// transcript — 352 segments
0:00 Intro
0:03 I've used Gemini for Argon non-stop for the last 7 days and I had to come on and
0:07 share what I found. You know, most people are going to use Argon just like
0:11 every other AI, but there are three specific problems where if you just
0:15 point Argon to it, it can help you build a real business and no one on the
0:19 internet is talking about it and no one's telling you how to do it. I'm
0:23 going to tell you. By the end of this episode, you're going to know how to use
0:26 Argon three ways. The first is vibe marketing. How do you use Argon to get
0:31 customers to your startup? I truly believe it's the best AI for writing
0:34 right now and I'm going to show you how to set it up to write just like you and
0:39 all the skills and everything. Second, we're going to talk about vibe
0:42 manufacturing. Argon is really good at helping you design personalized physical
0:46 products. I'm going to show you exactly how to do it, all the CAD files and
0:50 everything. And third, how can you use Argon to point it at legacy SaaS
0:55 products that people are still paying for and just build a better version? I'm
0:58 going to show you that. I'm going to give away some real sauce sauce around a
1:04 technique for actually getting customers there. I'm going to walk you through how
1:07 I would do it. And I want to give a quick thanks to Google for sponsoring
1:11 today's episode and giving me early access to Argon. Everything you're about
1:15 to see I actually built and these are all my personal opinions. And if you
1:19 stick around to the end, I'll show you six hacks to get way more out of Gemini
1:24 for Argon than most people will figure out for months. So, let's get into it
1:28 and I'll see you at the end. >> The startup by [singing] the spot cast.
1:32 >> [music] >> Okay, things to know about Google Argon.
1:35 What Gemini 4 Argon Does Well
1:40 What is it amazing at? Well, it's amazing at marketing with AI. So, that's
1:44 important. We need a tool that actually brings you customers. I'm going to show
1:47 you how to do that and where it's good at it. It's good at physical AI, aka
1:52 vibe manufacturing. So, how do you take STL files and send it to bamboo and 3D
1:57 print it? I'm going to show you how to do that on demand using Argon.
2:02 And it's good at complex engineering. So, I think that you can point it at,
2:07 you know, very legacy SaaS products, add your own spin to it, uh pick a
2:15 niche, go even nicher, and it could rebuild it relatively quickly. And it's
2:21 it's this isn't vibe coded software. It actually it it is tested, it feels
2:26 really good. Um and I've just been so impressed with how you can actually
2:29 point to a complex engineering. I will I want to get this out of the
2:33 way. Anti-gravity is not my favorite harness. Um I think that Google has some catching up
2:39 to do with Anti-gravity, and the more I played with it, the more I I was missing
2:43 some of the other harnesses. Now, you could still use uh Argon in your cursor
2:49 or in other harnesses, open code, and stuff like that. Um I just wish that
2:56 Anti-gravity uh would catch up, and I think it will catch up. And I think that
3:00 and you know, one one reason I think people should actually pay attention to
3:04 Argon is Google has so much data, and they're making so much progress, and now
3:09 Rebuilding the Late Checkout Website
3:09 they're at the frontier level in terms of a model. So, let's talk about
3:13 marketing with AI. Uh and what I first want to talk about a website me and my
3:21 team built uh in 2021. This website took 40,000 hours to build. And it's a a website
3:27 that was built uh for my holding company, and it just has all the
3:30 different buildings for all our different companies. Here you see LCA,
3:35 our design agency, Boring Marketing, um which is our marketing with agents
3:40 company, careers, all this stuff, right? I can't tell you how much money this
3:44 cost, but I can tell you that, you know, the business has probably brought in
3:51 more than $40 million of revenue. Um and a large part of it has to do with this.
3:54 It has to do with this website and people always tell us how good their
3:56 website is. I asked Argon to recreate the website. Um, or actually I just said, "Hey, I
4:06 want a website, um, like this, but can you make it better, basically?"
4:10 And it created this. And it's first of all, beautiful. I actually gave it the inspiration for,
4:16 uh, I was like, "Make it more Mondrian style, Mondrian the the artist." And I
4:22 think it did such a a good job. You know, when I compare it versus the one
4:28 that probably cost us $200,000 to build, I mean, yes, this is really nice, but
4:32 it's it's it's sad in the sense that there's no cars, there's no movement.
4:39 Um, and the web, the internet has moved, right? And we want things that are more,
4:43 you know, that you can interact with. And you can see here, there's this like
4:47 world passport. Um, when you, you know, click on the different buildings, uh,
4:52 there's some sounds, and like it it gamifies going checking out all the
4:57 different building, which what you want people to do when you want to go to
5:01 their their website, right? Um, and I just think, uh, there's all these like
5:06 fun little games and things that you can do, like shoot a three-pointer.
5:11 Uh, boom, boom, boom. This is crazy. It is crazy that you're
5:16 able to create this. Um, and when I say marketing with AI, or
5:20 by marketing, what I mean by this is why are you creating a normal website in an
5:25 Argon age? You know, it it doesn't make sense anymore, right? Um, cuz you want to
5:32 stand out. Um, and if you could, you know, how do you stand out? Well, the
5:37 limit isn't the model anymore, the limit is now the imagination. Cuz when you
5:41 have a model this good, that with an engine this good, then, uh, you can
5:47 create stuff like this. Now, my first prompt looked like this.
5:51 >> [laughter] [gasps] >> So, and I first was like, "Hey, make a
5:56 better version of like checko.studio .studio" and made this. And I think we
6:01 can all agree that this is pretty brutal. This is really brutal. You know, it got
6:08 the concept, but it really felt like me You know when Mark Zuckerberg like
6:11 announced the metaverse and it was like that that like Zuck in in 3D and like
6:17 everyone made fun of him. Um we'll put that photo up there. It
6:21 felt like that. Um and I think the question and the big delta in this whole AI era is like how
6:28 do you how do you go from this all the way to this, right? What is the
6:37 way to do that? And one thing I learned about Argon specifically is it needs
6:45 good references for good creative. I don't believe that you can get great
6:49 creative out of Argon unless you give it references. So, it actually in in some
6:54 ways it's kind of a great cuz you don't want to outsource all your creative
6:59 thinking to AI cuz then you end up creating stuff that doesn't really stand
7:04 out. I think this it has the power to it, but you do need to like, you know,
7:08 give it give it that uh those references. Initially, it created like all these, you know, way
7:14 too many buildings and the buildings were everywhere and I was just like,
7:18 "It's too much stuff." and you just have to like have the creative eye to be
7:23 like, "It's too much stuff." Um another another example is like the initial
7:27 version felt like Roblox-y. Like it didn't feel premium. And we weren't
7:34 there yet. All that to say is this this was a suc- suc- success and one of the
7:37 reasons why frankly, I don't update this website is it's really hard to update. Takes a lot
7:44 of time, energy, and stuff like that. Whereas this, now I can just use Argon
7:49 Justine Moore's Apple Computer Museum
7:50 to to to update it. I saw a really cool example of someone using Argon. Justine
7:58 Justine Moore A16Z said, "I asked Gemini for Argon for an interactive museum of
8:02 Apple computers and didn't provide any images or data. It scraped archival Mac
8:07 photos, wrote a tool to clean the images, pulled 50 years of specs, and
8:12 built a 3D cover flow with UI that morphs to match the decade you're in."
8:17 So, when you look at this, look how good this is. Um there's just so much opportunity to
8:25 whatever it is you're building, not to build something basic.
8:29 And that's what I mean by marketing with AI, is that this would would have cost
8:34 you thousands and thousands of dollars before, but now you just, you know, you could
8:39 just go and create this. Um and this is the sort of thing that
8:44 goes viral on Twitter. It goes viral on TikTok. It goes viral on Instagram.
8:49 Um and then you can decide, by the way, what what business you want to create on
8:52 top of this if you don't already have a business, right? You can be like, "Oh,
8:56 you know, I want to go and I've got this audience now of people into Apple
9:00 products." Well, what can I sell those people? Is it services? Is it uh memorabilia?
9:09 We'll talk about at the We'll talk about physical AI. What what are the things
9:12 that, you know, you can do later with Argon there, but um I think that's a
9:18 really one big thing I learned that I wanted to share with with you all.
9:20 Using Argon for writing
9:24 The second thing I I learned and is Argon is really good at writing. So,
9:31 it's extremely good at it. And I'm going to let you in on a little tip here, how
9:37 to make Argon an incredible writer in one step. So, you say, I want you to
9:41 research everything about how I write. This is what I wrote. My name is Greg
9:44 Eisenberg. I write on X. I do email newsletters, how I speak on YouTube,
9:47 everything about me. I want you to internalize all that stuff into a skill.
9:51 Once you've done that, I'm going to test you on it. I'm going to give you a topic
9:54 and I want you to write an essay on it and we'll see how you do.
9:58 And what it does is um it uh basically creates the skills and
10:02 what I should have done is actually asked it to do slash subagents.
10:07 Um to send subagents to go and do that. So you can just go slash subagents on um
10:17 on on on a antigravity to go and do that. Um but it what it does is it
10:21 created a skill package. So it created a skill MD file with the core voice
10:25 operating system, sentence rhythm rules, mental models, the structure I use is it created a
10:31 voice and lexicon MD, like signature vocabulary, things I've
10:36 coined. Um and it's just all it's got all the stuff about me which is crazy.
10:42 Verbatim samples and templates MD um plus all the structural templates. This
10:46 is something that if I you know, I was hiring like a head of content, they
10:50 would be able to do. And they did it perf- perfectly. It internalized exactly
10:57 how I how I write and speak. And this is just so cool cuz I find that one thing that AI isn't that
11:05 good at is writing, you know, in it's crazy cuz it's so good at other things,
11:09 but I don't know why it hasn't been such a good writer. And Argon to me at this point is
11:17 probably the best frontier model for writing. If you look here, I basically said write
11:23 me a post on marketing engineers. And it wrote me a post and it was
11:27 amazing. It was like, you know, exactly it's kind of it was actually crazy cuz
11:31 it was exactly how I would write. The old world versus the new world. You
11:35 know, talking about micro products instead of lead PDF lead magnets.
11:39 Things that I've said in the past, tweet-size landing pages. You know, I
11:42 have this belief that launch a landing page initially not more than 280
11:46 characters and stuff like that. It had all this stuff. Now, if you don't have
11:53 content online about you like me, just literally upload it to your anti-gravity
11:58 and have it learn about you. Like you've written stuff before, right? So, you can
12:03 have it do this. I really like this 955 rule. AI does the plumbing, humans bring the
12:07 taste. And what's really cool is it even signed off I signed off my newsletter at
12:14 gregeisenberg.com with be well, Greg and it had that. So, if I like if I hired someone to
12:19 write this newsletter, I would have been like 9.2 on 10 probably.
12:24 I would have been stoked. I would have been like telling my co-founder, "Hey,
12:27 you're not going to believe this writer I I found." And so, I think that, you know, from a
12:32 vibe marketing perspective, marketing with AI perspective, you know, using this to write more
12:37 um should be on your list of things to do. I don't know what you're using to write.
12:43 If you're just doing just humans um or if you're doing some, you know, mix
12:51 of Astra or, you know, Opus um but in terms of what I'm seeing uh as a
12:56 writer, as someone, you know, who who puts out a lot of content there,
13:00 uh this is definitely something that I think is is quite good.
13:03 Physical AI and Vibe Manufacturing
13:06 Let's move on to physical AI. Now, a lot of you have been asking me about vibe
13:10 manufacturing. You know, I had this tweet that went viral around vibe
13:14 manufacturing and how you this is the age to actually go and create these
13:19 files called STL files, send them to 3D printers, and build businesses around
13:23 customized physical products. But how can you do it? Well, I'll today
13:29 I'm going to show you how Argon is actually amazing at it. I actually asked
13:33 Argon to build me this vibe manufacturing studio. So, this is a big
13:39 idea here. I hope I hope people take. So, my belief is that you could be able
13:44 to customize physical products. I think it's going to be a bigger a bigger and bigger thing. I
13:49 think the idea is there's millions of Shopify stores, right? What if you can
13:54 embed this widget that allows people to actually customize products on Shopify
13:59 stores. And they could buy it for, you know, I don't know, $39, right? So, you have
14:06 here, you can see how long it's going to take um to uh how much how much it's going to
14:12 cost for the weight, how much is bamboo, print time is going to be
14:15 uh and by the way, if you don't know what bamboo is, it's a 3D printer. I
14:18 think it's a Chinese company. What suggested retail would be uh one
14:23 printer overnight, you know, and it takes 8 hours for four units. Um I think
14:31 this idea around uh basically embedding physical AI customizers on Shopify is an extremely
14:38 huge idea. Um now, you might be asking, "Okay, like what what would you actually create?"
14:45 Well, you know, think about Let's talk about the um I don't know. Your uh snowboard company.
14:54 I know Toby from Shopify started with a snowboard company. It was his first
14:57 Shopify store. Maybe you create, you know, famous hills, famous mountains, and you
15:04 allow people to customize it. Um maybe you create, you know, famous uh
15:09 snowboarders and allow people to customize it. Maybe you could put your
15:12 own uh face and and your own body in there, and you allow people to customize it.
15:19 When people customize things, they feel more attachment to it. And
15:24 gifts in general, by the way, physical gifts and these niche cues I just see seeing
15:29 more and more of a thing. And you know, one example here is an
15:35 espresso portafilter porto portafilter station. Um you know, I would pay for this and
15:42 you can see the demand signal it goes through Reddit, goes through R the sub
15:48 red R slash espresso who spent $2,500 on machines and gladly pay $49 for custom
15:54 portafilter stations. Just makes total sense. And I think this is like a big business.
16:04 Argon nailed this. You can see here, you know, it it went initially
16:09 I asked it to to create wireframe, show the layers and then show the CAD. Like
16:14 when I see this it's almost like an indescribable magical feeling that
16:18 you're able to create this. And there are so few people doing
16:22 physical AI. Um so worth playing with and I was impressed with how well
16:32 Argon succeeded here. One of the cool things you can do is see this export
16:37 ready to print STL file. Well, that STL file can send to your
16:43 bamboo printer if you own it. If you don't own one, you can do Slant 3D print
16:49 on demand API. So I think a lot of people are going to be like "Oh, I don't
16:53 want to own like 20 3D printers in my house or my office." Okay, don't. Just use an API and and
17:01 have it print there. You you literally all you have to do is aggregate demand,
17:07 come up with an idea that folks are going to resonate with, but you know
17:12 your niche better than other anyone else, right? So you have that insight
17:18 and then you go into Argon and you basically say like I basically was like
17:22 hey just build this embeddable Shopify widget so so I can show people. It was
17:29 literally a one prompt and then start testing it. See if your audience
17:34 likes it. Uh get feedback. What what do they wish you know you can have but I
17:38 think this idea around niche Qs physical AI is just getting started.
17:41 Made-to-Order Shopify Store
17:44 How I would actually like get to work on this so you have Argon build a product
17:50 customizer, you embed in your store. So the customer picks like a size, a color,
17:54 you have a live 3D preview so they can see what's going on. The customer pays on Shopify. I just
18:01 start with Shopify, they're the number one, right? Um course you could do it in the future in
18:06 other places. You check the custom print file so you've tested size limits and then you
18:12 print and ship the product. So it could be your printer or partner uh or
18:18 partner. By the way, you can also just sell this as a service
18:23 to other Shopifys. So you can actually just set up a business like here's a
18:27 business idea for you. Like just go and do this for other companies. Um
18:32 how you know, you go to them and say like I could help build more fandom for
18:38 your audience by having physical products so it's on their desk, they see
18:42 it every day. It's when they walk into their homes, they see it you know,
18:45 greeting them and they're going to be thinking of your brand and you're going
18:51 to make 50% margin, 40% margin, 70% margin depending on how you price it. Is
18:55 that something you'd be interested in? And maybe split the margin with them,
18:59 make it a no-brainer offer. These are things that you can now do. So the third
19:01 Building Boring SaaS
19:05 way I want to talk about where Argon really excels is building SAS products.
19:13 So boring SAS products. Now, I already hear the comment section. People are
19:17 going to say, "Well, Greg, SAS is dead. What is the point?" >> [sighs]
19:22 >> Well, I take a deep breath for that one cuz I see that on X all the time. People
19:26 saying SAS is dead, AI can do SAS. Therefore, you know, it's it's it's over. Those are
19:33 Indie Hackers selling to Indie Hackers, but the vast majority of
19:38 uh people uh are fine to pay for software that helps their make their lives easier.
19:46 Now, here's the way I would approach building a SAS product, and I think it's
19:51 different from what how other people are talking about it, and then I'll show you
19:56 how Argon goes and does it. So, first, go and find a recurring,
20:01 ideally boring, problem. You can find these problems on g2.com,
20:06 Capterra, Idea Browser something I built. By the way, has a It's free literally free to
20:13 sign up to get a daily idea. Um and then from there, I would go and
20:19 build a very focused problem. Sorry, I would build a a very focused product. Um
20:25 you know, you can you can be looking on G2 and on these other places,
20:30 um even Reddit, to go and see what are people complaining about and then
20:34 obviously put that into your product to make something better. There's so many
20:40 legacy pieces of software that are just you know, they're legacy, they're old,
20:42 they haven't been updated in so long. So, there's an opportunity now to
20:46 rebuild something that really fits the needs of people that maybe is actually a
20:51 little bit more niche uh in a good way, right? Um and then you go and make a useful free
20:59 tool. Now, a lot of people are go out when they they think about building SAS,
21:04 they think, "Oh, I'm going to go and build something, charge $50 a month, and
21:08 I'm going to get 10,000 customers. I understand why that's the approach,
21:11 but I actually think that getting a free tool out there is such a
21:16 good way to get it in front of likely buyers, then make your second or third
21:21 or fourth tool paid and just try to think about what's one tool
21:27 that can go and attract a lot of people and then from there you can go and
21:31 essentially create a product studio around that ICP. And your your studio
21:36 could be powered by Argon because Argon has become such a good way to actually
21:43 build software that actually works. And when I say that is we've all vibe coded
21:47 stuff, but when it comes down to deployment, it hasn't been good or it
21:51 hasn't been secure. And for my testing, Antigravity has done a quite a good job
21:58 um of that. So, I'll give you an example. I gave uh I gave Antigravity the uh this prompt,
21:59 Commercial Cleaning Quote App
22:05 "Build a working web app for a commercial cleaning company that turns
22:08 site visits notes into a reviewed price proposal." This is a This is a focus replacement
22:13 for coding spreadsheet. The main experience should be creating a quote.
22:17 And then you give it uh customer complaints, source links, anonymized
22:22 workflow here. Um and if they don't have it, you you know, here this would be your G2, this would
22:28 be your Idea Browser. Um but if you don't have it, you can actually just say
22:32 go and find it and it should be able to find something. Um and then uh find and inspect the
22:38 project, explain your implementation plan briefly, then build it using the
22:42 existing stack where possible. Then you just you just write it all out clear
22:48 clearly. Um you say don't deploy it, send proposals or anything, and then
22:52 give an honest list of what remains unfinished. This format is the format
22:58 for building these micro SaaS's that helps get you uh you know, in the door with a lot of
23:04 these companies. When you look at what what was built, um, let me pull it up actually. It's
23:13 basically these like interactive margin, uh, calculators. When you look at what was
23:20 built, I mean, again, this is not the most beautiful thing. Um, and that's,
23:25 like I said, if you want to get beauty out of Argon, you really need to give it
23:28 references. And you got to work with it to get it to that point. It has the capacity to do
23:36 it. It just it for whatever reason, in my experience, hasn't gotten it on
23:41 prompt one. But where it nails it on prompt one is on the engineering and all
23:48 these little details that, uh, you can see here, uh, it does such a good job
23:53 at. I mean, I'm not going to bore you with all the boring, you know, parts of this particular
24:01 workflow, but this format, um, in terms of, you know, these quotes and stuff
24:04 like that, uh, I think is is a big idea. Um, one that could I can see people
24:11 doing. And if you're going and you're competing against someone that charges
24:15 money for this, and then all of a sudden you're like, "Hey, this is free."
24:19 Like no one talks about that. No one talks that there's thousands of these
24:23 like boring SaaS products that charge money, then now you can go and sell it
24:30 for free. Literally zero dollar dollars. In the past, I would be scared about, you know, vibe
24:36 coding something, but we've reached this point with Frontier AI
24:42 that you can build something like this. And it it works. And it it it it it's probably even
24:48 better than what the legacy products have. So, the question becomes then, what do
24:58 you do when you when you sell a free a pay thing for free. Once you built the trust, eventually you
25:03 sell them something. So, totally was impressed with how Argon
25:10 performed here and happily surprised. So, having played with Argon for about a week, I found out
25:12 6 techniques to get the most out of Argon
25:18 that there's six techniques worth using to get the most out of it. It's this
25:22 isn't going to be rocket science to to people who use AI a lot, but I will I
25:27 will say it's going to help. So, the first is plan. /plan, just agree on what to build before it
25:35 starts changing files. So, like for example, with my latecheckout.studio,
25:40 you know, example, I should have used plan mode. If I would have used plan
25:44 mode, it would have been a lot better. My first version would have been a lot
25:49 better. The second is sub-agent. So, you can do /sub-agents.
25:53 That splits independent jobs, so research and and checks can run in
25:58 parallel. So, the example I gave around the marketing agents
26:02 creating the different skill files for writing better, like in sub-agents, I
26:07 would have got better output, and it would have been less prompting
26:11 and stuff like that. So, sub-agents you might as well use it.
26:17 And it's basically this concept that you know, you're you're not giving
26:20 a a task to one thing and it has to go and you know, go into multiple different
26:26 directions. You're just basically saying, "Hey, you You're delegating."
26:29 It's just literally delegating to different agents to do different things.
26:34 /goal. One of the reasons why the latecheckout.studio ended up becoming really good
26:41 was I did /goal. You know, halfway through it, I was like, "All right, /goal, make this
26:50 Make it amazing." I gave it a reference image. And yes, it it does take more
26:58 tokens, but uh I notice that the output is a lot better. Uh so, for me it's definitely worth it,
27:03 and you have to think your you have to think about like the value for you, you
27:08 know, like think about how much, you know, we made so so many tens of
27:13 millions of dollars from that website. An extra several dollars in tokens is
27:18 worth it, um if if that's even what it cost. Um {slash} browser, you can actually
27:24 have it open the app and test the actual user journey. How cool is that? Um good thing to know.
27:31 Uh number five, {slash} learn. So, you can turn your corrections into re-
27:37 reusable project rules or skill. You might as well do that. Um so, you know,
27:42 you're going to have things that you're going to mess up on, and uh you're just
27:47 going to want to like have a better system, more skills, more project rules.
27:51 You're only going to get better output, so you might as well use learn.
27:55 And the last thing is initially, I was on medium. I turned it
28:02 up to high. Um high has been my my uh my my base on on this. Like, I've been
28:09 enjoying using high. I do find the output's just a lot better in general.
28:15 Um and you know, so I would say use high, like personally, like as the
28:20 default. Um but start on medium so you understand like how to use the thing.
28:25 Um these are just like six six techniques ha- to get the most out of
28:30 Argon. Um the biggest one for me being plan, sub- sub-agents, goal. I mean,
28:35 they're all kind of big, so they're all I mean, this is why I put them in here.
28:38 These are some of the biggest ways to get the most out of Argon.
28:39 Closing Thoughts
28:42 So, there you have it, folks. Gemini for Argon is out. Some of the things you can
28:48 do. I think we're entering this new era of frontier model where these these
28:54 models are just getting so darn good at doing complex problems.
28:58 I was impressed. Um I will say I was using Google AI Ultra, which is $100 a month.
29:07 Uh that gets you full access to Anti-gravity app and the CLI and the up
29:13 to 1 million output tokens. And thanks again for Google for hooking
29:20 that up. The $19.99 Google AI Pro membership does not include the
29:26 Anti-gravity app and CLI or Google AI Studio access. You don't get the 1
29:30 million output tokens. So, that's just more for like text generation or web search, little tasks,
29:35 but if you're looking to build, the Ultra is what you're going to want.
29:40 Uh and what can I say? I I you know, I I think I'll end with
29:50 you know, the the limit is really your imagination. Um and the hardest part sometimes is just
29:57 coming up with the ideas, finding the references, getting the inspiration.
30:04 Um and finding a niche or ICP that you're you're going to fall in love
30:07 with. I always say marry the d- marry the niche, uh date the product. You can always like
30:15 try different products, but you're going to want to love the niche to serve them.
30:21 Um had a lot of fun playing with Argon. Interested to see what if there'll be
30:26 updates to Anti-gravity. Um I hope that gets better soon and I'm
30:33 excited to see what happens in Argon uh or Gemini 4.1. So, let me know what
30:37 you think down below. I read every single comment. Thank you uh uh, for for coming along
30:44 this journey with me. Um, I'm here to serve you. I'm rooting for
30:49 you, and I can't wait to see what you