ready to blow up your  channel. I'm just going to go to dashboard. Let's  
251:22 just see how the actual functionality of the app  is. This looks pretty good. Um, I don't really  
251:27 like the light blue in the back, so I'll adjust  that as well. But everything else is looking good,  
251:30 right? Like the margins are basically all the  same. Um, we also have emphasize contrast. So  
251:35 now the generation prompts are autopop populating.  And yeah, I mean, I'd say this is pretty great. I  
251:40 don't like the background blue on the um inside  of the app page, like the dashboard to generate  
251:47 the settings. That component looks weird to me.  I'd like some collection of soft gradients on the  
251:54 background of the hero header in the landing  page. And then I'd like you to turn all pill  
252:02 shapes right now into those rounded squares  instead. I think that just looks a lot better.  
252:08 For instance, the signup button in the navbar,  the real-time generation from your prompts widget,  
252:15 and the dashboard in the hero header um image. I  want all of those little pill shapes converted.  
252:24 Aside from that, change define to thumbly. I  want you to make sure that those two buttons,  
252:30 filter by CTR and plus new thumbnail, are just on  one line. Right now, they're both on two. Adjust  
252:35 the verbage if necessary. And then everything  else looks really clean. Yeah, that's about  
252:44 it. All right, this is looking a lot cleaner. We  have everything that we need on the landing page.  
252:49 Um, I like the fact that the cards no longer  look that ugly, atrocious blue. At this point,  
252:54 all we really have to do is provide  all of our API keys and tokens. So,  
252:58 as mentioned, I basically want to reproduce  the functionality of AI studio.google.com,  
253:03 except I want to be able to provide a thumbnail  and then a bunch of my own faces and then just  
253:06 have it replace the image in the uh thumbnail, or  rather replace the person's face in the thumbnail  
253:11 with my own. This is a pretty common thing to do  on YouTube nowadays. Basically what you do is you  
253:15 find like outlier really high quality thumbnails  and then you just replace them and make some minor  
253:19 adjustments. Um I'll show you guys what I mean.  If you head over to this app here, one of 10, and  
253:24 then you go down to, I don't know, videographer or  something, and you know, just look for something  
253:31 like this by this fella here, Kristoff Oliver. I  mean, like, is this the best thumbnail? Actually,  
253:36 no. Let's use this one. This one's probably  nicer. So, you see how there's like, you know,  
253:40 some fella here in the middle and then there's the  text around it and then there's some little image.  
253:43 So, I mean, like, you know, I think you'd probably  design something better than this, but still. I'm  
253:46 just going to save this image. Head over to AI  Studio. I wanted to just switch my account to  
253:51 one that actually has credits. What I'm going to  do is I'm going to stick this puppy inside. Then,  
253:56 I'll say replace the man in the thumbnail with  the man in the profile picture provided. Then, I'm  
254:05 going to look for a picture of myself. I like this  one. This one's pretty funky. So, I'm going to  
254:10 say replace the man in the thumbnail that includes  how to build a portfolio with a man in the profile  
254:17 picture holding a trophy. Okay. So, what I want  basically want to do, what I'm instructed to do is  
254:24 to take this and then just swap it out for this or  something close to that. So, I press enter. You'll  
254:29 see that it's not going to do the most amazing  job right off the bat, but it's going to do an  
254:33 okay job. And basically the whole idea is we're  just going to run this multiple times until we get  
254:37 like three or four options. And then we're just  going to present them to the user and then they  
254:40 can just select one that they really like. And  scrolling down here, you can see that we've since  
254:43 replaced that man just with me holding my trophy.  So it didn't get like exactly what I wanted to do.  
254:48 Obviously I don't actually want to be holding the  trophy, but you can see that I'm in this fell's  
254:52 room. And for the most part, it looks like me. I  mean, there's some minor little changes in terms  
254:55 of the skin texture and stuff. Um I think my skin  texture is worse and it's like done some done some  
254:59 touch-ups, but you know, it's got my hair. It's  got me holding the trophy. didn't add any extra  
255:04 AI fingers. And yeah, that's more or less what  we're doing, right? So the question is, how do  
255:08 we actually do this in our app, not inside of this  as studio.google.com thing and it's pretty easy.  
255:13 There's just like a bunch of API routes for image  generation. We just basically need to reproduce  
255:17 the same flow that I just did here over there. And  then we need to um connect API keys and tokens and  
255:22 stuff like that. So I have an API key over here.  Uh I'm just going to I don't know. I guess I need  
255:27 to find a way to copy this API key. So why don't  I go to get API key here. We'll create a new one.  
255:31 And I'm just going to call this um thumbly. Going  to click create. We're now going to have the key  
255:38 generated right over here. Okay. And I'm just  doing this in a new one because I want to be able  
255:43 to turn it off. I'll then go back to anti-gravity.  And then, okay, it's just done some changes for  
255:49 me. I don't know exactly what. Then I'm going to  open up av file and then I'm going to say Gemini.  
255:55 I think this is Gemini, right? What is this one  technically called? uh maybe like nano banana API  
256:03 key and I'm going to paste that in. Um there are  a couple of other things that we need if you think  
256:07 about it. So this is not enough. We obviously have  to go through the whole superbase spiel and then  
256:12 get a bunch of images. Sorry, I get a a database  from there. So I'm going to set up a database on  
256:17 Superbase just like I've done before. Head over to  superbase.com or rather I think it's database.new  
256:23 is even faster, right? Yeah, there we go. I'm  going to do it inside of proposal generator.app  
256:28 maybe gyro. Okay, never mind. I need to go and  delete one of these old ones, so bear with me  
256:31 one second. This is in the leado database, right?  So, I'm just going to delete this project. Leadzo  
256:36 DB. Cool. Thank you. Delete. Um, and then  I'll go to database new. Once that's done,  
256:44 now it's going to say, hey, let's make one. I'm  going to call this one um Thumbly DB. Again,  
256:50 I'm going to generate a password. Copy that  password over. Enable. Oh, cool. They also now  
256:55 have automatic RLS, which is cool. So, I'm going  to create a new project there. And what I want is  
256:59 I basically want to um paste in my database  password. Let's say superbase DB password.  
257:08 Paste that in. Um I don't know, superbase  project URL. Paste that in. Previously, I  
257:14 did this just directly inside of the um you know,  agent chat, but this time I'm actually going to  
257:17 copy and paste it over to the CNV. So, Superbase  publishable API key. I'm going to paste that in.  
257:26 I think there are a couple other ones. You know,  I just want to give it my API key as well. So,  
257:31 we'll go superbase project ID. Paste that in.  You can see it's the same thing as in my URL. So,  
257:38 basease project name I'm just coming up with all  excuse me, I'm just coming up with all this stuff  
257:45 um myself. Like I'm coming up with these names  myself. Um these aren't really important. The  
257:49 model can understand the context themselves.  Okay. So, I I have more or less all of that,  
257:54 which is cool and pretty clean. Um, and then yeah,  now it's just provisioning my database and API  
257:59 input. It's actually creating them for me. So once  all that stuff's good to go. I'm just going to  
258:04 paste that in. Any other additional functionality  we need? I don't think so. So I'm going to say is  
258:09 I've added a bunch of uh credentials for nano  banana and then superbase um into the env file.  
258:17 You now have everything that you need in order  to actually build out the functionality of the  
258:20 app aka the API calls. I'll go call or find the  um specific endpoint to reference. But basically,  
258:26 we're going to be replicating the functionality  of um Google's AI Studio just inside of my app.  
258:33 Let's see. ENV. Cool. And then, you  know, there should be the ability
258:43 to grab the code. Okay, cool. And there  is. So, I can grab this code. And then  
258:48 we're just going to go rest here. And  then what I want is I basically just  
258:54 want to copy all of this stuff. And  then I'm just going to paste this in.
259:05 Here is some REST code straight from their  dashboard. And then I'm just going to paste  
259:10 this in, which should be everything that they  need in order to actually like generate the  
259:13 the thing. Keep in mind we're going to  be supplying two inputs. The provided  
259:20 thumbnail image and our prompt uh I guess  three inputs. The provided thumbnail image  
259:29 um images of the person's face and then a prompt check there. Let me see. Are there any docs  or something like that I could use to give  
259:43 me more data? Okay, check their docs if  you need more on how to do this. First,  
259:51 test all of this via curl to make sure it  works. And then once it works, you can put it  
259:58 on the app. And then finally, I'm going to just  provide some base um data here. So, you know,  
260:04 earlier I had that portfolio builder image. So,  I'm actually just going to supply these images.  
260:08 Um, so this is going to be one image. I think  we can just open this in a new tab. Okay. No,  
260:14 I guess we can't. So, let me just save the  image. This is going to be the supplied. Okay,  
260:23 this one over here. Then I'm going to move into  anti-gravity. Just upload this to the agent. I've  
260:30 provided you three things. supplied the image to  insert a face into um face the image to get the  
260:46 face from and then prompt and the prompt is going  to be here. Let's just save this too. Uh this is  
260:56 pretty big actually. I don't know if we'll be  able to feed all this stuff in. Let's say face
261:05 And uh it should be able to pull all  of this data up from just the names.
261:11 Okay. In case like unless we get blocked, we  should be good. Just going to double check this  
261:17 because we we have supplied like a lot of image  files here. I mean this was like 4 megabytes.  
261:21 This here was like I don't know megabytes. So  in total it's going to be like five or six megs  
261:25 which is actually pretty chunky. Maybe they like  automatically resize the images or something so  
261:29 it's not taking up a massive amount of tokens. But  um you know I always do kind of worry like oh am I  
261:33 going to run out of usage and stuff. Okay. So I'm  gonna allow this conversation. It can do more or  
261:37 less whatever the heck it wants. And then um just  double checking thumbly. Yeah. I mean like this  
261:46 honestly looks really clean. I'm liking this quite  a bit. Nice. And then we have like this infinitely  
261:50 revolving social proof section here which you know  you stick your own social proof in. And then all  
261:55 of the um buttons and stuff like that are square  now which is cool. And everything is just a little  
262:00 bit more squared off. I don't actually like this  shadow. There's a slight shadow in the background  
262:03 here, which hurts, but not the end of the world.  We have the ability to change this. This is nice.  
262:08 Cool, cool, cool. Awesome. I mean, like, you know,  I think this is basically like the the MVP of  
262:13 um a high quality app with the landing page. So,  wonderful. I'm just going to wait and then um let  
262:18 you guys know when we're done this next step.  Looks like it's already making API calls, which  
262:22 is cool. Just using everything that I provided it  below. Generative language. Google APIs appears  
262:27 to be the endpoint. That's pretty nice. And  so it's gone through and actually verified  
262:30 that this works and it's created um another step  inside of implementation plan. So now it's going  
262:35 through and then integrating this directly into  the app. So it's not just going to be doing it  
262:39 via its own curl requests, but I'll be able to do  it via uploads files and stuff like that. Okay. So  
262:44 just signing in here. We've now integrated the  um database functionality. Going to go over to  
262:49 generate and see if we can do this. So you know  I can obviously add images and we just did. So  
262:54 now we have this cool little blur effect which is  really nice. Let me see. Keep the style identical  
262:59 to the inspiration image. Make the main subject  highly cinematic with dramatic neon backlighting.  
263:03 Contrast. So no, I'm going to say but replace  with the user. Oh sh uh crap. I need to actually  
263:09 provide images, right? So let's upload some faces.  Let's do face. This is now me. Okay. And then here  
263:17 inside of the default prompt instructions, I'll  say replace the thumbnail image with the supplied  
263:25 user cinematic. and I don't know dramatic lighting  on the face. Replace the face of the person in the  
263:36 provided thumbnail image with the face of the  supplied user. Cinematic and dramatic lighting  
263:42 on the face. Let's save the config. I'm just  going to refresh this. Has that been saved? Okay,  
263:46 it is. This has been saved as well. Okay, that's  great. You know, might as well test one additional  
263:51 piece of functionality here, which is if I could  just paste the URL in. So let's go back here,  
263:57 paste that in. Okay, cool. So we've actually  we've actually now got this in the URL, too,  
264:01 which is clean. I'm going to click generate  variance. And this is really like the test  
264:05 of the functionality. So I'm going to give  it a few seconds. Then I'll double back with  
264:09 um you know, changes if it didn't work, and  then maybe some next steps if it did. Okay,  
264:13 we've actually generated one of the options here,  and it looks interesting. Um for whatever reason,  
264:19 look, it's like the background's kind of white. I  don't know. And we've only done one, so we want to  
264:23 do three, right? Obviously, we're going to want  three. I see there's some click functionality,  
264:27 which is kind of cool, but um yeah, we're going to  have to do more. So, the first thing I mean, the  
264:32 dramatic lighting looks freaking badass. Like, you  know, it's not exactly my nose, to be honest. It's  
264:37 a little bit different for me, but still really  cool. Hey, we've um generated only one, not three.  
264:43 So, just double check and make sure that there  was no issue in the background that prevented us  
264:47 from doing um all three simultaneously because we  want to populate that with three thumbnail images,  
264:52 not just one. Also, when the images came out, they  had this sort of like white sheen or overlay. They  
264:59 looked kind of transparent. I just want you to  remove that. Um, so that I just can see the actual  
265:04 image immediately. Uh, and yeah, I think that's  about it. What's interesting is while it did that,  
265:10 it consumed three credits. Okay, so the credit  consumption functionality is there. However,  
265:15 when I saved this, it saved it twice. So, that's  another thing we'll have to change. This time  
265:19 we're going to do the same thing just with the  supplied version instead. And then I should just  
265:24 be able to click generate variance and we should  be able to do three, right? Since we're generating  
265:28 three times. So credits available um on the left  hand side, which you guys can't see. Say 197 right  
265:34 now. Okay, now we have three provided variants.  These all look awesome. I'm loving this. Loving  
265:39 the the really badass lighting on the face. That's  neat. Um so now what I'm going to do is I'm just  
265:43 going to select one of the variants that I want  to keep. So why don't I select this one? That one  
265:45 looks probably the most realistically me anyway.  And then I'll click keep selected thumbnail. It  
265:50 says thumbnails saved to your gallery, which is  awesome. Okay. And then we go back to overview.  
265:54 And then we can see this new one here. Beautiful.  Okay. So, I guess what else we need? Like the  
265:58 first thing comes to mind is we just need a way to  like delete these, right? Provide a way to delete  
266:03 um the thumbnails under recent generations.  Basically, I want to be able to delete ones  
266:07 in my gallery so that I don't have to see them in  the database anymore. Um aside from that though,  
266:11 everything looks like it worked really clean. Oh,  um, instead of like the alert that comes up, um,  
266:16 just make like a nice modal, a nice toast message,  um, that then allows you to, um, I don't know,  
266:23 open it in a new tab or download the image or, uh,  go back and generate a new one. Also, just little  
266:28 things. I mean, these intelligences are so dang  good. Notice how in this one here, I don't have  
266:32 the O behind my neck, but in this one, I I do.  Like, it's, you know, it knows the placement of  
266:37 the O and where it should be and stuff like that.  But yeah, um now that we're done with all this,  
266:41 what we're also going to do is we're going to  uh like after we get the little X functionality,  
266:46 we're going to hook this up to Superbase and then  it's just going to give me that migration script  
266:50 so that I could use this in like real production  scenarios. Then on from there, it's connecting  
266:55 it to Stripe so that the payments work and then  yeah, everything's just going to be 100% live,  
266:58 which I'm really excited for. Speaking of Stripe,  I forgot, but I do need to provide the keys there  
267:02 as well. So, let me head over to Stripe, sign  into that sandbox. It's right over here. And then  
267:07 I'll give both the publishable key and then secret  key. So stripe publishable key, stripe secret key.
267:18 Then I'm just going to save this and say, hey,  I've also added the Stripe publishable key and  
267:22 Stripe secret key. I want you to connect the  payment feature as well. You may have to adjust  
267:29 the SQL migration script. So um do this first.  Okay. And the really cool thing is I can just  
267:36 queue up messages. So even if I change my mind  or do a bunch of things, I can have this message  
267:40 sent. I can then read that message and then just  um incorporate it in the flow. I don't actually  
267:44 have to wait until it it all finishes. So that's  pretty clean, huh? Pretty stoked about that.  
267:48 While we're waiting, why don't we test this on  another thumbnail? So I'm going to head back over  
267:52 to generate and then I'm going to paste this in.  Going to set that. I'll say replace the face in  
267:57 the provided thumbnail which is the face applied  to use cinematic and dramatic landing the face.  
268:00 Also replace 100 to 1,000. Let's see if maybe we  could get um a slight little variant of the actual  
268:08 number in the background as well. Generally  speaking, my experience with these AI models,  
268:12 you only really want to suggest one major change  at a time. So maybe what I could also do is add  
268:16 like a little tips feature or something like that  so that people know um how to prompt these things  
268:20 and get the best results. Nothing like clean work.  This one here looks real good. I like it. We have  
268:26 the 1000 in the background. Also dramatic lighting  on the face. That's cool. And this one has  
268:30 dramatic lighting on the face with my actual skin  texture. Thank you, Rosa. Um, honestly, I think  
268:35 all all three of these are really badass. So, I'm  going to keep this one. And as you can see, I'm  
268:39 going to have a toast message instead. So, I can  download this image. I can open in a new tab. Uh,  
268:44 that one doesn't seem to work, so I'm going to  have to modify that. And then I can also click  
268:47 generate another. It'll take me back to this page.  So, really, really clean functionality. I mean,  
268:50 this is basically exactly what I wanted. Just  wish it could be packaged in a slightly better UX,  
268:54 but we'll we'll solve that later. It's now going  through and just doing a bunch of Stripe stuff.  
268:57 So, creating API routes, resolving TypeScript  errors, and so on and so forth. I'll let it go  
269:01 and just do that thing. um as it wills and then  we'll double check how all that works with SQL  
269:06 and we're done with that. Okay. Now I'm going to  use the uh big thing inside of Superbase schema  
269:12 migration script the file that it created for  me. Then just hold command and press enter. I  
269:18 realize I used to think that you couldn't have  comments in here but uh you can. It's just the  
269:21 comments that the previous model did for me  were I think like with two hashtags instead of  
269:26 um two dashes. As long as there are two dashes,  you can basically pump this whole thing in and  
269:30 there are no problems. And it's always good  once you're done with that just to go back  
269:33 to database and then you can see how everything  interacts. So see the model here is set up ID,  
269:37 user ID, original image base 64, prompt, generated  URL, selected URL, and created at as well as ID,  
269:42 email, credits, Stripe, customer ID, subscription  ID, default prompt, face images B 64, and then  
269:48 created at. This is where we're basically going  to store their status. Okay? And then over here is  
269:54 where we're going to store the number of credits  that they have. So if they have a subscription ID,  
269:57 then they will have been signed up. Um,  obviously if they have credits, any credits,  
270:01 then we'll populate that. That's where we're going  to get this from. Um, and then if they're at zero  
270:05 credits, then when they attempt to make a thing,  it'll check their credit balance. And, you know,  
270:09 because they don't have any credits, that won't be  able to uh work. The primary key of thumbnails is  
270:13 ID. Primary key of users is ID. These two things  are sort of mapped together. And then, yeah,  
270:18 everything else here looks pretty solid to me.  So, just heading back to anti-gravity. Right now,  
270:21 what it's doing is just finishing up that little  migration from uh mock routes, which are when  
270:27 you store all of the database files and like, you  know, your your local computer to ultimately like  
270:32 live supabbase instances. And it's just doing a  build right now to verify that everything works  
270:37 uh the way that we told it to do it. And as you  can see, I've just signed into this supposed new  
270:42 account here and uh we are now at zero credits  available. So, it's pulling it directly from the  
270:47 database. Meaning if we want to run the full flow,  we now have to go through create a new account  
270:52 and then run basically the entire signup flow.  So I'm just going to go back to anti-gravity,  
270:58 verify that everything is good. Awesome. Live  database has now replaced our prototype API  
271:03 structure. I've now done it all. So we should be  able to test this. Just going to head back first  
271:09 to Superbase and then I'm going to turn off the  feature of authentication that forces us to um  
271:14 confirm our email. So you can do that in signin. I  just do that because it's a little bit faster. And  
271:18 I realized after that first build, I was spending  just as much time, you know, signing up as I was  
271:23 doing everything else. Now I'm going to click  sign up. And this is sort of the desired user  
271:26 flow. So we'll go [email protected]. Continue with  email. Let's see what happens. Just says signing  
271:31 in. There's no password. So this is a problem. Um,  notice how before we always had an email address  
271:35 and a password. Now we just have this like email.  So I think this is probably a blind spot where it  
271:40 didn't really realize, oh, okay, we're connected  to Superbase. Now we have O. So we're going to  
271:43 need an actual like login page. Hey, the login  right now only requests an email, but I need to  
271:49 authenticate with Superbase. I'm pretty sure we'd  need an email and a password. No, I just tried  
271:53 to create one and it didn't really work. There  were some errors. Okay, so now going to see if  
272:00 maybe I'm already logged in or something.  We go to payment. Yeah, no, we can't. So,  
272:05 I mean like the middleware routes are working as  we could tell with dashboard and stuff like that.  
272:08 Everything redirects us to login. It's just the  actual login doesn't seem to be um set up yet. So,  
272:13 we'll double click that. We'll double check that.  Um, it looks like we also added some additional  
272:18 landing page stuff here, which is super cool. So,  this added a whole new section based off what we  
272:22 added. AI power generation, AB testing, clean  modern aesthetics, and so on and so forth. We  
272:27 also have some FAQs here based off how it works.  So, that's really cool. And then we have the ready  
272:32 to blow up your channel. Um, Gemini will do this  from now and uh every now and then. I don't really  
272:38 sometimes it's sort of annoying when it acts  without necessarily asking you, but in this case  
272:42 it was what I wanted because really really early  on in the prompt I said, "Hey, I want you to make  
272:46 this longer." And then it said something like,  "Oh, I don't have enough context to make this  
272:49 longer." So I guess it just like generated its own  context based off of our conversation history. One  
272:53 thing I really like about this app is just how  fast it is. I mean, each of these routes are what  
272:58 are called preloaded or pre-filled. And so this  is me like really quickly going from one page to  
273:02 the other. And just notice how this whole page is  just like automatically done. Um, I also really  
273:06 like these brief little animations where you, you  know, start at the top and work your way down,  
273:10 loading each element slowly in succession. You  just get the, uh, visual appearance of very fast,  
273:15 very performant website. Okay, we have some  new superbase schema. So, I'm just going to  
273:19 jump over here and then what this is going to do  is go through and then add the rest of our users,  
273:25 tables, and stuff like that. So, back to SQL  editor, paste this in, run. Okay, the updated SQL  
273:31 script seems to have worked. So, I think we just  added one or two additional records here and then  
273:36 allowed it to add like the email and the password.  Heading back to Thumbly, if I go click sign up,  
273:42 we're running into Superbase URL is required.  So, they do have some runtime error. Now, I'm  
273:47 just going to paste all of this in. Okay, this new  signup page seems to work. We'll go nickniki.com.  
273:54 Then, Nickniki. Um, I'm going to click sign in.  Invalid login credentials. The reason why it's  
273:59 telling me this is because we obviously don't have  an account. So, I'm going to go back to sign up  
274:03 instead. I like that it keeps the email address  and the password. I don't have to redo it. Okay,  
274:07 we now have the secure checkout with the starter  plan with 100 thumbly credits. So, now what I  
274:11 can do Okay, because we're we clicked the starter  plan. So, I can go Oh, actually, did we click the  
274:17 starter plan? Hold on one sec. I don't necessarily  just want to buy the starter plan. I might want to  
274:22 buy pro. Let me try agency, actually. Okay, we'll  go nickniki.com. And do we have the account? Oh,  
274:30 you see because we exited out of the um payment  flow, it immediately jumped us into dashboard, but  
274:40 it selected the starter plan automatically. So, I  need to change that. Right now, it's automatically  
274:45 selecting the starter plan. You should allow the  user to select which plan they want on the payment  
274:50 page. It should also take into account which  button they click on the landing page because  
274:55 there are three different payment options there  as well. Just going to enter this in and then  
275:00 I'm going to proceed to check out just to see  what happens. When I click proceed to checkout,  
275:05 nothing occurs. And I think the reason why is  there might be some problem here. Additionally,  
275:11 when I click proceed to payment, nothing occurs.  Do we need to open up a Stripe checkout window or  
275:16 is the payment occurring locally on the platform?  Keep in mind, I'm using a Stripe sandbox, so we  
275:21 should be able to test the payment and have all  of the events and stuff like that run true. Hey,  
275:25 more pointedly, um, since I am now signed in, I  should be able to go to the dashboard even though  
275:29 I haven't done the payment. If I go to overview  and then click generate new and then actually go  
275:35 down to settings, can I upload my face? So, we go  supplied face. Okay. And I'm just going to refresh  
275:41 this. Am I still here? Okay. So, I'm not actually  keeping the reference faces in. And this could be  
275:46 because my account is not like set up yet because  the payment hasn't gone through. I'm not entirely  
275:50 sure. Let me just make sure everything else works  here. So, face that goes through. All right.  
275:54 If I click generate variance, what's going to  happen? Please provide an inspiration image and  
275:58 a prompt. So I have the inspiration image here.  The prompt is right over here. So we also have an  
276:04 erroneous error message where it says please  provide an inspiration image and a prompt. Hey,  
276:08 I provided an inspiration thumbnail and a  generation prompt. And I clicked generate  
276:12 variance, but I'm getting an error called please  provide an inspiration image and a prompt. I think  
276:16 this is occurring because I don't have a reference  face. However, that error is not representative of  
276:22 the real problem. If this are if this is the case,  I've also noticed that when I click upload face on  
276:27 reference face, then I click save configuration,  nothing happens. And then if I refresh,  
276:33 my reference face is gone. So make sure that  the save configuration button works. If it's  
276:38 not working because of some frontend error,  fix that. If it has to do with superbase,  
276:45 let me know if there's anything I need to do on my  end. And I'm just going to continue going through  
276:48 this and seeing if there are any other errors  that I could spot. Obviously, I'm going to sign  
276:52 out again because I'm on a local dev server. This  is all occurring in real time. So, any updates I  
276:56 make to this is going to be reflected. Um, but we  do need to make sure that if credits available are  
277:01 zero that we immediately jump back, right? So,  some other design decision. If a user has zero  
277:05 credits upon signin, we need to redirect them to a  payment page where they could select um, you know,  
277:10 how many credits to purchase and stuff like that.  That should occur every time a user is at zero.  
277:15 Now, if you're noticing, I'm not actually  just like doing waiting, checking the feature,  
277:21 and then proceeding to the next one. I'm now  at the point where I'm just dumping in a bunch  
277:24 of changes. And the reason why is because most of  these changes are relatively minor. Uh, you know,  
277:28 it's probably just some slight little database  misconfiguration or whatever. And so, I can  
277:31 actually proceed in this way. Um, have it working  on a change while I'm like modifying the rest  
277:36 of it. On the landing page under everything you  need to win the click, I don't like the shadows  
277:43 in general. I'd actually like you just to remove  shadows across the entire UX of the website. It  
277:47 they just seem a little bit off um given the fact  that most of the website does not have shadows and  
277:52 does not use them. Under the frequently asked  questions where it says how does thumbly insert  
277:56 my face? Can I use any images inspiration etc. The  title text is good. So how does Thumbly insert my  
278:04 face? Is written in like this nice black text,  but the subtext that says when you upload your  
278:09 reference photos in the dashboard settings, our  models learn your facial features. That's a little  
278:14 too dark. Can you lighten that just by 10% or so?  In this way, I'm sort of arbiting the time between  
278:19 the changes that are occurring here and then um  you know my testing. So, signing in, going back  
278:24 to the dashboard. Let's go to top up. Okay, cool.  And now we have the ability to select these three  
278:28 different plans, which is cool. Okay, so why don't  we pretend I'm doing pro. And then every time I  
278:32 click redirecting to Stripe, looks like there's  this problem which pops up. So, I'm actually going  
278:35 to copy this over then say every time I try making  a payment by clicking the proceed to checkout  
278:41 page, this error occurs. But I mean like you know  we could also parallelize this and do a bunch of  
278:45 them simultaneously. No real need. Um but yeah I  mean like you know we're getting closer for sure.  
278:50 One other change it just made is when I click on  starter you see it goes payment question mark plan  
278:55 equals starter. So I'm actually on starter but if  I go pro it actually opens up this page to pro.  
279:00 So these links do something now which is kind of  cool, right? I can actually click get started and  
279:04 then go to a specific payment page. I'm unsure  if this is because I'm logged in or logged out  
279:09 though. I'm going to click sign out. Then I'll  go back here and then see what happens. Okay,  
279:13 so login question mark plan equals pro. That  means it's actually going to keep and conserve  
279:18 this query parameter as I browse through the  rest of the website, which is pretty neat. Okay,  
279:21 now the redirection script has changed. It's  pretty clean. The way that this always works is  
279:25 you can just put in whatever um numbers you want  so long as your card details are 4242 repeating.  
279:30 So I'm just going to click pay. Please do not save  this card because it doesn't work or exist. And  
279:37 it looks like after the payment, we're being sent  back to the main page. So, I'm just going to say,  
279:41 "Hey, it looks like the Stripe Checkout payment  um went through, but then it redirected me to the  
279:47 slash payment route." Uh, if I go to dashboard, it  continuously pushes me back to that route despite  
279:54 the fact that I've already made the payment. So,  we just need to find a way to validate that the  
279:59 user's actually done the payment. That seems to  be the main block right now. Okay, it's telling  
280:02 me that it's finished this, which is cool with new  secure endpoints and session IDs and stuff like  
280:06 that. Okay, I'm just going to try a new account  here and see if maybe there's a problem with  
280:10 that. Maybe it just has to do with the fact that  I had an old account and it's redirecting me based  
280:15 off some old flag. Then I'll click processing  there. That payment looks good. No, it must be  
280:20 a more broad issue. So, check out Stripe here.  It looks like we are receiving the payments. So,  
280:26 that's going through. Um, for whatever reason,  we're not actually capable of connecting that back  
280:30 to a person. This is my fake card, [email protected].  I've made tons of payments. Uh, I'm just going to  
280:36 relay this information over to my agent. If agents  have a hard time dealing with stuff like this,  
280:41 it's best just to give them as much context as  humanly possible so they don't get stuck on one  
280:45 loop over and over and over again. Okay, looks  like I've finally made it into the dashboard. So,  
280:49 we finally solved that persistent error um to do  with payments. As I'm sure you guys can imagine,  
280:53 you know, adding payments is like adding  another whole u problem that you have to solve,  
280:57 which is why I wanted to push it off as long as  possible. But, uh, now that we have I'm seeing  
281:01 that the credit monitoring is good, settings  look good. But I'm going to try uploading  
281:05 faces. See if we could um diagnose any issues  there. The saving configuration button looks  
281:11 like it works if I refresh this now. Okay.  No. So, it's telling Oh, okay. Cool, cool,  
281:17 cool. I got worried there for a second, but I  guess we had a bit of a lag. Um, that's okay.  
281:21 Okay. Going to take a peek at an example. Why  don't we just go to thumbnail generated? Oh,  
281:29 no. That's uh that's me from before. Let's do this  one here. And then um it doesn't look like we're  
281:36 saving the prompt variants. That's one change. I'm  also noticing this is significantly laggier than  
281:41 it was before. Uh make the main subject highly  cinematic. Replace the face with the face of  
281:52 the supplied face of the reference image. Use  dramatic cinematic lighting on the face. Okay,  
282:01 let me try saving this configuration. Just seeing  if maybe it's an error or if maybe I didn't fully  
282:06 understand what this is doing. If I refresh this  now, do we have that default prompt down below or  
282:13 does that need to be dealt with as well? And why  is it so laggy? It may just be my computer. I am  
282:18 editing a bunch of videos. So, okay, that doesn't  look like that filled in, which is unfortunate.  
282:23 Um, why don't we do supplied here? Generate new  variants. Okay. Make sure the generation prompt  
282:31 auto prefills with the default prompt in settings.  Right now it isn't. It should pull directly from  
282:38 the default prompt instructions. It's also  a little laggy. The app takes significantly  
282:44 longer to load now than it used to. Diagnose  whether this is because of some persistent error  
282:50 behind the scenes or some issue with prefilling  or something else. Okay, taking a peek at this.  
282:56 It did give me a very sexy beard. Let's be real.  Nick definitely can't grow that level of facial  
283:01 hair. But now like the internet will be able to  tell the difference. Cue that one up. Thank you  
283:06 very much. After the save, it's going to pop up  this little thumbnail saved. Um let's download the  
283:16 download the image and just see how that looks.  See if there are any problems there at all. Nope.  
283:20 That looks like definitely fake Universe Nick,  which is what we wanted. I'll go back to generate  
283:25 another. Now I can do more. Okay. And it looks  like this is taking some time to load. I mean,  
283:31 it's saying zero. Okay. Now it's saying 97. I  think I know the problem. I think in order to  
283:34 fix whatever payment BS that it had to do, it's  now fetching data from the database on every page  
283:40 load, which is adding a fair amount of latency and  then causing the app to lag a lot more. So, it's  
283:45 obviously not what we want, but it's what's going  on. That's why every time I load settings, it  
283:49 takes a second before this comes in. Ideally, what  you want to happen in a full stack app like this,  
283:54 at least a modern full stack app, is when you sign  in, you want it to automatically cache and then  
283:58 like like pull all the data and then cache it.  Um, and then that way everything is pre-filled,  
284:03 all the images are prefilled, the credit amounts  prefilled, so on and so forth. And then instead,  
284:06 what you want is you basically only ever want that  to change if you materially change the database.  
284:11 So when you change a database, then it says,  hey, you know, you should probably grab the new  
284:15 data as well, just uh just to be 100% sure. And in  Nex, they manage this through a variety of means,  
284:20 but a big one is like mutating the database um  or or setting like mutations or or whatever as  
284:25 opposed to always fetching totally new data. So,  I think that's what's going on. I'm just going  
284:29 to instruct the agent, see if we could fix that,  and then assuming that there's no other changes,  
284:33 I'm just going to push this puppy live. Okay,  looks like it's loading significantly faster now,  
284:37 which is nice. We can jump around from settings  to generate to overview, and we don't have that  
284:41 like really persistent lag like we had before.  Um, we also have the ability to, you know,  
284:45 delete if we wanted to. I think what I'm going  to do is I'm just going to open this in, well,  
284:50 not an incognito window. I guess I'll just open  it in a new tab. And then what I want to do,  
284:54 whoa, I don't know what the hell just happened  here. No, thank you. Let's not do that. Thank  
284:58 you. Thank you. Um, I think what I'm going to do  is I'm just going to try the mobile features now  
285:03 on the dashboard because mobile optimization is  pretty big, right? Just want to make sure this is  
285:08 going to work. And it looks like it is. We have um  we have a little nav bar down at the bottom now,  
285:13 which is kind of cool. And we can jump back and  forth between these three. So, that's nice. I am  
285:18 a fan. Thank you. And now, let's just exit out  of the dashboard, go back to the landing page,  
285:23 and just see what we're looking at. So, looks  like that mobile optimization prompt worked in  
285:27 so far that, you know, the um little section here  is mobile optimized. And then all this stuff's  
285:34 mobile optimized. I'm just going to pretend sign  in now. Okay. Okay. And then yeah, we immediately  
285:39 jump to dashboard where we have functionality.  Let me just try deleting this. Yeah, I want to  
285:43 delete it. And then if I just refresh this puppy,  is it gone? No, it's not gone. So that's the final  
285:50 change we need to make. And then uh yeah, I'm  pushing this puppy live. In addition, I've also  
285:55 set this up. So it's now at thumbly-g.netfi.app.  I just said, "Hey, put this puppy on GitHub and  
286:00 Netlefi. Let's do this thing." And uh yeah, this  is actually working really, really good. So,  
286:05 I'm just going to go over to sign up here and I'm  going to go [email protected]. Would be a hell of  
286:10 a name if I could pull that up, though. And then  we'll go proceed to checkout and I'm just going to  
286:15 enter some info. Apparently, my card holder name  is 234234. Take that, Nicholas. This has jumped  
286:21 us directly over to dashboard and now I have the  ability to do everything. So, I'm going to upload  
286:26 some face. Uh, what was it again? Was it supplied?  Yes, it was. Keep the identical to the inspiration  
286:32 image. Replace the face. to face the reference  image. Use dramatic cinematic lighting on the  
286:35 face. Cool. I'm just going to save this. Says  that it saves successfully. Oh, hold on a second.  
286:41 That's not my face. That's my face. Good god,  Nick. You know, you've been developing an app for  
286:47 a long time when you start making trivial errors  like that, huh? Okay, now I'm going to generate  
286:53 three variants. Um, we're going to see how  long this takes. So, counting down. 1 2 3 4 22.  
287:01 Okay, so right at 22, we had an unexpected error  occurred. So I imagine this probably actually went  
287:06 all the way through. There was just some problem  with updating the DB. Uh this occurs as mentioned  
287:11 when you go from local to some sort of like  cloud setup or or you know actually like hosting  
287:16 it live. So I'm going to go through and just give  this an unexpected error occurred. Do a little bit  
287:21 more final debugging and see what we need to do  to make this puppy. I would imagine the issue is  
287:26 probably related to the fact that um when we were  running this locally, we didn't have a timeout,  
287:31 but services that like manage the calls  for you, like uh Netlefi, for instance,  
287:39 typically have timeouts of like 15 seconds or so,  maybe 20 seconds. So, I think we probably just  
287:43 ran into that. Okay, just ran out of usage again  after developing this for a good hour or so. So,  
287:47 I've moved over to Claude Code. Uh I'm still going  to try pushing and generating the variants here.  
287:52 We're going to see how that goes. 22 Mississippi,  23 Mississippi, 24 Mississippi, 25 Mississippi, 26  
287:59 Mississippi. Okay, so right at 26 seconds is when  we got the result. Um, I imagine that's probably  
288:04 why this one's kind of weird. I think that's the  first one that's like actually objectively failed.  
288:09 Uh, but yeah, no, I like this one. So, why don't  we just keep selected thumbnail? Awesome. Looks  
288:13 clean as per usual. And uh, yeah, I mean like  we'll just click generate another and we can just  
288:17 keep on going as many times as need be. But that's  it. We just launched our app and it works. And  
288:25 there's one final thing we need to do, which is  that god darn security audit. Almost forgot. So,  
288:29 let's do that. Okay, pasting this directly into  cloud code. As we can see here, now reading all  
288:34 the client pages and doing the config. And let's  see how we do. And it's just building to verify  
288:39 that everything compiles now. And then I'm going  to do a final check just make sure that, you know,  
288:44 the app functionality works the way I want it to.  You'll quickly realize that the vast majority of  
288:48 time that you spend on this is not actually the  building. It is just the giving of vibes plus the  
288:56 verifying it works to the way your vibes require  it working. So I made a bunch of these changes. Um  
289:03 okay, there's some rate limiting off middleware.  Okay, so I'm just going to say implement the  
289:10 middleware with check stripe web hook secret etc.  So I'll probably have to go and grab some stripe  
289:20 stuff, but I'm just going to go and verify that  this actually works now. Awesome. We still have  
289:23 the ability to do everything that we had before,  which is nice. Okay, I'm taking a look at this  
289:29 is so interesting. Like if you really zoom into  one of these things, you see every little pixel  
289:33 that this constructed. It's wild. Um anyway, I  mean like that dead inside completely. But yeah,  
289:40 uh we're now going to run a with check migration  on the database. So that's going to be cool. And  
289:45 then we'll also add some web hook secrets. So I'm  going to head over there and then just verify that  
289:49 um I got all this. Okay. Well, that one took a  little bit longer than I wanted it to because  
289:53 we did have a bunch of back and forth with  the payments on Stripe, but suffice to say,  
289:57 still good to go. Uh this is a good example now of  everything. It's not only, you know, kind of CRUD  
290:02 create, read, update, delete, like a traditional  app. Doesn't just do stuff that you ask it to. We  
290:07 also, um, are wrapping around a pretty popular  and powerful API, the Nano Banana API. And then  
290:13 we're applying some prompts and then some sort of  creative applications here to generate really cool  
290:18 outputs. Uh, and then we're wrapping it all in a  really nice UX and connecting it to payments. So,  
290:22 at this point, like you guys basically have  everything that you need in order to just build  
290:25 whatever full stack app you want. Uh, I think  what I'm going to do next is just do one final app  
290:29 combining everything that we talked about. Maybe  make the UX a little bit more complicated. So,  
290:33 instead of just having three or four, you know,  two or three things on the left hand side, we have  
290:37 like what might uh be considered more of like, you  know, like a really successful enterprise app. And  
290:42 then I'm just going to run through the motions  that I've already showed you guys multiple times  
290:45 at this point before talking a little bit about  the economics of vibe coding and then wrapping up.  
290:50 Man, I am so excited for this app because as  somebody that produces a tremendous amount of  
290:54 content nowadays, having an app that can basically  allow you to focus on one uh place, let's say  
290:59 YouTube or maybe your blog or something and then  have that same content, the same concepts and  
291:05 ideas immediately distributed to a bunch of other  platforms is fantastic. My issue with the way that  
291:10 most people do this is they just do it in a really  poor way. And so, you know, the end uh output  
291:15 quality of, let's say, a blog post converted to  like Xthread snippets or whatever ends up being  
291:20 so bad that, you know, the AI slop um overruns it  and then the quality is really poor. So, my idea  
291:25 today is I want to make an app that's so dang good  that, you know, you don't get that. And it's still  
291:30 written in your tone of voice. It's written really  succinctly. It's written really intelligently.  
291:34 And I was trying to come up with like a good  name for this app just before, you know, like  
291:38 clearly if you guys have paid attention so far,  the names so far have been afterthoughts, right?  
291:42 But um this time I think I already got one. It's  going to be called splinter. And the idea is okay,  
291:48 you're going to put in one piece of content which  why don't we just call this like pillar content.  
291:55 This is going to be like the pillar upon which  everything else rests. And maybe it'll just be I  
291:59 don't know like a blog post or something or maybe  like a video URL, right? And then what I want is  
292:04 I want this to be able to create content for a  platform like X. I wanted to be able to create  
292:09 content for a platform like LinkedIn. I wanted to  create content for Instagram. I wanted to create,  
292:16 I don't know, scripts, maybe like long form  scripts. And then I also want short form scripts,  
292:23 right? So we could do stuff. And then like  eventually, and I'm not going to get to this now,  
292:27 I don't think, cuz it's kind of a big ask, but um  we should also be able to create like video like  
292:32 full stack end video. Um that would be my that  would be my end goal. So, this is basically what  
292:38 I want the app to do. I want you to be able to  provide one piece of content and then immediately  
292:42 just like, you know, creates high-quality content  that's specific to each platform. And obviously,  
292:47 we're going to need to do this through some  sort of dashboard. In addition, okay, I don't  
292:51 just want this to be basic where we had like our  little dashboard on the left hand side with like,  
292:55 you know, three things and then we have some other  like central section um over here. I want this to  
293:00 be more interactive and I want this to have some  sort of like really detailed onboarding as well.  
293:05 Um, this is naturally going to involve like a  fair amount if you think about it. Not only are we  
293:10 going to have to do everything we did in the last  app, which was to uh summarize a landing page,  
293:15 Stripe payment integrations, um, OD DB, some sort  of API wrapping, and then like a good way to,  
293:21 you know, present that to the user. We're going  to have to go further because it's now not just  
293:24 one like API that we're wrapping around, but  it's realistically going to be multiple. And  
293:28 then we're also going to use our own custom logic  sort of um on-site logic to well not on-site sorry  
293:34 but like our own proprietary stuff to do things  like manage the tone of voice the prompts and so  
293:39 on and so forth. And perhaps one of the coolest  parts of this is I think the last app used AI as  
293:43 just like an like an afterthought to help generate  you know a list of filters in this AI is going to  
293:48 be like the main deal and uh hopefully we all  know like AI is the big thing right now. So,  
293:53 it allows us to sell both uh AI and our value  props, but also AI through the apps. And my my  
293:58 goal when building apps is I just want to  build stuff that like I actually would use  
294:01 uh if I wouldn't actually use it as somebody in  a space. Odds are I probably don't understand the  
294:06 problem enough to even be building this app in  the first place. But yeah, you know, if I could  
294:09 save myself a bunch of time with this, that's  going to be a solid win. And then if I could  
294:12 save other people time of that and stuff like  this as well, then that'd be even better. Okay,  
294:16 so the very first thing I'm going to do is I keep  turning back to this 16 or 17-year-old kid. No,  
294:21 he's 16 years old. His name's Leon. And what he's  been doing recently is he's been building like  
294:24 really cracked high-quality landing page designs  for Gemini. And so he comes up with these very  
294:29 very in-depth prompts. And then he ends up with  like a literal interactive app that you can like  
294:33 move your your fingers around. Unfortunately, you  didn't take a video of this, but this is like,  
294:37 you know, an app that there's like little strings  that are basically here being um processed every  
294:42 second through I don't know if it's like web web  GUI or something like that, but it's some sort of  
294:46 3D library. And um he shares his prompts online.  He's really cool. I used him in my last video.  
294:50 I wish I had half that uh brain at 16 years old.  And so what I'm doing just to start us off is I'm  
294:55 going to copy this prompt that he's put together.  Okay? And then I'm just going to ask Gemini to  
294:59 make me a landing page. And I just want it to  make me a landing page first before I do the rest  
295:03 of the app because I just don't trust that it'll  be able to do both the very in-depth programming  
295:07 required to build some sort of cool 3D generative  thing and then also do the app. So basically what  
295:12 I'm going to do is I'm going to have it like  generate the the landing page and then we'll  
295:14 build the rest of the app out afterwards. Um, so  you know, kind of non-traditional, but basically  
295:19 what I've done here is I've just pasted all this  stuff in. Okay, I'm going to press enter. And I'm  
295:23 not going to read this whole prompt to you, but  this is basically, you know, act as an elite,  
295:27 award-winning creative developer and digital  generative artist. Your task to write a single  
295:30 self-contained executable file that renders an  ultradetailed pixel perfect and breathtaking hero  
295:34 section for a next generation tech company.  Critical constraints, do all this fun stuff.  
295:38 Here's the theme and the content. You know,  provide some information about what I want and  
295:42 stuff. And stylistically once we have this set up  um you know I'll make some minor modifications to  
295:47 it so it's not a one to one clone or rip of this  uh and then we can actually move on with the app.  
295:52 Now while I'm doing that just to save on time I'm  going to voice dump everything that I think this  
295:57 app is going to include. Um and then I'm going to  send it over to claude code so that I can have two  
296:01 of these instances operating simultaneously.  One that's probably a little bit better at  
296:05 architecture design than the other that you know  is better at the front-end design that I'm asking  
296:09 for. So, uh, I'm not going to repeat it all over  to you guys, but a I want to design an application  
296:15 that takes a single piece of pillar content and  then immediately rewrites it, regenerates it,  
296:20 and recreates it, and then syndicates it across a  variety of platforms. Okay? And I just gave that  
296:24 over to Claude Code, and I asked it for a spec,  which is basically a highlevel description or  
296:29 overview of what this app would look like. And I  just want to run you guys through this here while  
296:33 Gemini is designing the actual generative part.  So this includes tech stack um app architecture  
296:39 some info on the database schema some info  about the landing page and marketing site.  
296:42 So after we're done with this, we'll use this  to modify it. Pricing and Stripe integration,  
296:46 authentication and signup flow, onboarding  flow, the actual content syndication engine,  
296:51 uh info about the dashboard, the content library,  AI configuration and voice model, scheduling and  
296:56 calendar, analytics and reporting, settings and  account management, API and webbook architecture,  
296:59 and then also some direct platform publishing  and then AI video generation stuff in the future,  
297:04 which basically we're just going to like add stubs  or little nuggets that we could use to build that  
297:08 out later if necessary. I've then given it some  information on the text stack and then asked it to  
297:13 come up with some other things. So for instance,  we're going to use Next.js15. We'll use Tailwind  
297:17 CSS plus the Shad CN/ UI component library. We'll  have Superbase Realtime for live updates. We'll  
297:23 we'll do a bunch of other stuff as well. ZOD  validation. This is important for security cuz  
297:27 I asked it to make sure that it was as secure as  possible on the first shot. What I'm going to do  
297:30 is I'm just going to modify a couple of the pages  here. I'm probably not going to use a calendar.  
297:34 Okay. Analytics look good. Um platform connections  for now. I'm just going to leave those out.  
297:38 Um, AI voice prompt configuration is good. So,  I like this. This is like your tone of voice.  
297:44 Just make sure that gets that in right. And then  settings, account, billing, and team. All that  
297:47 stuff looks good. Here's like the data objects.  I don't want to spend too much time on that,  
297:51 so I'm not going to worry about it. Um, that  looks solid. On the right hand side, it looks like  
297:55 this generative component has now been done. So,  that's pretty sexy. Now, it's going to go design  
298:00 the rest of the website. While we're looking  here on the lefth hand side from cloud codes,  
298:04 uh we have a bunch of different feature sections.  The landing page pricing and stripe integration  
298:08 is going to start at 19 a month or 190 a year. So  it's going to be 10 months basically giving two  
298:12 free months to people. Of note, just before we do  this, the input method is really clean. Um I asked  
298:16 it to be able to either paste raw content. So just  copy and paste a blog post, article, a transcript,  
298:21 or a series of notes, or paste a URL. And the  whole idea is, you know, if we paste a URL,  
298:27 um it's going to go through and then scrape the  whole website for us. So we didn't even have to  
298:30 touch any of that. Um, additionally, I want to  be able to do things like paste a YouTube video,  
298:33 although I think right now I'm probably going  to leave that out just uh as the demo. And then  
298:37 once I have those core functions and features  in, then I'm going to do more. But you should  
298:41 also be able to upload things like MP3s, waves,  M4As, OGs. Um, all these things should be auto  
298:46 transcribed. And the idea is it's just going to  be like your one-stop shop portal for like, hey,  
298:50 I just produced a piece of content. I'm going to  plug this in and then I'm going to have copy and  
298:53 pasteable ready content to distribute to all these  different platforms with minor adjustments. And  
298:57 obviously, I'll also be able to like change the  prompt anytime I want. Um, so very valuable for  
299:01 me. Hopefully it'll be as valuable uh for other  people as well. Okay, very satisfied with the  
299:06 first iteration of this landing page. As you see  here, there are some issues with this WebGL stuff,  
299:10 but these are real particles that are being  rendered um in real time. Um I don't like how this  
299:15 neurallex stuff is all over the place. Obviously,  that's that's okay. I've since told it to slow  
299:19 things down a little bit, but yeah, I mean like  the text on the left hand side of the page looks  
299:22 really really clean. Obviously, we have like nice  little mouseover buttons and and stuff like that.  
299:26 Um, I'm going to change all of this signap  stuff to uh Splinter, but for now, I mean,  
299:30 Gemini uh 3.1 and you know, prompts from Mr. Leon  Lynn over here are absolutely crushing it. So,  
299:36 pretty stoked and looking forward where and  looking forward to where the next uh step  
299:40 goes. After the next round of edits, we've also  added some additional sections. Common questions,  
299:44 ready to syndicate, and then finally down here,  um, a full- length footer. So, that's looking  
299:48 pretty sexy. The lines here are still moving  around a fair amount. So, I'm just going to modify  
299:51 them a little bit more. But, we've now implemented  some sort of like mouse um interactivity. Do you  
299:56 see how this thing is kind of like moving away  from my mouse as I move around the screen? So,  
300:00 I can technically control this. It's just  it's not super uh what's the term? Trivial  
300:04 or obvious to me as a user of the app how this  works. I've also had it change these lines. So,  
300:09 that's going to slow down pretty significantly.  And then we can see what the final product looks  
300:12 like. While I was waiting, I also had it go do  a little bit of research on other platforms. So,  
300:17 Repurpose.io lately, Constant Studio, Buffer,  and so on and so forth. And then why Splinter  
300:22 will solve it. So most are distribution only or  creation only. Splinter will do both. Basically,  
300:27 it'll allow you to both create content and then  distribute it quickly. Um voice and tone being  
300:32 an afterthought here. We're going to fix that  via um some pretty strong prompt engineering. I  
300:36 think most people that are designing these sorts  of repurposing apps are more software engineers  
300:39 than they are prompt engineers. I would consider  myself much more prompt engineer than a software  
300:43 engineer. So that should allow us to get like  very high quality tone. Okay, these particles  
300:47 are looking a lot better. They now move according  to my mouse, which is cool. And then these little  
300:52 um things are no longer facing the opposite  direction. I'd say they're probably a little  
300:56 too many of them. So, I think I'm going to remove  the entanglement ratio here just so we have three.  
301:00 And then I'll probably also just adjust the color  of the particles a bit. So, you know, it's I don't  
301:05 know, like kind of lightish yellow, maybe goldish,  something like that. Also see if maybe I can make  
301:10 the particles just a little bit bigger. Right  now, I feel like they're kind of compressed,  
301:13 which I'm not a fan of. still is kind of neat just  seeing this little streaming animation. Pretty  
301:17 interesting seeing the logic and kind of the math  that uh ended up aggregating all these things  
301:23 together. I mean, I'm just moving my mouse around  and there's clearly a deep relationship here,  
301:27 but I can't exactly figure it out. I think that's  because they're modeling this in 3D instead of 2D.  
301:32 Becomes a lot simpler when you see them in black  like this. There's definitely some 3D going on  
301:37 here as we could see. And honestly, I think I like  the black. Um I like the 3D translation. So, I'm  
301:42 just going to keep it at that. All right, awesome.  So, now that we're done with this, um, now I'm  
301:46 just going to modify the content on this page. And  I'm just going to do it by giving Gemini the rest  
301:50 of the instructions and saying go for it. Um, you  know, we're going to want to clean out some of the  
301:54 contacts and stuff. Uh, but I'll show you guys  what that looks like. So, I'm going to say to  
301:58 adjust the copy on the page to match the purpose  of the app, which you can find in this side of  
302:01 this spec. I'm not actually going to like manually  do any of this. And then it'll go through and then  
302:06 just adjust the rest of it. And yeah, I didn't  actually realize, but this is this is definitely  
302:10 like a portal that we are moving through in 3D  space. So, it's actually more complex than I  
302:15 thought it was. Looks clean to me. We now have  everything here set up. All of the syndication  
302:21 text is in now. Um, it's still saying synapse,  which is kind of annoying. I don't think it got to  
302:25 that. Cool. Good to go. So, let's build the rest  of the app now. Um, for us, it's going to be as  
302:29 simple as just sending all of this information  over to Gemini and saying, "Excellent work.  
302:35 I'd now like you to build the rest of the app  as detailed in splinter_spec.md. But there's one  
302:43 change. I'm going to say along the way document  each update in an updates.md file so that we have  
302:50 a history of tracked changes in case something  occurs or I need to feed this into another version  
302:59 of the model of Gemini. And the reason I'm doing  this is because this is a pretty crazy long app,  
303:06 right? There's a fair amount that's going on here.  Um, I anticipate I'm probably going to run out of  
303:09 tokens at some point because I ran out of tokens  about an hour and a half into the development of  
303:13 most apps so far. What I want is I basically just  want like a list of everything that it has done  
303:18 up until the time that it did not do the thing.  Um, and I'm just going to store it in a file that  
303:22 I can then give to either another version of  Gemini later on or clawed Opus 4.6 or whatever  
303:27 the freaking model I want. And in this way, this  is almost like a GitHub um repo before I get all  
303:32 the pushing logic and stuff in. But it gets to  take advantage of AI summarization features. So  
303:36 I'm not going to have to load the codebase every  single time. I want to like analyze and you know  
303:39 see what the problem is. This section cracked me  up a ton because I don't know if you guys have  
303:44 uh seen the draw the rest of the owl meme, but  basically it's like how to draw an owl. Draw some  
303:51 circles. Draw the rest of the freaking owl. Um  I feel like we've just done this so far and it's  
303:57 doing that. So, thank goodness for AI, huh? This  initial implementation plan is going to be heavily  
304:02 modified since we're no longer just building  a one kind of uh file uh website. Instead,  
304:08 we're going to be distributing some of the stuff  across big file structure. Here you can see that  
304:12 it's also set at the updates.m MD, which is quite  nice. So, we're going to see technical decisions,  
304:16 feature implementations, and significant changes  applied to the Splinter project. And my hope is by  
304:20 doing it this way and then later on feeding this  to um you know claude I can have both of these  
304:25 models both claude code uh sorry claude opus 4.6  and then gemini 3.1 pro high basically like debate  
304:31 about which features are better which things  to keep and then which things to remove. Agent  
304:35 teams in cloud code does a variation of this right  now where basically they have like a shared chat  
304:40 file and then they chat back and forth. Um, this  updates to me is just a slightly more streamlined  
304:44 version of doing that without me having to  waste $400 in tokens cuz boy does that agent  
304:48 teams feature consume a lot of them. All right,  so it's built the MVP implementation plan. Just  
304:53 taking a quick peek through this. This is going  to need everything here, which is fine. Hopefully,  
304:58 we are very used to getting all those API keys  now. And yeah, we're going to initialize Nex.js  
305:03 directly inside of this root directory. So,  it's going to do all of these steps here. Oh,  
305:07 hold on. I see that it's not actually taking that  into account. I'll say initialize in the same um  
305:15 folder as root hero.html will be our landing page.  Just want to make sure that's super clear. And in  
305:22 this way I get to pause the outputs. I'm not going  to waste a bunch of usage on you know building it  
305:27 out and so on and so forth. That's where taking  a peek especially initially at what um Gemini is  
305:33 doing can be pretty valuable. I always liken this  to sort of like a trajectory sort of thing. Let's  
305:39 say over here is um I don't know South America,  okay, on the left hand side and then over here  
305:45 is the westmost coast of Africa and hypothetically  you are a boat trying to make it from one side to  
305:51 the other. And so this is you, right? And then  your goal is you want to go over here. Well,  
305:55 logically speaking as you are leaving the port on  the eastmost side of South America. Okay. Um there  
306:01 are a lot of different directions you could go.  You go this way, you go that way, you go this way.  
306:06 only a very small subset of these directions will  actually get you to your goal. Obviously, you're  
306:10 going to need to go like here, right? And even  then, I'm a little bit off. And so, initially,  
306:15 I guess what I'm trying to say is the direction  that you pick even in like the first kilometer  
306:20 as a boat heading to to to sail is really really  important for determining the final placement of  
306:25 the app. Um, even if I was like one tiny little  degree off, that could be the difference between  
306:30 me making it directly on target and then I don't  know like a one degree difference or something  
306:34 could be the difference between me making it to  city that's like 100 km away. And so I usually  
306:41 like to be around here to steer the model and  then once it's sort of like in this territory  
306:46 or whatever, you know, if it's like on track well  enough that I think it's like going to be okay,  
306:50 then I don't really worry too much about that. Now  on the right hand side here you see that the error  
306:54 um the agent actually ran into an error. This is  another reason why I like being around. This can  
306:59 occur for a number of reasons. This could occur  for token issues. It could occur because I don't  
307:03 know like the API is just out. It could occur for  rate limiting reasons or just because sometimes  
307:08 um some of these packages don't actually interact  nicely with Gemini and then it kind of struggles  
307:11 with the terminal. But yeah, I'm just going  to be around for the first couple minutes for  
307:14 a build that this is this comprehensive because  I don't want it to like make a wrong assumption  
307:18 and then build something completely different to  what I want and then later on I'm just going to  
307:22 have to go back and and redo it all anyway. Okay,  you got the app right over here. So, it's looking  
307:26 pretty sexy. If I click get started, we now can  sign in or sign up. So, I'm just going to pretend  
307:32 that I've done all of that. Now, we have the  onboarding. So, initialize workspace. Cool. So,  
307:40 this is like all the platforms that I'd want to  connect to. It's good. This is really interesting.  
307:45 Formality index, humor coefficient. I like that,  but I think we're going to have to add some to it.  
307:50 So, I'm just going to go back to the model and  then I'll say change voice profile to a default  
307:58 um to an advanced advanced setting advanced let's  say button where I can modify the prompt directly.  
308:09 That looks pretty cool. Dry run simulation.  Hey, what's going on? Execute run. This is  
308:17 going to be like when it actually does its  thing. Cool. Also add a um an additional  
308:26 step right after the dry run that shows the  output. I'll wire you up to Claude Opus 4.6.
308:41 So we can test this next. Okay, we're now going  to enter the dashboard. Okay, so this looks like  
308:46 a pretty simple dashboard still. I do like  the style we have. We're going to build out,  
308:51 you know, a generative interface. It says  over here. I'm probably going to want to  
308:54 move this stuff a little bit over to the left.  I'm not seeing anything under generative engine,  
308:58 voice profiles, platforms, or settings. So I  guess right now we have the onboarding all the  
309:02 way up to the dashboard. So that's pretty cool.  Still, if I click log out, I go back here. Then  
309:06 if I go back here then uh Okay, cool. So yeah, I  think I'm understanding where we're at with this.  
309:11 Uh it's looking pretty clean and I'm glad that I  managed to do all that stuff with just a little  
309:15 bit of prompting. We still have that cool little  3D thing, too, which is really nice. Okay, so next  
309:20 up, I'm going to feed in these instructions and  then I think what I'm going to do is I'm going  
309:24 to have it continue the initialization.  Um this is going to be tough. I mean,  
309:30 I really like this. I really, really like this,  right? But this is going to be pretty tough. Uh,  
309:36 I guess we could probably autogenerate a  prompt that includes some settings with this
309:42 for the sliders. We'll use this in a prompt  to fill variables like formality and humor.
309:56 And then this dry run simulation. We'll  execute the input. We should get a new  
310:01 one here. Systems nominals.  Cool. Awesome. Okay, cool.  
310:05 Let's also start building out the dashboard.  I'd like you to replicate the attached image  
310:15 in terms of content. Now, I'm going to go to  dribble over here. I'm going to go dashboard
310:24 and let's see what we can do here. I want  something that's just like much higherend,  
310:28 cleaner. I do like the squareer look. Um, to me  that just implies like a high-end brand. So, I'm  
310:33 going to look for something that does have that  square look. Let's try filtering by new instead of  
310:38 um popular. That way, maybe we can get things  that other people haven't built before. Okay,  
310:44 this looks pretty clean. I like this.  Let's see if we could feed this in. Oh,  
310:49 sorry. We need to copy this.  We can't just do that. Okay.  
310:52 uh in terms of content and style except I  want you to stick to our pre-established
311:02 um square no border radius. That's the little  um smooth corners. I don't really like that.  
311:11 I also want you to add significantly more  functionality as per the initial doc initial  
311:19 splinter spec.md. And this is looking pretty  clean. There's an advanced settings feature  
311:26 here now where we can actually change things which  is quite nice. I'll pretend I'm executing a run  
311:32 running style transfer. Okay, great. Awesome. So,  this looks kind of silly. Not a fan of the design  
311:39 for sure, but um you know, it's cool that it has  a lot of the stuff laid out. Interesting. Yeah,  
311:45 this is like highly opinionated stuff clearly. Um  all right, so I think what we're probably going to  
311:49 want to do is on the left hand side, we're still  going to want to have like a like a database,  
311:57 sorry, database dashboard panel. Um we're not  going to want to mention the model. I do like  
312:02 the system log. That part's cool. All right, cool.  Let me just give this some stream of consciousness  
312:06 feedback. Hey, this looks cool, but it's quite  opinionated. I don't like the dark borders. I'd  
312:11 like a higherend sort of Lux design internally  that still sticks to um very similar to the design  
312:18 on the landing page. I'd like you to remove all  mentions of the specific models that we're using.  
312:23 I want some sort of lefth hand side pane that I  could use to select different views. I really,  
312:28 really like the system log. I like the  credit usage, the generation, source items,  
312:34 active platforms. These are fine. replace recent  archives with a graph. That graph should go really  
312:39 in-depth on the number of generations and which  platforms we've connected and stuff like that.  
312:43 We also need the rest of the app. So, I need a way  to see specific and individual generations. I also  
312:50 need a way to create a new syndication. Right now,  when I create when I click create new syndication,  
312:54 it takes me back to the wizard. But I want this  to take me to a page where I can insert content  
313:00 either using URL or MP3 file upload or video  upload or paste in a bunch of text and then it'll  
313:09 do the syndication for me. Remember, we're just  dry running all this stuff for now. Uh I'm going  
313:13 to include some API credentials for anthropic in  a second. So you could hook up the cloud opus 4.6  
313:19 six integration and change the rest of the  language on the actual app side to reflect  
313:24 like normal app usage. For instance, I don't like  the term disconnect. I don't like the matrix. Um  
313:30 I don't like, you know, some of the other  language that you're using here. It seems a  
313:34 little too tryhardy. Okay, after about 40 minutes  of development, we did run out of tokens. But  
313:38 never fear, that's one of the reasons why we're  using two models and not just one. So what I've  
313:43 done is I've exported our conversation history  and then that updates file. And now I'm just  
313:47 uh doing next steps which in my case involves  feeding in the enthropic API key. So I'm going to  
313:51 go and find the anthropic API key here. I'm going  to call this uh for splinter. Then I'm going to  
313:59 copy it over right there. Okay. And then what I'm  going to do is I'll say let's wire up Claude Opus  
314:08 4.6. Test a bunch of prompts. Give me a few uh a  few dry run examples. And then I'll pick the ones  
314:17 I like the most. Let's say test 10 prompts.  Give me a few dryr run examples of befores,  
314:24 afters, and I'll pick the ones I like most  and we can integrate them. In general,  
314:28 the design quality of Opus 4.6 will be a little  bit lower. So, just keep that in mind. However,  
314:33 um there are some other trade-offs. You know,  it's a little bit faster and then addition to  
314:36 being faster, I think architecturally, it tends  to make like a better functioning application. So,  
314:40 that's why I frontloaded the design with Gemini.  And then after the design and stuff like that was  
314:45 all wired up, I did all like the little scripting  or I'm doing all of the little scripting and  
314:48 backend passes with um you know cloud opus 4.6.  This database is now set up. So I can head over  
314:56 to let's do two things. Under authentication I  just want to go to sign in signin providers and  
315:02 then I'll remove that little confirm email step  and then under project settings I'm going to go  
315:06 to API keys just so that this has all of the keys  that it needs in order to you know do whatever it  
315:11 wants. So I already gave it the publishable key.  We'll go um superbase secret API key and then I'll  
315:18 paste that in as well. And the idea is now it has  everything that it needs in order to manage the  
315:22 Superbase instance for me. I don't really need to  be involved in this process at all. And I'm doing  
315:26 this while Claude is basically just running  a quick little test on the prompts. I'm just  
315:30 going to pick the best prompt for now and then  I'll worry about like optimizing the hell out of  
315:33 it later. Okay. And whereas anti-gravity was using  its own built-in browser, uh Claude Code typically  
315:38 uses Chrome DevTools. So that's what it's doing  right now. jumping on taking a screenshot and then  
315:42 it's seeing an internal server error. So whatever  it did either didn't compile or there was just  
315:46 some problem on the back end. Um so that's what  it's fixing up now. As you can see the syndication  
315:51 page here for create is much lower quality. Um  and as mentioned you know cloud is just nowhere  
315:56 near as good as Gemini at least as of the time of  this recording and designing those really high-end  
316:00 uh looks unless it's like per perfectly basically  trying to duplicate functionality which is  
316:05 unfortunate. Um, it is going to make some minor  changes here because it it can't actually click  
316:09 this generate button. So, it's cool that it  can kind of run that testing loop on its own,  
316:13 but all of this is just in contrast with Gemini's  high quality design. And it looks like it created  
316:18 a LinkedIn post and then also an X and Twitter  post. Just taking a peek here. It took us 3 years  
316:23 to hit 1 mill AR with our SAS product. Here's what  actually moved the needle and what didn't. So,  
316:26 this is a um phrase that I explicitly said that  I didn't want in the prompt. So, obviously,  
316:31 I'm going to have to do some adjustment there. We  sell to dental practices only dental practices.  
316:35 The temptation to go horizontal was constant,  but saying narrow am meant we could out build and  
316:38 outmarket anyone in the space. We charge more than  felt comfortable from day one. Honestly, the price  
316:41 anxiety never fully goes away, but underpricing  would have killed us. Okay, just while we have  
316:45 this, I'm going to go back to the actual page.  Make sure that nothing else was changed here.  
316:50 Cool. So, the design here is quite nice, right?  Hopefully, we can we can see that. The question  
316:54 is where do we go from here? Get started. Sign up.  What's the wizard looking like? Okay. Okay. And  
317:04 then we're going to have our prompt here. Try run  simulation. Sure. Oh, is this now actually using  
317:11 real cloud code? No, it's not. Okay, cool. Uh,  sorry. Um, Opus 4.6. We'll sign off. Enter the  
317:17 dashboard. So, the dashboard here is looking a lot  simpler. Um, you know, we we're missing a fair few  
317:24 of those cool design features that we had before.  And then we also have little rounded corners which  
317:30 I don't like. So, I'm just going to dump a bunch  of instruction into Claude and see if it can solve  
317:34 this. Hey, I want the design of the dashboard to  reflect the design of the landing page. Right now,  
317:42 there's a very big discord between the two.  Notably, the corners are rounded in the dashboard  
317:47 and not rounded everywhere else. There's also  color discords. We're using this soft purple,  
317:52 which uh is correlated with which is typically  associated rather with LLMs. I don't want pill  
317:57 style things anywhere here. And in general, I want  you to use the same fonts in the landing page for  
318:03 the new syndication or create page. I like how  we've laid this out so far with the URL, file,  
318:08 or paste text, but rather than have three buttons  that the user has to click, I just want them to be  
318:13 able to do it um do all of these simultaneously.  So, just have one field inside of that field  
318:18 allow users to either paste text, import a URL, or  upload the file. It'll eliminate some complexity,  
318:24 keep things really streamlined and simple. also  allow them to select which platforms they'd like  
318:28 to syndicate content to at the bottom. Right now,  the generating for button, generating for section  
318:33 rather, doesn't allow us to um make any of those  changes. Finally, make sure that the app looks  
318:38 similar to the landing page. As mentioned, it's  very important that we get this one right. Oh,  
318:41 this is much better. Okay, as we see here, we  now have that perfect design unification. I  
318:45 like the new syndication button up top. Um content  library, votes, profiles, analytics, config, and  
318:50 API access don't fully work yet, but this looks a  lot cleaner. So, we paste a URL. or drop a file or  
318:56 type our content here. If I go into my finder, if  I go into my finder and drop it, does that work?  
319:02 Oh my goodness, it does. That's really clean.  Likewise, if I paste a URL, so I don't know.  
319:08 Let's do this one here. Okay, for my big claw code  course earlier and paste that in. Does that work?  
319:14 URL detected. Really clean. Okay, we also have  blog x, Twitter. Okay. Okay, this is much much  
319:19 nicer. Thank goodness. Um, obviously this is  going to be difficult to see the differences  
319:24 between these lines. So, looks fantastic. Let's  add mild differences between the lines on the  
319:31 generation throughput graph. I also want to build  out pages for content library, voice profiles,  
319:37 analytics, setup, and API access. So, work all of  those out. Add uh whatever relevant features make  
319:43 sense. And if there are any additional features  that you think would make sense on the lefth hand  
319:47 side panel, please add them. I'm uh attempting to  increase the at least perceived sophistication of  
319:53 the app right now. Okay, I actually just stepped  away to grab myself some breakfast and came back  
319:57 and it looks like Claude is still monkeying away  at this. It's been probably 30 or so minutes.  
320:03 I've had very little actual hands-on involvement  and uh yeah, you know, like the agentic workflow  
320:09 time horizon of this specific model tends to be  quite large. Um, I'll say Gemini actually has  
320:14 a slightly longer one, but when you prompt  it right, where essentially you say, "Hey,  
320:17 here's a big road map of things. Here are a bunch  of changes that we've already made. Here's the  
320:22 uh whole spec of the thing that we ultimately want  to build." And then you just let it run loose. It  
320:27 can operate quite independently for quite a  long time. Um, and so this has gone through I  
320:31 think three or four context windows at this point  considering it's about 200,000 or so tokens. You  
320:35 know, I've spent a few million tokens um using the  cloud or anthropic API. That's multiplied by fast  
320:41 mode here. This sort of approach isn't the best  for everybody because of costs, but in my case,  
320:45 I'm happy to trade off that cost for um time  as previously and will continue to do. Checking  
320:51 back in on the app, it's looking significantly  more fleshed out. We have a content library up  
320:55 in here with the actual posts themselves. We can  go grab these posts and do something we wanted  
320:59 to. There are voice profiles now. Um there is a  brand voice, casual, blog. You can adjust that  
321:06 using this little section down below. You can also  add writing samples for best results which will  
321:11 improve the probability that an AI model actually  outputs what it is that you want. Generation  
321:15 volume right over here along with tokens used,  average numbers per session, platform breakdown,  
321:19 everything that I basically like talked about  in the spec. Finally, automation rules. So,  
321:24 um this is like a workflow thing where you can  automatically syndicate new content. Basically,  
321:27 every time you add new content, it'll  automatically go through a big flow. We have  
321:31 like enabled and and disabled and stuff like that.  Right now, these are all just stored on the local  
321:35 database. um it's left to be seen essentially  whether or not I could scale that up. We have  
321:40 profile platforms. This is where I can connect  and disconnect and so on and so forth. And ideally  
321:45 this would open up a little like connection  panel. And then uh you know we have notifications  
321:49 supposedly throughout email with the ability to  export all data. And this is really the section  
321:54 that I'm I'm the most excited about because uh you  know if you make your apps like agent compatible  
322:00 via API um I think that's going to be a massive  market later on. So, what I've done is I've added  
322:04 an API access section with like supposed  production keys and stuff. This isn't fully  
322:08 fleshed out, uh, as I'm sure you can imagine, but  still pretty reasonable. Um, what I'd like to do  
322:12 next is basically want to verify that all this  functionality works and and actually like does  
322:17 what it is that they imply that it does. And so  for this, I'm probably going to use a distributed  
322:23 um cloud code either agent team or I'm just going  to ask it to run through one by one by one. The  
322:27 idea would be, you know, spin up a se separate  instance to test the API key while you're spinning  
322:31 up a separate instance to test the notifications  while you're spinning up a separate instance to  
322:35 test the automation rules. We'll see how all  that stuff goes. Hey, this looks fantastic.  
322:38 I don't think there's anything else here that I  really want on the app. Um, maybe on the dashboard  
322:43 there is a little bit of white space at the bottom  because the system log is just just a little too  
322:48 long compared to the connected destinations.  I think we could probably fix that by adding  
322:52 one additional row of connected destinations.  That'll um match everything up quite nicely.  
322:56 Everything else looks really great. My next desire  is to test this end to end every specific feature.  
323:02 Just going through a really simple setup here  for LinkedIn O because I want to verify that  
323:07 this actually works. So, uh, we're going to add a  couple products to the app. So, share on LinkedIn,  
323:12 we have to do this. So, request access. We also  need, um, sign in with LinkedIn using open ID  
323:18 connect. So, I'm going to request access to that  as well. And the good news is, you know, you don't  
323:21 actually have to know how to do this yourself.  This is just stuff that uh in this case Claude,  
323:25 but Gemini as well can just walk you through.  Then under OOTH 2.0 settings, we need to add this  
323:30 redirect URL. So I'm going to go up to O. Just  going to grab the authorized redirect URLs. And  
323:36 then I'm going to paste this in. And then I'm also  going to copy both the secret um into this ENV  
323:43 file. And then I'm going to uh okay, added restart  dev server etc. And then when this is done,  
323:51 I need to go to setup platforms. Connect LinkedIn.  Okay, cool. So, I'm now going to check out what's  
323:58 going on over here. We'll mock. We go to setup  platforms and then connect LinkedIn. This is going  
324:06 to open up a little LinkedIn dialogue. I'm then  going to sign in using my own credentials. Okay,  
324:11 now I have a phone number that I have to verify.  So, I did just get this. I'll type that in.
324:22 Now I'm going to sign in as myself. Cool. And it  looks like LinkedIn status is now connected. I'm  
324:28 going to head back over here and say connected  just fine. So we should be fine with this. Um  
324:35 find some way to store all of that info so we  don't need to reconnect next time. All right.  
324:43 So, we just did this um you know with with  LinkedIn. So, I should be able to go to new  
324:49 syndication now and then I don't know. Let me just  like find an actual post that I made before. Maybe  
324:53 my my blog. It's a good source for this stuff.  What's a fast one right over here? Actually, let's  
325:00 test this functionality. We'll click generate. I  don't know if it's actually feeding in this. So,  
325:07 that'll be the tough thing, the thing to double  check. We are on a generating tab. It's a fair  
325:12 amount of content. So, you know, I'm not expecting  it to finish basically immediately. Oh, also we're  
325:16 syndicating to LinkedIn supposedly. We should have  like a final check before we actually publish it.  
325:22 Okay. Okay, cool. So, it's double checked to  basically see whether or not we're connected.  
325:26 And so, X doesn't have publish, but LinkedIn does.  Most business small business owners are terrible  
325:30 at hiring. Honestly, it's not because they're bad  at business because nobody teaches you how to do  
325:33 it properly. Biggest mistake, hiring based on gut  feeling instead of actual data. You meet someone,  
325:36 they seem sharp. Cool, cool, cool. All right. Uh,  we still have some issues here. So like I don't  
325:41 know hashtags. Nobody uses hashtags anymore,  buddy. Remove that. Explicitly add it to the  
325:48 prompt. Uh we should never generate any sort of  hashtags. Also, I'd recommend against emojis in  
325:53 general. And then let me see what are we doing  that is sort of unfortunate. Most small business  
325:59 owners are terrible at hiring. And honestly, it's  not because they're bad, but it's the biggest  
326:01 mistake. Okay, so we have to avoid um LLM isms.  Like essentially the main LLM writing pattern  
326:10 that pushes it into uncanny valley territory is  when you jump between extremely short sentence  
326:16 lengths and then long sentence lengths and then  do that over and over and over again. That's not  
326:20 how most human beings will write unless they're  trying to be direct response copyriters. And for  
326:25 the most part we are we are not. So find a way to  adjust the prompt. Cool. And then everything else  
326:29 looks pretty good. So I should be able to actually  like push this. Um, thing is I don't really want  
326:34 to because then I'm going to have to publish this  on my own LinkedIn, right? Ah, I guess we're going  
326:38 to have to test it. All right, why don't we go to  another page here and then just verify whether or  
326:45 not this is being published to LinkedIn around  personal profile. If it is my personal profile,  
326:49 I'm just going to delete it basically immediately  because I don't want all my all my fans to see  
326:54 that. How do I do this again? Good god, man. Okay,  here we go. And we should be able to right click,  
327:01 press delete post, right? Okay, cool. Okay, just  ran into one issue right before I did this. So,  
327:06 I'm just going to Oops, that's not what I wanted  to paste. Um, I wanted to paste this. There  
327:11 you go. So, we ran into a problem just as I was  about to do this. I I don't know if it like out  
327:17 um just ran out and sort of timed out or  something, but okay. There we go. There we  
327:20 go. That's better. Okay. And now we should be able  to go to dashboard. I would hope anyway. Okay. It  
327:27 is going to force me to do this signin flow. We  do have some more platforms here as you could  
327:32 see. Fix the onboarding wizard as well. It should  now um function off of live data. Right now it's  
327:36 just doing the basic pass with templated stuff.  Okay. Setup. All right. All right. Here we go.  
327:43 Content library. Go back to new syndication here.  Go drive blog. Why don't we get something maybe a  
327:48 little bit shorter like this. Paste that in. Click  generate. We'll exit out of X Twitter so that we  
327:54 only have the syndication option for my LinkedIn.  Then we will go to my LinkedIn and then see if we  
327:59 could actually publish this. This is a blog post  I wrote basically all about interacting with my  
328:04 YouTube comments. I said uh you know the statues  that we are building aka our YouTube comments last  
328:11 way longer than the statues that people throughout  history have built because we don't know if people  
328:16 are going to be looking at these comments in  the year 2148 and the year 20,048 or what.  
328:21 So there's a lot of potential life there. Okay.  Clicked publish. Refreshing. Not seeing a post,  
328:27 which is good. I just clicked publish.  I'm not seeing a post on the LinkedIn. Uh,  
328:34 let me just on the generated content. Really helps  to be hyper specific. Did this copy work? Yeah,  
328:45 the copy worked. Okay. So, just want to double  check. Going to refresh this. And then I'm also  
328:51 going to head over to left click and see maybe if  we published it to there instead. I should be able  
328:56 to pull this up just by going back to the LinkedIn  page. Four left clicks. So, I can click on that.  
329:01 Go back to the company page and then we'll go  page posts. Okay. I'm not seeing a page post here,  
329:06 which is nice. Oh, wow. 675 followers. Did not  have any idea 675 followers in that company page.  
329:12 That's funny. Looks like there was a specific  issue with how we pushed this, which is okay. So,  
329:16 we're going to use actual API generate calls now.  And then I have the post right over here. Just  
329:20 going to head right over here. Click publish. Not  seeing any sort of notification. That's an issue.  
329:25 So, I'll head back to Cloud Code. I'm not seeing  any sort of notification after I click publish.  
329:29 Honestly, the experience doesn't feel very  snappy to me. I think the cursor might also be  
329:34 uh reason why. Find a way to just improve the  perceived customer experience when they click  
329:39 that button. Whatever it is, toast, menus, and  whatnot. We'll go back here. I'm just going to  
329:45 quadruple check. Make sure I didn't. Did I post?  No, I haven't posted. And then I'll do the same  
329:50 thing for leftclick. Let's go leftclick LinkedIn.  Just double check that that didn't post to either.  
329:57 It might have. Just unsure if I'm posting to  company pages right now or my own. Okay, doesn't  
330:02 look like I'm posting to either. One thing I'm  seeing here is we actually have been storing these  
330:06 in the database now, which is cool. Um, our local  database to be clear, but still a database. And  
330:10 so, you know, we can copy this. We could post this  to LinkedIn and whatnot if we wanted to manually.  
330:15 Um, but I do want to check to see whether or not  we could post u specifically like uh, you know,  
330:21 using that flow that we we showed you earlier.  So, I'll go back to new syndication. We're just  
330:24 going to do one more. Okay. Yep, we got it. I  got to delete the hell out of this thing ASAP,  
330:29 please. Thank you. And uh yeah, we've just  verified that this actually does post. So,  
330:33 fantastic work. That's incredible. Um, super  cool. Super cool. I'm going to go back here. Hey,  
330:38 I tried publishing and the first time I clicked  it, nothing happened. I waited five more seconds  
330:43 and then I clicked it again and that time it  went through the whole publishing loop and  
330:46 it actually appeared on my LinkedIn. Diagnose why  this might be the case. Go ahead and fix. Another  
330:51 cool feature is in the top rightand corner we  have a new. So I can actually just click new,  
330:54 go back here and then do another one. Um and  then I can also store all of that as you guys  
330:59 saw before inside of the content library along  with the specific generation for that um rung. So  
331:05 that to me is really really neat. Also the fact  that we have sort of the ability to go listical  
331:09 view is cool. You can tell that this really ran  with the spec that I gave it and it's um it's done  
331:14 a good job. I don't know. Okay, cool. Yeah, even  the search works. So very high quality stuff here.  
331:19 I'm going to head over to voice profile and see if  I could maybe set this as the default casual blog,  
331:24 do some editing and just double check what's  going on. So, as you can see, we can select  
331:28 these here. Um, I don't know. Is there any way  to like go deeper into what the actual prompt is,  
331:32 though? I'm not seeing that. I'm just seeing  like the ability to calibrate formality, humor,  
331:36 and emoji usage. So, I think I'm going to have  to add that feature. Hey, under voice profiles,  
331:40 I'm noticing that we can calibrate formality,  humor, and emoji usage, but I don't see any way to  
331:45 adjust the actual prompt itself. What I want you  to do is to add just like we have in the wizard  
331:50 some sort of advanced drop down that I can click  on that allow people to actually modify the uh the  
331:55 underlying text as well. Under analytics, I want  you to make sure that we are actually reflecting  
331:59 real analytics. Uh right now I'm not sure if we're  pulling from our database or what's going on. I do  
332:05 see that the recent activity here is updated with  the LinkedIn post, which is pretty solid, but I  
332:09 want to make sure that everything else is good,  too. Finally, under scheduling, um verify and  
332:13 test that this works for LinkedIn. It looks like  we have a scheduled post coming up. That said, I  
332:17 don't want you to actually publish it on LinkedIn.  Um, I just want you to verify that everything  
332:21 works up until that point. I've now been given  this initial schema. What I'm going to do is I'll  
332:25 head back to Superbase, go to the SQL editor, and  paste this in. Just verifying that these comments  
332:29 work. Okay, it looks like they do. I'm now going  to hold command and press enter. We're going to  
332:33 run this. And we see success. No rows return. I'm  going to go back to Cloud Code. And it looks like  
332:38 my Gemini subscription is finally coming back. The  rate limiting said it would be 1228. So, we have  
332:45 just another 10 minutes here before I can get  back to that. Um, which I think I'm going to do  
332:48 for like a final design. But that should allow us  to push through probably most of like the database  
332:53 migrations and stuff. Superbase SQL added. You  know, just to be clear, I don't like combining  
332:59 multiple um different models like while I'm in  the middle of a development session because I  
333:03 like all of the context that Cloud Code has to be  used in the cloud code thread and all the context  
333:07 of Gemini has to be used in the Gemini thread.  generally speaking, you get much higher quality  
333:11 results because uh this thing has processed so  many tokens at this point. You know, has so many  
333:15 so much history of like past design decisions and  stuff like that that it just inherently gets what  
333:19 it is that I'm trying to say. If I were to try and  just have Gemini, hey, pick up where whatever left  
333:23 off, you would basically have to do the entire  job of retraversing all of those failed paths. So,  
333:28 we're going to push this through with cloud code  and then at the end I'll probably do like a little  
333:31 design or redesign. Um maybe do some finishing  touches, some some final polishing with Gemini.  
333:36 Okay, so I'm going to say let's go 100% superbase  and we should be good. And it just got back to  
333:40 me saying that the superbase migration is now  complete. So it's also given me a summary with a  
333:45 bunch of API route os. It's verified the build and  whatnot. Verifying the build to me is not enough.  
333:50 I need to actually go through the whole testing  um end to end. So I'm going to have it log in,  
333:54 put in some credentials, and actually like do said  testing on its own because it has the ability to  
333:58 screenshot things and whatnot. Then once all this  stuff is done, it's just a matter of pushing that  
334:03 to um you know actually like a like a like  a Netlefi host and then also adding payments  
334:07 and stuff like that. The reason why I'm not adding  payments right now is uh I sort of had a change of  
334:11 heart. I thought a little bit about um what would  Stripe require in order to set all this up and I  
334:16 think Stripe would work better with like a live  published app since uh it has like a fair amount  
334:20 of callback functionality and I want to be able to  give it a URL. So I I'll do all that stuff after.  
334:24 Anyway, I'm not touching anything here. Model is  now proceeding autonomously and it's doing set  
334:28 testing without me. spotted a couple of issues  so far with the voice profiles table. So, it's  
334:33 going to go through and then actually apply the  migration. I can see this doing a bunch of silly  
334:37 BS here trying to solve some Superbase issue. So,  I told it just to give me whatever uh it needs  
334:42 and I'm going to go through and actually I can see  this doing some silly BS issue. Um what I'm going  
334:46 to do is I'm just going to run this directly  in the Superbase editor instead. It's saying  
334:50 alternative install this via Homebrew to run it  from here. I mean, like it can, but I don't know.  
334:54 I don't really want it to waste my damn time or my  tokens. So, I just created it. Um, should be good.  
335:03 Cool. Yeah. And you can see I've already done the  creation. So, we've saved ourselves some tokens  
335:07 there. I would hope. And uh yeah, I mean like  that's why I keep an eye on it. Good mental model  
335:11 here is pretend you're at like uh some security  desk job and part of your job is just like  
335:16 reviewing CCTV footage and there's like 10 CCTV  things in front of you. You're just like kind of  
335:19 glancing around at each. It's basically what we're  doing here. Although um you know with a little  
335:23 bit less context switching since I'm using fast  mode. But everything that I've done here, you can  
335:26 absolutely replicate with like two or three cloud  windows open. The reason why I'm starting with  
335:30 GitHub first is because I want this to be version  controlled. And to be honest, like you could have  
335:35 done this from step one. I just uh I'm not working  with anybody else in this project. It's just me,  
335:40 myself, and I. So this local folder is enough  for me until I get to, you know, the part where  
335:44 I'm ready to go. You can see it's um looking to  ask a question. Uh h I should probably do this  
335:49 privately because I don't want anybody to take  my tokens. Cool. We'll do that after it's done.
335:56 um push to Netlefi. Use my credentials  since we're just demoing. I'll remove  
336:02 them later. And that way I'm going to  have like my env keys and stuff like  
336:05 that so I can actually go through the  uh the generation process and stuff.
336:11 Cool. So now it's setting up Netlefi  deployment and then now we just need to  
336:15 connect to Stripe. So I'm going to head back  to Stripe and then I'm going to open up that  
336:19 uh publishable key thing. So signing in should  now open up a sandbox. So what I can do is I can  
336:25 actually copy the publishable key. Oh, that's not  it. I'll head over to ENV and then I'll also say  
336:32 Stripe publishable key. Add that. Also go Stripe  secret key. I'll add that. And I realize it's  
336:40 probably going to have to redo the update. Hey,  just a heads up. I've added Stripe publishable  
336:44 and secret keys tov as well. Our next step is  going to be to connect it to Stripe to get our  
336:49 payment integration set up. I'd like you to handle  this end to end autonomously. Once it's done,  
336:55 um, I'll do an endto-end test of the app myself  and let you know if there's any final changes.  
336:59 Cool. So, we've since pushed this app live and we  can actually see it at splinter-app.netlifi.app. I  
337:05 don't really think I needed that dash app there,  but the model made that choice and that's okay.  
337:09 Um, I'm liking this. It also looks like we have  that nice little I didn't really notice, but we  
337:13 have that nice little 3D geometric background.  Um, we're going to have to find ways to insert  
337:19 these pipeline visualizations. I've realized now  that we haven't done that on the landing page,  
337:22 but it shouldn't be too hard. I think we'll be  able to sort that out pretty quick. I'm also  
337:26 going to want to widen the landing page a little  bit because notice how the hero header is really  
337:29 wide and then everything else is really kind of  compressed. So, I'm just going to voice dump this  
337:32 and then um have it run through some final  designs. Hey, the landing page looks solid.  
337:38 Just checking the live Netifi link. Now, I'd like  you to add more platforms and social proof. Um,  
337:43 so just add more company logos and stuff. Aside  from the hero header, I find that everything  
337:48 underneath is pretty squished. Like it's not  very wide. It's very narrow. So I'd like you  
337:51 to significantly widen the design. Also, we have  processing pipeline visualization, tone analysis,  
337:56 matrix, multi format, compiler. These three are  placeholders that were meant to be filled in.  
338:01 What I'd like you to do is actually go through our  app and then fill this in with relevant sections.  
338:06 um not screenshots, but actually just mini builds  that uh elucidate various parts of the app,  
338:12 whatever the the specific feature is. Make sure  to widen the pricing and then the call to action  
338:16 as well. And just in general make this a little  bit longer and a little bit sexier. All right,  
338:21 let's take a peek at that actual landing page now.  We should have everything essentially ready. Oh,  
338:27 that is so cool. I don't think I'm going to get  over just how interesting it is to move my mouse  
338:32 around. Um, looks like we added significantly more  context here. So, we could see these little steps.  
338:37 And wow, that some of this is actually dynamic,  which is wild. So, future content creation isn't  
338:43 about producing more. It's about distributing  smarter. Here's how we scale our reach 10x without  
338:46 hiring a single additional writer. LinkedIn X,  blog, whatever. That's cool. We can then actually  
338:51 move these things around. That's really, really  interesting. Oh, wow. Nice. So, we can move the  
338:58 emoji usage and stuff like that around and then  add various features and and buttons. And this  
339:04 over here has a bunch of different sections for  different types of formats. Hot take. If you're  
339:10 still copy pasting the same content across every  platform, leaving 70% of your engagement on the  
339:13 table. Hell yeah. Threads. We added some more  steps down here. So, clean. We then added some  
339:19 data. We added an additional pricing tier,  which is nice. Creators love Splinter. So,  
339:23 we have some supposed big creators. And then  we can start our free trial or book a demo. I  
339:27 don't know if there is a demo feature though.  Is there? Yeah. Okay. So this is where we add  
339:30 our calendar. It's pretty neat. I'm probably  not going to do the demo stuff. Um anyway,  
339:34 that's fine. We'll start free trial. So now I'm  going to add some context. So we'll go Nikki  
339:40 Sarif. Go [email protected]. Actually, why don't we  use a real email? We'll go a very simple password.
339:51 So the creating is now taking us to the  onboarding. This is super clean. So maybe we just  
339:57 want LinkedIn for now. We're going to set up our  voice. So move this around. Looks like we don't  
340:03 have access to that advanced section just yet.  Let me see if I could pull something from my blog.  
340:11 Maybe we'll say this. Paste this in. Then I'll  click generate. We should go through a generation  
340:19 script here and it should be real. Okay. So  the output is what you get in production. Nice.
340:31 Cool. This is now taking my post and then  it's um iterating on it slightly. Wow. This  
340:39 is great. Nice. Oh, nice. Yeah. So I think  the the difference here and the reason why  
340:49 this reads so well is because I specifically told  it do not have those crazy short sentences with  
340:56 like super long sentences immediately after  and then alternating cuz it's like blah blah  
341:01 blah blah blah blah blah that's not pattern  that's signal right it's like you know that  
341:05 that's AI just the way that it writes it this  has solved that for me which is clean okay  
341:10 nice anyway finish setup apparently we're  all set so we'll go to the dashboard here  
341:15 okay And we ran into our first issue, which is an  application error. So I'm going to paste this in.
341:22 This occurred right after the onboarding wizard.  And now I'm just going to continuously push until  
341:28 I get all the way through the app. Now,  logically, what we should do after we are  
341:31 done our little onboarding wizard is we should  go directly to a pricing page. I'm okay with  
341:37 people making like their initial request that  little LinkedIn rewrite thing without having  
341:41 to sign in because while I know some people abuse  that um the abuseness I think will be outweighed  
341:46 by the convenience and then the wow factor of  actually just being able to pump content in and  
341:50 then getting it like nicely formatted and tuned  for LinkedIn. So I'm going to keep that in. But  
341:55 immediately after like we do want to force them to  pay and at that point they'll be in a little bit  
341:59 of like reciprocity like they will just have used  our app for free which allows us to say use for  
342:03 free and then in return um you know it's sort of  like a like a teeter totter or a scale in return  
342:09 they're going to want to balance that by paying  us and then I think I'm going to like make the  
342:13 intro offer like pretty affordable so people can  you know sign up get the ball rolling and stuff  
342:18 like that without necessarily having to spend  a lot and then um from there you know as they  
342:22 go grow addicted to Splinter her then uh you know  eventually the monetization will improve. Okay,  
342:29 this looks pretty good. Nice. All right, so why  don't we give this a little refresh here and  
342:33 see if our data is still in. Beautiful. So we have  this. The thing is we don't have any tokens. So we  
342:40 do have to ask ourselves how many credits. So free  plan. Okay, so we can do five I guess in total,  
342:44 right? So maybe we'll go new syndication up top  here. We can create content. But notice how we  
342:50 don't have LinkedIn or Twitter. So, one more  change I'm going to make is under Oh, actually,  
342:55 I guess that's to be expected because I selected  LinkedIn and then that one X or Twitter does not  
342:59 have an API and then blog is manual. When I click  new syndication, there's currently only LinkedIn  
343:05 X/ Twitter and blog available. I'd like you to  add all. We also need to force the user to connect  
343:10 at minimum LinkedIn and we need to do this post  onboarding. To be clear, are we giving them five  
343:15 free credits as in five free generations? Another  thing on LinkedIn is now that my app is live,  
343:20 we actually need to change the URL. So rather than  being localhost 3000, which is what it was before,  
343:26 we actually have to go back to Splinter and then  type in this, not that, this. So that should now  
343:34 give us the right redirect URI, which now if  I click on this, this will take us back here,  
343:39 which is good. So, I'm going to paste in this  password, click sign in, see if I have to  
343:47 connect via this. Okay, no, and I don't. That's  great. I'll head over to dashboard here. Now, we  
343:53 have one active platform configured. And um what  else we should do is we should offer some sort of  
343:57 like notification saying, hey, connect an active  platform. If a user has zero apps configured, then  
344:03 encourage them to connect to platform. Um, drip  this out in various places throughout the app.  
344:10 Okay, I'm going to add that here. And then I'm  just going to paste this in, then click generate.  
344:16 So now I'm actually going to run through and  I want to see how this interacts with credits,  
344:20 right? We're not doing payments here. We're  just doing um credits. It's your consumption.
344:32 Cool. This looks pretty solid. Okay, nice. So I  can do that. Let's check and see whether or not  
344:37 this is actually going to work by heading back  to my page. And then I'll also check my posts.  
344:44 Then if I can make the post, uh I'm just not  going to test it again. I'm just going to assume  
344:48 that this is going to work. Okay, it's saying  published. Refresh. Okay, cool. So, I'm going to  
344:56 now delete this. That's pretty great. Uh posting  functionality, which is the core functionality,  
345:01 clearly works well. Okay, going back to  the dashboard, we now have one generation,  
345:06 but no tokens have been consumed. So, we're going  to have to fix that as well. I'm not seeing any  
345:11 credit consumption. I just submitted a request as  a Okay, cool. So, credit consumption here hasn't  
345:17 been fixed, but I do have this, which is nice.  So, I have the whole post. That's lovely. Let's  
345:23 um see if we can create a voice profile now. Okay.  And then we can also adjust the system prompt,  
345:27 which is nice. Cool. Looks like there's a uh  default system prompt here. And then there's  
345:31 also the ability to add writing examples. Nice. I  really, really like this design. It's super clean.  
345:36 Very brutalist, which is nice. As you can see,  we have one data point. Um, it's picking up our  
345:40 time. So, 12:57 p.m. we submitted one, which is  really cool. And then it looks like we've outputed  
345:46 one LinkedIn post, but it's also writing one  Xthread. Also, I only published this to LinkedIn,  
345:52 but it's telling me that I've also published  an X thread. Perhaps we should just add more  
345:57 context surrounding output types in analytics  so that they know that you have the option to  
346:01 copy from Xthread whereas the LinkedIn post  was actually published. Okay, heading down to  
346:06 scheduling here we could see a content Q. So  this is the generation I can actually select  
346:11 uh something I have generated which is neat.  So maybe instead of like publishing what you do  
346:16 is you just generate a bunch of things. You then  select the specific platform you want. So in our  
346:20 case LinkedIn then you can set the date using this  cute little calendar. So, why don't I pretend it's  
346:26 2025 0228? Okay. And then I don't know, we'll  do 255 in the morning. That looks good. Let's  
346:34 add to Q. Cool. And so, now we do have something  coming up on Saturday at 255, which is nice. Now,  
346:42 moving over here to setup. It looks like I can  change my account details here. Let's just pretend  
346:49 I'm only saving my name. If I refresh this, you  see that that does not persistent. The setup page  
346:55 doesn't save my account details when I change it.  For instance, I changed my name from Nikki Sariah  
347:00 to Nikki Wiki. I press save changes and then  refresh the page and there was no um context  
347:05 there. I do like how under billing we do have uh  both the free plan and then we have starter, pro,  
347:10 and business. So, that's nice. Then we have  these little notifications and it looks like  
347:13 they're toggling fine, which is good. I do wonder  I don't think that this is 100% hooked up. So,  
347:18 we'll have to double check that. We have the  ability to delete an account. Oh, wow. Okay.  
347:22 Okay. And that autodeletes. So, we need to set  some sort of like feature that basically checks.  
347:28 I just clicked delete account and it immediately  signed me out. Make sure that there's some sort  
347:32 of are you sure message. Okay, that looks pretty  nice. Okay, now I guess this is good for me. If  
347:40 this actually did delete my account. Okay, no,  it didn't actually delete my account. I see. So,  
347:45 it just said that it deleted my account, which  is unnecessary. Never mind. It just said that  
347:49 it deleted my account. It didn't actually do the  deletion. Okay, now I'll generate new key example  
347:56 key. Okay, and then we need the curl details.  I'm just going to see if we could key here.
348:10 Let's reveal this. And I'll say API key  send a request to this URL using the format.  
348:21 Let me see if it makes any sense whatsoever.  API splinter.io. Okay, for the API request, it  
348:28 doesn't look like the domain is a live host. Um,  update everything so that it reflects the new URL,  
348:35 not api.splinter.io. And then finally, we have  this web hook feature, which uh I don't know,  
348:41 let's just go web hook test. And over  here, we can copy a web hook URL. We  
348:47 could basically paste that into our app and say,  "Hey, when this is complete, scheduled or posted,  
348:52 I want you to deliver to this." So then I'm going  to go back to content library. And then I'm just  
348:56 going to do one more. So I guess we'll go to  dashboard, new syndication here. Then I'll just  
349:02 copy all of this over again. Paste it in. Okay.  And we have all the rest of these buttons now,  
349:07 which is wonderful. Click generate. And then I'm  going to go here and then check to see whether or  
349:13 not a web hook was actually sent. If the web hook  wasn't sent, I don't believe it will be. I'm just  
349:17 going to notify the agent. Hey, the generation  occurred, but the web hook was not sent to the  
349:21 address that I added. Um, double check that all  of that is okay. I imagine there's probably some  
349:25 issue related to the fact that we're not on the  URL that we initially thought we'd be on. Yeah,  
349:32 cool. So, we have that the generation has  occurred. We didn't end up getting that.  
349:36 I can copy this, though. Go back to dashboard.  Now, I see we have two created. And there's the  
349:41 content library here with the the the second  post, which is nice. Okay, what else is it  
349:45 telling me? It's telling me that um all the code  examples have been updated. So, I should be able  
349:51 to go back to API access here and then give this a  little refresh. I'm giving this a refresh because  
349:58 I wanted to hardcode the change. Okay, just  taking a look at this code. Example still says
351:02 Okay, I'm going to go to upgrade and let's see  if that works. So, we can now select one of these  
351:07 three options. So, I'm going to go pro. Seeing  that it's opening up a Stripe checkout page,  
351:11 which is good. I'll go to USD and then I'm  just going to pretend enter some details  
351:15 here and then click subscribe. Looks like the  processing is occurring, but is it going through?  
351:20 Okay, cool. So, it looks like it went through.  Did we go back to checkout? Okay, cool. Yeah,  
351:24 we have and we've upgraded. When a user pays, um,  add some confetti or something upon the redirect  
351:30 only when they make it back after paying. Okay,  after a little bit more back and forth, you can  
351:34 see we actually have the right profile set up. So,  I'm going to head back to Stripe, generate a new  
351:38 key, and then actually reveal and then copy this  whole key. And I'm just going to see if this ends  
351:44 up working because that's the API functionality.  Okay, we're still getting unauthorized. Um, this  
351:53 using this superb basease session. I don't know  exactly what that is. I imagine this is probably  
352:00 going to solve it. Yeah. Okay. So, we we're going  to have to rewrite that a little bit. The good  
352:04 news is the name now saves along with various  password problems. Uh, we've now solved all of  
352:09 that. Wonder what managed subscription is going  to look like. I've never been there before. Oh,  
352:13 nice. Okay, cool. So we can actually like we we  literally have like our own managed subscription  
352:17 page which is Sweet. Um why don't I go to  dashboard just so I could take a look at that as  
352:20 well. Cool. So here are the generations and stuff  like that. Nothing upcoming because I don't think  
352:24 we have anything scheduled. Yep. Cool. Yeah. All  right. This looks pretty good. Generated formats.  
352:29 LinkedIn post two generated exert one generated  formats generated. Cool. Cool. This tells me what  
352:34 they are now. And then we have access to all of  our old posts. Awesome. So now I'm just going to  
352:39 mobile optimize this and then run security audit.  And then honestly that's it. We're live, baby. We  
352:44 can see we're actually sending over a live API  request to the Splinter-app API as well. So,  
352:49 this isn't just a front-end app for us anymore.  This is entirely um yeah, this is super clean.  
352:54 Super clean. So, yeah, pretty stoked about that.  We also have the API functionality. Obviously,  
352:58 we could paste in links as you guys have seen.  Not going to lie, this is pretty solid as far  
353:02 as apps go. We uh now basically have something  that's capable of doing everything that I want.  
353:06 Last thing is I'm just going to paste in this  security prompt. And just because we've done  
353:09 this what three times now, I'm not going to bore  you all with the details, but suffice to say,  
353:13 using Gemini 3.1 Pro in conjunction with Claude  Code for both a mixture of front-end design and  
353:19 then backend configuration, route creation,  and then payments, and so on and so forth,  
353:23 is pretty dang efficient. Now, I should note  that we used a ton of tokens this time, and I  
353:27 don't actually recommend using that many tokens,  like committing that many tokens to an app. Um,  
353:32 I just did this because I traded off speed uh  basically for money. I wanted to get this done  
353:37 as quickly as humanly possible because of an  upcoming engagement, not necessarily, you know,  
353:41 have three or four instances of cloud running  or uh, you know, take three or four hours when  
353:45 realistically I could have done it in 45 minutes.  So, I'm not going to push the fast mode on you.  
353:49 It's just an option. Uh, just keep in mind it does  cost a fair amount. This session was probably at  
353:53 least $70. And I'm sure you can imagine if  you develop an app every couple of days or  
353:57 work on your codebase every few days, that sort  of thing is going to add up pretty quick. For me,  
354:01 I'm happy to do it because of the arbitrage and  then the value that I got to produce. Hopefully  
354:05 you guys now see how to build an app just like  this. Not just a simple CRUD app, but a CRUD app  
354:09 that also has API wrapping functionality that does  OOTH integration with social media platforms even,  
354:15 which are some of the most secure and difficult  to authenticate and organize. And then finally,  
354:20 um, payment integrations as well, like with  Stripe. All right, so you now know how to  
354:25 vibe code an app. Congratulations. However, most  people that have gotten to this point are probably  
354:31 not just interested in building internal apps and  tools for themselves. Like, if you've gotten to  
354:35 this point, you're probably really interested in  building for other people, whether it's consumers,  
354:40 whether it's other businesses, maybe it's some  organizations that you're working with or whatever  
354:43 you want to help economize their workflow and so  on and so forth. And so, what I wanted to do next  
354:48 is round out the course with a module all about  pricing, packaging, and selling what you build.  
354:54 Now, most people that are here in your shoes,  okay, that are probably interested in um pricing,  
355:02 packaging, and selling what they just put  together, don't really have any experience doing  
355:06 so. And so, it's not that they don't have any like  vibe coding um business experience. It's that they  
355:11 don't have any business experience in general. And  so, what I wanted to do is just give you a very  
355:14 highle look at some tactics and playbooks that  you could follow to significantly streamline your  
355:20 ability to actually successfully go out there and  then sell this sort of stuff. whether you know at  
355:25 a small scale or whether at a large scale, whether  your goal is to replace your income or whether  
355:28 your goal is to, I don't know, start a business  that does $100 million. Now, I think there are  
355:32 a lot of people on YouTube that make courses on  vibe coding and stuff and um they're like, "Oh,  
355:37 you know, here's how you make all these super cool  crazy apps and here's how I spun up 500 clones and  
355:42 publish them to the app store." But not a lot of  these people have actually made money. So, I know  
355:45 that this is a vibe coding course, but hopefully  it'll allow me to dive a little bit into the  
355:49 business side of things because personally, that's  what I find the most interesting. Collectively,  
355:52 my businesses make over $300,000 a month in  profit. Um, and that's a pretty easy number to  
355:56 hit revenue-wise, I think, if you know what you're  doing. But profit-wise is a different story. So,  
356:00 all this to say, you know, I I do this sort of  thing professionally. I generate revenue outcomes  
356:04 and profit outcomes professionally for businesses  at this point. Let me run you guys through at  
356:08 least at a high level what I know and what you  guys can use to make some money with vibe coding.  
356:12 So, the big thing I want you guys to know is over  the course of the next, you know, 5, 10, 20 years,  
356:17 um, software will no longer be the moat. So back  in the day, right, 30, 40 years ago, if you were  
356:22 a cool startup and you were working on some new  computer system, okay, and you were smarter than,  
356:28 you know, your competitors and you spent more time  on it than competitors, you might have had like  
356:32 a small edge and maybe with that small edge, you  would have carved out like an operating system or  
356:37 something like that. And maybe potentially you  would have called that Windows OS and then you  
356:42 would have sold that at scale to people and that's  how you would have made your money. So if you  
356:45 think about it like Microsoft and Windows, Windows  uh uh was like software mode, software as a mode.  
356:51 Same thing with like the Apple brand of computers.  Same thing with like I don't know applications  
356:55 afterwards like Photoshop and like various  databases and stuff like that. Businesses that  
357:00 were built entirely off of software made sense  like 30 or 40 years ago because software back then  
357:05 was very difficult to make. Okay. Well, nowadays  software is very easy to make as hopefully I've  
357:10 just demonstrated here over the course of the  last six or seven hours or however long this  
357:13 damn course ended up being. So, we've transitioned  pretty heavily. Unfortunately, a lot of people are  
357:18 still stuck in that old school world where they  think, okay, that they're going to put together  
357:22 this incredible software solution that nobody else  in the market can do because they're smart and  
357:27 they're intelligent and they're they're visionary,  they're a forward thinker. The probability that  
357:31 you do that is extremely low. It's extremely low.  Not just because, you know, nobody's as special as  
357:37 we think we are, but uh because there are soon to  be 100 million times as many software engineers in  
357:44 the form of AI agents than there ever would have  been human beings. And so moes are no longer made  
357:51 through software. Your edge in the market is no  longer in the cool algorithm that you come up with  
357:56 or whatever the heck. Okay. where the moat is is  in distribution, which is fundamentally different.  
358:05 And rather than play by the software mode playbook  like everybody in their mom is doing right now,  
358:10 um I encourage you to play by the distribution  mode playbook. Hypothetically, why do you  
358:15 think Netflix is still Netflix in a world where  literally anybody could just tell their agent,  
358:21 hey, here's Netflix.com. can you go and recreate  everything about it and then I don't know you  
358:27 know host the the servers and set up databases  and set up all the infrastructure and then like  
358:31 have a little login page everything like that the  reason why Netflix is not broke you know today is  
358:37 because Netflix contains within it distribution  it contains a collection of clients vendors  
358:45 relationships their staff their team their culture  and so on and so on and so forth and these are  
358:51 things that an AI agent and until they, you know,  hook it up to the cloning vats and start pumping  
358:57 out clients, vendors and and staff members, um,  cannot currently do or reproduce. And so software,  
359:05 which used to be like the golden goose, right,  has basically been replaced almost completely  
359:12 with distribution. Okay? And in a world where  anybody could build any software, the only thing  
359:18 that makes your software different from somebody  else's software are the customers essentially the  
359:23 vendors, the relationships, and the staff on  your team um that manage it. Anybody in 2026,  
359:29 2027 and beyond with a couple of ideas, a high  degree of agency, which just means your ability to  
359:35 like enact your will on the world, go out and do  things, and a very small budget, just a few bucks,  
359:39 literally 15, 20 bucks, can now build whatever  the heck they want. the whole moat, the whole  
359:44 edge is just your ability to actually reach and  then retain people within an ecosystem. And if you  
359:49 can do that, right, and if you can focus your time  and energy accordingly, then you can make a lot  
359:53 of money. But if you're stuck in the software  days, then you're just spinning your tires,  
359:56 man. The whole world's up and moved on and you  don't even know it yet. Hypothetically speaking,  
360:00 I want to pretend right now that um you know,  we we we got a little clock here, and this is a  
360:05 terrible drawing of a clock, but you know, bear  with me. Just pretend that this whole thing is,  
360:10 you know, a 60-second um um hand. Imagine that the  entire life cycle of this vibe coded business that  
360:17 you are about to start is this hand. Okay? And it  traverses a whole 60-second length. Here is how  
360:23 much I'd recommend that you actually spend on the  vibe coding part. Okay? Over the full 60-second  
360:30 life of your business, I would recommend you spend  maybe 3 or 4 seconds on the actual building. Let  
360:39 me make this a little smaller so we can see it  better. Everything else, okay, literally the  
360:46 rest of all of your time, my recommendation would  be to spend it not on building, not on tweaking,  
360:53 not on making your server infrastructure better,  not on doing whatever the hell. My recommendation  
360:58 would be to spend it on marketing. And this isn't  a joke. If you spend 3 seconds or 12th or 5% on  
361:08 building, you should spend the other 1920ths or  95% on the marketing. I don't know if you guys  
361:18 have been in like the the build and public space  on X. I used to be uh uh you know following that  
361:23 fairly frequently because it was just interesting  to me how people would share their revenues and  
361:27 talk about the apps they were building. And back  then they said you should spend at least as much  
361:29 time building as you spend marketing. You should  spend at least as much time marketing as you do  
361:33 building. And I was like okay so 50/50. Um that  was like four or five years ago. That no longer  
361:38 applies. You need to spend like 19 to 20 times  as much time marketing than building cuz anybody  
361:43 could do the building. The building is literally  just like your entry ticket to get onto the ride.  
361:48 The whole ride itself is your poorly spelled  marketing. So without marketing and sales,  
361:54 to be clear, you have no product. You may think  you have a product, but it's not a product. It's  
361:59 just a hobby because without customers, you  don't have any sort of business. You need  
362:04 revenue and so on and so forth. There are also  some additional benefits to having um customers  
362:09 that go far beyond the money that they pay you.  If you have customers, what you get is you get  
362:14 faster iteration because you're listening now to  real market feedback. You're not listening to like  
362:19 your own opinions and your assumptions, which is  what most builders will do. Most builders are like  
362:24 heavily biased. I mean, who the hell builds  software products anyway? For the most part,  
362:27 giant nerds, okay? Myself included. I don't know  what the average human being that uses my product  
362:33 actually wants or thinks. So, everything that I'm  doing is just assuming. I'm using my own opinions  
362:38 essentially and I'm hoping that my opinions  are correct because I possess a strong level  
362:41 of empathy or whatever the hell. But the reality  is most of the time I'm wrong. You know, it's like  
362:46 a ven diagram. You know, maybe like if this is  my opinions, maybe only like 50% of my oh my god,  
362:53 I really need to I'm writing like a doctor.  Um only like 50% of my opinions are actual  
363:00 reality. And so the only part of like this ven  diagram that I'm actually moving the needle for  
363:05 and like building features into my app that  actually make sense, okay, are over here. But  
363:09 like the other 90% is all [ __ ] you know? It's  all like uh I'm just wrong. So why would you even  
363:14 like worry about your opinions? Instead, focus off  reality and ground truth. The only way you can get  
363:21 reality and ground truth is getting customers.  The ideal is sufficient number of customers.  
363:26 so many so that you could start averaging  out their opinions and beliefs and desires  
363:29 and needs and so on and so forth and then you  just start designing for them. So rather than  
363:33 letting your own assumptions build the product,  use customer requirements, use customer needs and  
363:37 actually do like customer surveys. Launch your  vibecoded app today. You know, hypothetically,  
363:43 you've seen how easy it is to do. You know,  I can build one of these in just an hour or  
363:46 two. Okay? And then contact 20 people and have all  of them use your app for free and then tell you,  
363:51 hey, I liked this. I didn't like that. Hey, I  thought this could have been better. Hey, this  
363:54 didn't really solve my need. Hey, I wish it did  this. I wish it did that. Take all that feedback  
363:58 and then use it to design the next version of  your app because it's not the first version of  
364:01 your app that's going to make you any money. It's  like the 20th version of your app that's going to  
364:04 make you money. In addition, because customers and  distribution is really the moat, your whole focus  
364:09 should be on maximizing the customer experience at  every step. It should not be on making the nicest,  
364:14 sexiest, most polished app. And I know this is  kind of annoying cuz I did just spend the last  
364:18 part of, you know, five or six hours hopefully  teaching you guys how to do this and how to like  
364:23 make a really good product. But in reality, the  product, as we've seen, is really only like 5% of  
364:27 the whole pie now. So, focus instead on acquiring  as much distribution as quickly as possible. I'm  
364:32 going to run through some specific ways on how to  do that. And then once you have that distribution,  
364:36 um try and satisfy the needs of your customer.  And then generally and consistently iterate on  
364:40 the product. Okay. So, highle advice aside,  how do you actually do the big thing that I  
364:46 think a lot of people are wondering? How do you  price your product? Well, what I want to do next  
364:52 is talk about essentially my core principle for  pricing. And this is a core principle that I've  
364:56 used to generate tens of millions of dollars in  revenue across a variety of businesses, small  
365:00 to mid-size businesses, my own agencies, digital  marketing businesses, uh a couple multi-billion  
365:05 dollar portfolio companies. And um this is  something that I just I I continuously return  
365:09 to you because it's the easiest and simplest  way to price. So, it's called VBP or valuebased
365:18 pricing. And so in value based pricing, what  you do is you price based off the value that  
365:23 your app delivers, not on your costs or your  competitor's pricing. You know, and I think a  
365:29 lot of people are like, well, I have to anchor to  my competitor's price, don't I? And to me, I say  
365:32 I mean like obviously your competitor's price is a  factor and people are going to be doing some price  
365:36 shopping. But if you focus on the value instead of  at what your competitors price rather than running  
365:41 uh down to the very bottom and it being like a  race down below like a lot of these freelance  
365:45 platforms and stuff where somebody bids nine,  another bids eight, another bid seven till you're  
365:49 at freaking zero. What we do is we just establish  how tremendously valuable a product is and then  
365:54 charge such a vanishingly small percentage of said  value that the customers are bought in regardless.  
366:00 I mean to them it's like a it's it's a pittance.  And I know this because I also purchase software.  
366:04 um what ultimately a lot of B2B software users  are looking for is something that maximizes their  
366:09 return on investment and minimizes the friction  involved in making the purchase decision. So let  
366:13 me explain what I mean by all this. There are two  types of value that you can produce at least in a  
366:17 B2B environment. In a B2C environment you have  additional types of value. Those types of value  
366:21 are nonmaterial and so it will not be discussed.  You have the ability to directly reduce their  
366:27 expenses which is the money that your product  will save a customer if they use you. And then  
366:32 you also have the ability to significantly  increase their revenue in opportunity cost,  
366:38 which is also termed the additional money  customers can now earn because they use your  
366:42 product. I think direct expense savings are pretty  simple, right? If I have product X and product X  
366:48 currently costs the user $100 a month and I say,  hey, I just built product Y. Product Y only costs  
366:55 $5 a month. What do we have there? What I've done  is, if you think about it, I've saved the customer  
367:01 $95 every month. And so the direct expense savings  are 95 bucks. Easy. Now, an opportunity cost. Now,  
367:12 opportunity cost is a little bit different.  Instead of focusing specifically on what the  
367:15 delta between, you know, their old thing and then  their new thing is, what we focus on instead is  
367:21 how much time do they save and what is the  cost of that time. So I'll give you a brief  
367:30 little example. Let's say hypothetically you  are currently a salesperson. You work at big  
367:36 company. Big company pays you on average about  $100 per hour. Right now, as part of your day,  
367:44 you do some [ __ ] admin task. That admin task  involves filling out some CRM or whatever. And  
367:50 on average, every single day, you have to spend  2 hours doing this. Okay? So, they're paying you  
367:57 $100 an hour. You're doing an admin task that is  worth 2 hours. If you hypothetically could build a  
368:03 software solution that enable this salesperson  to spend zero minutes doing this admin task,  
368:08 how much money are you saving this salesperson  right off the bat? You're saving this company  
368:13 right off the bat. You're saving $2* $100 an hour  or $200 an hour. That is your direct expense. But  
368:23 if you think about it, what do sales people do?  Salespeople are tasked with increasing the revenue  
368:29 of the business that they work in. Hypothetically,  let's say if instead of focusing on admin time,  
368:36 they then focus on sales, aka calling people or  whatever. in those two hours they could contact  
368:43 50 people and if in one in 50 people end up  purchasing a small product or something like  
368:48 that okay which is their conversion rate and if  the average cost of that product is let's say  
368:53 uh I don't know $500 then by saving 2 hours a day  not only are you adding $200 sorry $200 an hour  
369:03 I meant $200 not only you adding $200 back to the  core business that you built the software platform  
369:08 for you're also adding If you think about it,  an additional $500 in revenue. So this is minus  
369:15 $200 of expense. Okay? And this is plus $500  in revenue. And if you think about the total  
369:25 amount now that this software product is saving  or generating the business, we have come to  
369:32 $700 in value and that's per day. Okay. So the  expense portion is about saving money whereas  
369:40 the revenue portion is about reallocating the time  that your software solution is now saved to allow  
369:45 a person at a business to focus on higher revenue  things that actually produce for a company. And in  
369:51 practice usually these expenses are much smaller  than the total amount of opportunity cost aka  
369:56 the revenue that you are unlocking. So you know  how earlier we built a thumbnail generator? Well  
370:00 here's a quick example of what that might look  like. You have direct cost savings. Let's say  
370:05 right now whatever system that they are using  cost them $50 a thumbnail and they generate 20  
370:10 thumbnails a month. That means that by switching  over to nyx thumbnail company.co, also known as  
370:16 thumbly, uh you'll save $1,000 a month. But when  you take into account the time savings revenue,  
370:22 okay, you move this 2 hours per thumbnail design  period to 5 minutes. What you're doing is you're  
370:28 saving 115 minutes. And hypothetically, let's  say these 115 minutes times 20 are 2,300 minutes  
370:36 a month. Let's say with this additional time,  which works out to be approximately 40 hours,  
370:42 this designer or something can actually make  an additional $1,000 a month as well. Well,  
370:46 now the total value created is $2,000 a month.  What we're doing as somebody that is developing  
370:51 a thumbnail product is we're pricing at a very  small percentage of the total value we produce,  
370:55 so maybe 10%. So, we charge 200 bucks a month. And  then the customer ends up making 10 times as much  
371:00 money as they spend with us for using our service.  Now, obviously, this is going to be an average,  
371:04 right? Every customer is going to be different.  Some people are going to produce way more value  
371:08 than your $2,000 a month estimate. Others are  going to produce less value. But because you  
371:12 can't price individually per every new customer  that you onboard, you have to pick a specific  
371:16 package price. What we've done now is we've  basically used facts and logic to work out what  
371:21 a strong value would be for a core product and  then how much to charge for it instead of relying  
371:27 on things like competitor analysis or worrying  about what the dude in the next cubicle over is  
371:32 doing. And so this is a much more sustainable and  higher ROI way of pricing. You're pricing based  
371:36 off value. You're not pricing based off of what  everybody else is doing. This young viewers is  
371:41 how you zigg when everybody else zags. Once you've  established some sort of price for your product,  
371:46 my recommendation is I actually establish  this as my top price. So, for instance,  
371:51 um let's say I kind of screwed up the uh writing  here, but let's say this is like your top tier,  
371:56 this is your mid tier, and this is your entry  tier. You know how there's typically multiple  
372:00 subscription tiers for a product, right? Assuming  you're doing a monthly thing. What you do is you  
372:03 take the top and then you set that one at uh you  know, let's say 200 bucks a month. And what you  
372:08 do is you set your middle at approximately half of  that and you set that at about 100 bucks a month.  
372:13 but you uh reduce the ROI to compensate. So in  this case, you know, previously I'm offering a  
372:17 10x ROI on my top. Well, now I'm offering an 8x  ROI on my mid and then a 6x ROI on my entry. And  
372:23 so in this way, you have like a reasonable product  differentiation. You have a reasonable spread of  
372:27 values. And this is enough to get you started  before, you know, you can get enough customers,  
372:30 you can actually run like strong analytics to  determine whether or not people would actually  
372:34 prefer a $37 a month product or a $58 a month  product or whatever. And that's honestly all there  
372:39 is to it. To make a long story short, what I do  anytime I'm pricing an app, uh, Pel, an automation  
372:44 service, a back-end tool, even like a big business  deal, back when I was um, operating a content  
372:48 writing agency and we were trying to work out how  much to price our product, uh, we do it all based  
372:52 off of value pricing, which is determining how  much value our product has to the user to begin  
372:57 with, an expense, an opportunity cost, and then  just charging a small percentage of that. And  
373:01 that usually ends up being somewhere between 10 to  20% of the total ROI. And as a business owner, I'm  
373:06 thinking like, hey, I could put my money in the  S&P 500, I could make 8% a year, or I could give  
373:10 my money to NYX Co. and then I'd make a,000% on my  initial investment. What am I going to do with my  
373:16 money when presented that choice? Obviously, I'm  going to give it to Nyx Co. right now. Uh, whether  
373:20 or not somebody chooses to invest in you isn't  just a product of this math. It's also how much  
373:24 they believe the math, and that's another part of  marketing. But I'll leave that for another day.  
373:28 Hopefully, at least now you guys understand how to  price set products. But that takes us to the next  
373:33 big thing, okay? which is well there's pricing the  product but how do I actually get my product out  
373:38 there okay how do I actually market this amazing  thing that I have developed well there are three  
373:45 major ways the first is outbound a lot of people  shy away from outbound but outbound was the very  
373:50 first thing that I did when I started my first  business my first real business about 10 years ago  
373:54 I went door to door literally knocking on business  doors coming in trying to shake their hands and  
373:59 then sell them on my little marketing package it's  very difficult but this is ultimately what works  
374:04 So in this case, you go directly to people who  need your product. You can consider this like a  
374:08 sniper rifle approach. The two most popular  methods right now are cold email, which is  
374:12 where you actually craft an email, usually using a  template that appears to be personalized onetoone  
374:18 communication between you and the person that's  reading. And then you say, "Hey Pete, I love what  
374:22 you're doing over at XYZ Business. I think right  now you guys are currently spending all this money  
374:26 on the software, and I can save that for you with  XYZ product. I would be happy to give this to you  
374:31 for like a major discount because you're going to  be one of my first five uh you know signups. I've  
374:35 also attached a video here that runs through how  to do X Y and Z. Um let me know if you have any  
374:39 questions. If I can help you out at all would be  happy to. Okay. And so in this way you're reaching  
374:43 out directly to them. This is you. This is them  and this is you bridging that gap. That's why it's  
374:50 called outbound. The other main way to do this  is through cold DMs. And typically you do this  
374:54 on B2B platforms like LinkedIn. Cold DMs stands  for direct messages. It's just the same thing  
374:59 as cold email. It's just instead of using like an  email inbox, which tends to be the the highway to  
375:04 the soul, um we're instead using like a LinkedIn  message box or like an Instagram message box,  
375:08 but it's the exact same idea. Some caveats here.  This works really well for B2B products, but it is  
375:14 pretty difficult for consumer products, aka B2C  products, since most of the lead sources where  
375:19 you're going to scrape this information do tend to  be B2B. Like for instance, if you think about it,  
375:23 LinkedIn is entirely a commercial platform. If  you sign up to LinkedIn, odds are you're signing  
375:27 up for employment or business purposes. So, you  can feel pretty safe that most of the people that  
375:31 you're scraping on LinkedIn, you're doing so  because it is a B2B sort of transaction. Hey,  
375:35 I can help you at XC Corp. Hey, I saw you're  employed at whatever position. I think I can  
375:38 assist you with that and so on and so forth. If  they're not on LinkedIn, crap gets a lot harder,  
375:43 right? So, for the most part, anytime I sell B2B,  I start with Outbound. The other big benefit to  
375:47 outbound is it's just really fast. It is way  faster than any of the other methods on the  
375:50 list. It's also way harder, right? you sort of  have to put your freaking ego on the line. And  
375:55 um I can't tell you how many times I got thrown  out of businesses when I was knocking on doors.  
375:59 Security racing after me being like, "Get  the hell out of this plaza." Uh but you know,  
376:03 it it worked and it was what enabled me to make  my first uh probably $15 $200,000. That said,  
376:07 this is not the only thing that you can do. It is  not the only way. There are a variety of others.  
376:12 The second is inbound, which I think a lot more  people here are probably familiar with. Inbound,  
376:17 if you kind of think about it logically, is  kind of like what I'm doing right now. What  
376:21 I'm doing is I'm creating very valuable content.  I'm building a very strong and large audience of  
376:26 people that enjoy consuming my content. And then  what I'll do is I'll direct a small percentage of  
376:30 you guys to some paid products so that you guys  sign up to my service or whatever. The issue with  
376:34 inbound, despite the fact that, you know, if I'm  here and then they're here, despite the fact that  
376:41 they're now coming to you, is it's a lot slower.  It takes a lot longer to build. Okay. And what a  
376:46 lot of people don't realize is they'll watch some  vibecoded course or whatever and then be like,  
376:50 I'm going to make my own app business. It's going  to be awesome. And then they start making content  
376:53 on it. What they don't realize is like when you  make content um on the on the wider internet like  
376:58 on YouTube or LinkedIn or Instagram or whatever,  what you're basically doing is you're like in a  
377:02 sea of a billion people and then you're  given a megaphone. And then you're like,  
377:06 "Okay, I have a megaphone. This is awesome." And  then what you don't realize is every other person  
377:10 also has a megaphone. And then the big people that  are pretty established have megaphones that are  
377:14 way bigger than you. And then the small people  like yourself have these tiny little things that  
377:17 barely produce any noise uh whatsoever. And so  growth on inbound is extraordinarily difficult  
377:23 as a beginner unless you have some pre-established  success or pre-established wins. And most people  
377:28 that are starting these sorts of things that  immediately turn to I'm going to post one thread  
377:31 on Instagram every day for the next year. Uh  they they don't understand this. And so what  
377:35 ends up happening with a lot of inbound is because  it takes so much longer to build and people are so  
377:41 unprepared for this, they end up just screeching  into the void, never achieving anything. and then  
377:45 quitting before they actually build up the amount  of momentum required to succeed. If I could graph  
377:50 sort of like the differences between inbound  and outbound. And I love graphing stuff. So,  
377:54 in case you've watched any of my other videos,  get ready for a freaking Nyxive graph. Basically,  
377:59 this is the amount of time that it takes to see  any sort of win. And then this is the amount  
378:03 of money that it takes. Okay? And I'm going to  I'm going to pretend that blue here is inbound.  
378:08 And I'm going to pretend that orange here is  outbound. And you guys are going to see why  
378:11 I prefer outbound in terms of the total amount  of money you could make. Um, if you use inbound,  
378:16 okay, what happens is you start at this time step  one and then you make no money for a long time  
378:21 and then all of a sudden your brand picks up and  then you make a ton of money. With outbound, it's  
378:26 basically the exact opposite because every single  time you send a piece of outreach, it consumes a  
378:32 little bit of your time. What ends up happening is  you usually have some success initially and then  
378:37 you start tapering out or plateauing and hitting  a ceiling. And so if you think about it logically,  
378:42 both of these have major trade-offs, right? But  with outbound, what happens is you make money much  
378:46 much faster. You know, this might be 1 month. This  over here might be one year. And so with inbound,  
378:52 what we've done is we've sacrificed all our  money in the first year for money after year  
378:55 number one. And with outbound, what we've done  is we've sacrificed money after year 1 for money  
378:59 before year 1. Um in reality things are never  fully black or white and so the recommendation  
379:04 is usually to do some sort of blended take of both  where you know at the beginning you start mostly  
379:10 with outbound. Okay. And then as time goes on you  shift more to inbound which allows you to generate  
379:16 revenue quickly without also tapering out. Um that  said there are a variety of different ways you  
379:20 could do this sort of thing. That's where business  programs like u mine maker school and so on and  
379:23 so forth come into play. So I'll I'll leave that  with you and you guys could do whatever you want  
379:27 strategically there. The third major way that you  could market as somebody that develops and builds  
379:30 products is instead of trying to build your own  distribution via you know outbound like we talked  
379:36 about earlier or inbound where people come to you.  What you do is you just take somebody else okay  
379:43 somebody like Nick Sarif and you say yo Nick can  you market my product for me and then I'll just  
379:48 pay you 30%. And then what happens is Nick, you  know, can use a combination of both inbound and  
379:54 outbound methods to do that marketing for you. But  the idea is because he's already developed sort of  
379:59 pre-existing distribution, it's way faster. And uh  you know, all you're really doing is you're paying  
380:04 a percentage. So it's more predictable for you.  Maybe it won't necessarily be cheaper, but it's  
380:08 probably cheaper than you initially buying a bunch  of software platforms, getting a bunch of cold  
380:12 emails, uh getting a a beautiful studio and all  that stuff. So, the whole idea is you just partner  
380:17 with someone who has pre-established distribution  and then you make it a win-win for them. You're  
380:20 like, "Hey, man. This software does exactly what  the heck you really like to do. Uh, I think I can  
380:25 really help your viewers and stuff like that as  well." And rather than you having to build a whole  
380:28 new product from scratch, why don't you just use  my product and I'll just pay you a percentage. You  
380:32 like my product, right? Yeah. Okay. Awesome. So,  it's a win-winwin. As mentioned, it can be inbound  
380:36 or outbound affiliate, though. Inbound tends to be  the most popular one because it's more scalable.  
380:40 And then yeah, you're just monetizing somebody  else's audience or work, which allows you to  
380:43 like quickly jump cut in uh basically return for  some predictable outflow of your money. Okay, so  
380:50 hopefully I've now showed you guys how to not only  price but also package together multiple tiers and  
380:55 then also get started with the marketing of your  product. My recommendation to be clear is to start  
380:59 with some form of outbound laser or sniper rifle  target specific people whose needs you are solving  
381:07 and handling with your product. And once you have  a solid base set up, maybe 10, 20, $30,000 a month  
381:12 in recurring revenue, then you can transition  over to maybe building a brand um um you know,  
381:17 doing the whole megaphone analogy that in reality  tends to be the best of both worlds. Um affiliate  
381:22 is also open to you at any level of the stack.  So whether you guys are early on, whether you  
381:26 guys are a little later on, I do highly recommend  looking for those sorts of partnerships and then  
381:30 finding opportunities to, you know, plant your  seed and have other people water them from here  
381:34 on out. Obviously, I can't cover every possible  business permutation. Some people are going to do  
381:38 like high ticket SAS style stuff where you have  to sit down and do demos. It's going to be more  
381:42 sales. Other people are going to do low ticket  SAS stuff where basically people guide themselves  
381:46 through their onboarding checklist and stuff. Um,  you know, that tends to be more marketing. But  
381:50 hopefully this at least tells you guys a little  bit about where the market is going and then  
381:54 how to practically take what I'm talking about  here to improve the quality of your guys' lives,  
381:58 businesses, and so on and so forth. Okay, that  was a hell of a course. I had a lot of fun putting  
382:02 all that together. If you guys have made it to the  end, y'all are real ones. I have uh one quick ask,  
382:07 two, I suppose. The first is if you guys like this  sort of thing and you guys are curious about how  
382:11 to monetize your own vibe coding business,  definitely check out Maker School. It's my  
382:15 day-to-day accountability program where I actually  guarantee you your first paying customer for an  
382:19 AI or automation business in 90 days. If you don't  get customer number one in 90 days, I I refund you  
382:25 in full. So, it's a full 100% guarantee. Inside  of Maker School, you'll learn how to put together  
382:29 really classy and clever offers, sort of like  the offer I'm using to get you into Maker School,  
382:34 as well as do things like set up outbound  campaigns, uh, productize your business, and  
382:38 then eventually scale to inbound, like I talked  about here. Additionally, I'd love it if you  
382:41 could check out my podcast with Mr. Jack Roberts.  We both talk on AI automation. It's a blend of  
382:46 news and then us just goofing around, showing  off our boulder shoulders. So, no pressure.  
382:51 Do whatever the heck you want. Lastly, you'll  find everything that you need down below in the  
382:54 project description. And then yeah, if you really  like this sort of thing, then please subscribe.  
382:57 Helps out my channel. Comment and engage with the  video. And I'll catch all y'all on the next one.
$

VIBE CODING FULL COURSE: Gemini 3.1 + Antigravity (6 Hrs)

@nicksaraev 6:23:04 58 chapters
[AI agents and automation][content creation and YouTube][developer tools and coding][productivity and workflows]
// chapters
// description

🔥 Join Maker School & get customer #1 guaranteed: https://skool.com/makerschool/about 💎 All course files: https://drive.google.com/drive/folders/1eGTwguPbANKzhtA7EFq4YSLsZnqA_ZCH?usp=drive_link 🎙️ Listen to my silly podcast: www.youtube.com/@stackedpod 📚 More free courses → Claude Code (4hr full course): https://www.youtube.com/watch?v=QoQBzR1NIqI → Agentic Workflows (6hr full course): https://www.youtube.com/watch?v=MxyRjL7NG18 → N8N (6hr full course, 890K+ views): https://www.youtube.com/watc

now: 0:00
// tags
[AI agents and automation][content creation and YouTube][developer tools and coding][productivity and workflows]