you2idea@video:~$ watch X3yfVo2oxlE [23:13]
// transcript — 298 segments
0:00 Justin Jackson: Adam, is there other ways you're using it? This
0:02 is actually the first time I've talked I've heard you talk about
0:05 using AI at all. I I couldn't I didn't know if you were, like,
0:08 for it or against it. Adam Wathan: I used to be more skeptical than I am now. And I
0:15 think the thing for me is I was seeing everyone talk about, like, one shotting things and, like, vibe coding things. And to
0:23 me, that feels like bullshit and is just not not interesting. To
0:28 me, it felt like, okay. I like programming. Like, why would I wanna not program? But when especially
0:37 when the agent mode stuff came out, I think I started to see
0:43 how it's more useful. Because I forced myself to download Cursor
0:51 and just try to basically build an entire project, which was
0:56 like this this course platform template that we released not
0:59 too long ago. Yeah. So it's like a Next. Js website that's, like, pretty dynamic and does a bunch of
1:04 different things. And I just forced myself to build the entire thing without typing any code and just making the
1:11 computer type the code for me just to really, like, throw myself in the deep end and learn, like, okay. You know?
1:18 Because in my head, I was like, okay. Well, if I have to tell
1:20 them exactly what to do, isn't it just faster to do it myself?
1:24 And it turns out, no. It's still not faster to do it yourself.
1:27 Brian Casel: Like, just the fact that it can it can build multiple files.
1:30 Adam Wathan: Yeah. And you can what it feels like is it feels
1:33 like having your feet up on the desk pair programming with someone next to you that you can just, like, talk to like a human
1:38 being that understands what you want them to do and they do it.
1:42 You know? Brian Casel: Yeah. And and also, like, there's a lot of the
1:46 skeptics will say, like, oh, but it's just gonna spit back, like
1:49 like, code slop that I can't use. Adam Wathan: Yeah. Then you're not being specific enough with
1:53 what you're asking it to do. Justin Jackson: So did you have to babysit it at all? Like
1:57 Adam Wathan: Yeah. Like, extremely, honestly, but not not in a it's still faster, you know?
2:03 Brian Casel: You're you're still approving every Adam Wathan: Yeah.
2:05 Brian Casel: Page and every change. So you're so experienced
2:08 developers can read the code that it Yeah. That it spits back
2:11 and you can improve it. But the other thing is that, like, this
2:15 stuff is improving so fast. I mean, I'm using today, I'm using
2:20 mostly Claude Sonnet four. Even 3.7 was so much better than the models from, like, six
2:27 months ago. Justin Jackson: Mhmm. Brian Casel: Like, it's it's not only better at, like, just
2:33 programming and, like, and just, like, knowing, like, the like,
2:36 the Rails conventions and everything, and and it's also better at reading my project. Mhmm. And like, the when it when
2:43 it creates when it generates new code, it first reads my code
2:48 base and understands how I've implemented certain features in
2:51 other areas and then implements based on that. Like, it it it is
2:55 really good at, like, adapting. And then you can take it to the
2:58 next level. Like, you can have, like, a day one with Cursor, like, just
3:01 fresh install Cursor, try it out. But once you're using it for six, twelve months, now you're getting into, like,
3:07 Cursor rules. Yeah. Now you Adam Wathan: really started doing that on the very first
3:11 day. Basically, anytime I found myself correcting something that
3:15 it did, I added it to a rules file to make sure I never had to
3:18 ask it to do that again. So just like little opinions in, like,
3:21 our tailwind projects. If it used, like, a bottom margin, I always use top margins instead. So add it to the rules file.
3:26 Don't use bottom margins unless you're doing this. Or when I ask
3:29 you to, like like, a common thing that I need to do, like, a
3:33 workflow so I guess, like, here's the other thing that I I kind of became acutely aware of when I started using these
3:38 workflows is just like how much of the program I do is not programming. It's just like grunt work. There's a it's like
3:44 90% grunt work. So Steve has an icon in Figma. I need to take that icon, bring it into my project. I need to
3:52 optimize the SVG so it's not big. I need to convert it into a
3:55 React component. I need to add props to it so that I can add
3:57 classes to it when I'm using it. And the workflow for me has
4:00 always been right click in Figma, go to this tool, svg0mg.com, paste it into the thing, copy the thing out of
4:06 that, go to transform dot tools, paste it in to the HTML to JSX
4:11 converter, copy the JSX out of that, create a new file in Versus Code, paste it in, name the file, whatever.
4:18 Now it's just like in the cursor chat, I just say, like, add this
4:21 icon to my project, paste the SVG code directly from Figma. And because I'm told it, anytime I paste in an SVG, please
4:27 optimize it. Please name the component files using this convention. Make sure they always have these props. And,
4:33 like, that's now it's just like a one sentence thing. And I wasn't getting any, like, mental stimulation out of those
4:39 steps that I was doing before. No. It's like grunt work. So,
4:43 like, to me, the the real unlock has been realizing that this
4:46 actually lets me, like, program more, not less. You know? Brian Casel: Like, dude, it it's like that we should be spending
4:56 our mental cycles on design. Mhmm. Designing the designing the structure, the architecture, the the user experience, and our
5:05 skills as experienced Tollstack designers, developers is now to
5:11 direct AI to carry out our design. You know? And we are still in touch with the code base.
5:18 We can still craft it however we want to. And we can craft cursor
5:22 to build Adam Wathan: Yeah. Brian Casel: In the way that we want to. But we that's where
5:27 that that's why I actually believe that the more experienced you are in this industry, the the better
5:33 Adam Wathan: A %. Brian Casel: You are in this new world. Adam Wathan: I've tried to use it with, like, things I don't
5:36 know how to do, and it's a fucking disaster. Like, I was I was trying to, like, prototype, like, a Swift UI app the other
5:43 day, which I've never done in my life, and you instantly fall
5:46 into that trap of, like, AI being, like, confidently wrong, but you can't tell because you don't know how to do it
5:52 yourself. But when you're so I I've until, like, I really this
5:57 really clicked for me, I always kind of thought the idea with AI
6:00 was, like, okay. This is gonna, like, let me do stuff I didn't
6:02 know how to do before. Like, total opposite way to think about it.
6:05 It's like, this is keyboard shortcuts on steroids for the things I'm an expert in. You know? Mhmm. It's like when I
6:13 started using Sublime Text, and I had multiple cursors and all
6:17 this stuff. And it's like, I can just take, like, what my brain
6:19 wants to put on the screen and make it appear faster. Like, that's what my relationship is with AI too.
6:26 Yeah. Brian Casel: Wow. For for people building new stuff, though, I I
6:30 I do think that there's an element of, like, now that I build with AI, I can be more ambitious about the types of
6:37 things that I build. I actually think that Instrumental Components, in the way that I built it and and and the
6:44 extensive functionality that's built into it would not definitely would not have happened without this style of
6:51 build. Adam Wathan: Mhmm. Brian Casel: Or or I would have built a much simpler
6:54 Adam Wathan: version of it. I found it useful to separate those two things, though. Like, the using AI to learn and using
7:01 AI to code as, like, very discrete tasks. Like, I've never had success trying to learn get Cursor to do something I don't
7:11 know how to do. Like, I just I get bad results with bugs that I
7:14 don't understand how to fix. But if I go into ChatGPT and, like, I'm trying to do this,
7:18 what's the right way to think about this? What happens if I do
7:20 this? Like, it's the perfect thing to just, like, ask dumb questions to. You know what I mean?
7:26 Brian Casel: I've been doing that pattern where I I I go to ChatGPT, which the nice thing about ChatGPT is that it has the
7:33 memory built in. Adam Wathan: Mhmm. Brian Casel: Like, it has all my past chats, it knows a lot about
7:36 me and my preferences. So I use that as like my strategist that
7:41 where I bounce ideas and like how would I go, how would I architect this? Yeah. But like like that, like, so I'll like
7:47 get the approach there. And then I might I might also double
7:50 check-in Claude. Like, are you are you suggesting the same concept or are there
7:54 differences? Adam Wathan: Mhmm. Brian Casel: And then when I'm actually, like, ready to build,
7:56 I come into Cursor and, alright. Now let's do it. Yeah. Justin Jackson: Is it similar for you, Adam? You've got Cursor
8:00 for coding stuff and then ChatGPT for the learning stuff? Yeah.
8:04 Adam Wathan: Yeah. Right now, although I do wanna try some other stuff. A lot of people on our team are using Claude code
8:09 and liking that a lot. I've heard a lot of people have been having better results with that in Cursor and candidly kind of
8:15 like appeals to my sensibilities as, you know, it's like a terminal based thing, and I kind of like that it's separate from
8:21 the editor. There's something about cursor. Maybe it's just that I haven't spent enough time configuring it
8:25 where it's just, like, there's six pop ups at the same time,
8:28 and I don't know which one hitting tab is gonna actually auto complete or, you know, there's just Yeah. All these
8:34 competing, like, completions and, like, AI helpfulness going
8:37 on that it feels a bit a bit noisy. Brian Casel: Don't use the, like, the inline, like, pop up
8:43 things that you see, like, oh, that overlay the Adam Wathan: Mhmm.
8:46 Brian Casel: The current file. I'm I'm all about the chat sidebar. Yeah. You know?
8:49 Justin Jackson: Interesting. Brian Casel: And I also think that that is a pattern that is
8:53 going to become super common across all apps. Right? So like
8:56 Adam Wathan: Everything every app needs to have a chat now. Brian Casel: They and they are. Right? Like, Descript Descript
9:02 now has a cursor like I think they call it, Underlord. Justin Jackson: Wait. Are they good? It's like it's Are these
9:08 are these sidebar chat AI chat things good? Because I've Brian Casel: I played with the Descript one yesterday, and it
9:14 it's not great yet. But it but, like, that's and and Figma is
9:17 now Adam Wathan: I haven't tried the one in Figma, but me me and
9:20 Steve always talk about this with Figma. Do you remember when
9:23 Figma announced their first, like, AI stuff? It was, like, at
9:26 their Figma config Yeah. Last year? It was all, like, press a
9:30 button and get, like, a mock up of, like, your UI and all the
9:33 designers, like, revolted against them, which to me totally makes sense.
9:37 Because when I'm working with Steve in Figma, the opportunities to use AI, which I think maybe it does a better job
9:43 at now. I haven't looked into But the obvious opportunities to
9:46 use AI is just when Steve is doing tedious shit that he shouldn't have to do. So it's like, okay. I just designed
9:52 this, like, sidebar thing, and I've got, like, three of these
9:55 links. And now I just want, like, seven more of them. And, like, you can do it pretty quickly, copy and pasting stuff,
10:02 whatever, but now you wanna put sample content in there. You
10:05 know, it'd be nice if it just did all that for you. Like, we
10:07 with the template that we were working on, I'm working on, like, a course platform thing. So I'm coming up with, like, a
10:12 fake course. And we come up with a concept for the course. And then I can just say, add, like, three more course modules
10:18 with about five lessons each that, like, continue Yeah. With
10:20 the same sort of theme and the same sense of humor. And it just
10:23 boom. You know? That would have taken, like, so much brainpower
10:27 before. Justin Jackson: Yep. Adam, for you, I'm just trying to think,
10:32 how do you think this is going to affect your business? Adam Wathan: Yeah. I I think it already is, you know. There's a
10:41 lot of, vZero is a competitor with our business at the end of
10:47 the day. But yeah, I'm not sure. Like I think we've been seeing
10:55 effects of it for like a long time already. There's a lot of things that do make me like nervous that people
11:01 want like llm. Txt versions of our documentation, right? And
11:05 I'm like hesitant to do that because our business depends on
11:09 traffic to the website where people are gonna be noticed that
11:14 there's products for sale, you know? Yeah. And if everyone is
11:17 just looking at a markdown file that's pure documentation with
11:19 no ability for us to like advertise that lives in their editor, you know, then we have no distribution anymore.
11:26 Yeah. Yeah. I'm not sure what my takeaways are from that really
11:31 or my conclusion from it. I think one thing is that I think it's gonna be more important than ever to have, like, an
11:35 audience. I Brian Casel: yeah. That that is, like, the conclusion that I keep
11:40 coming back to. It's like if if it's zero if if traffic if organic traffic from search is on the decline, the answer to
11:48 that is personal influence. Justin Jackson: Yeah. Yeah. Adam Wathan: And then yeah. And then the other thing for for us
11:55 is just to, I don't know. Like, I think if you're building a
12:01 tool that costs money and it solves a problem, then you are you still can find customers. You know what I mean? And I
12:10 guess that's like still true for us regardless. Although like we are like competing with AI in a lot of
12:16 ways. Although I've become a little bit more optimistic about
12:18 it over the past few days because I do kind of think like there's something to the experience of like browsing
12:25 curated designs on like Tailwind UI that is easier than blank
12:29 canvas trying to get AI to do something nice when maybe like
12:34 you don't even really know what you want. So we've been thinking
12:38 about, like, how can we leverage AI to just, like, produce more
12:42 stuff that has our fingerprint on it. Justin Jackson: Yeah. But directed by you and Steve. Yeah.
12:47 Exactly. Brian Casel: It's like the the thing that comes to mind now is,
12:50 like, yeah, you're you're saying things that, like, like, v z
12:53 vZero or our competitors, but, like, it's also, like, everything is using Tailwind now.
12:57 Adam Wathan: Yeah. And it's frustrating because everything on Earth is using Tailwind, and there's, like, nothing we can do
13:01 to benefit from it. It's purely, a burden in a lot of ways. It's
13:06 just like, I have like over a million dollars in salary obligations that pay people just to fix bugs in a free tool that
13:14 people don't pay me back for. You know what I mean? Justin Jackson: Yeah. Yeah.
13:19 Adam Wathan: Yeah. So it's kinda brutal. Like, I'm I'm very ready
13:24 for like my next act in a lot of ways, Justin Jackson: honestly. You
13:28 Adam Wathan: know? Like, I like, we're working on like adding
13:32 support to Tailwind UI or Tailwind Plus as it is now for like, adding JavaScript support to the vanilla HTML stuff, which
13:40 we never had before. Yeah. In the past, I would've it would've
13:43 been obvious to do that as, an open source library. And, like,
13:46 really though, I don't think we're gonna do that this time. I think it's just gonna be part of the commercial product and
13:51 you have to have a license to use the JavaScript because what
13:54 is the incentive to make it free? Just so that somebody else
13:57 can go and use it to build a tool that, like, competes with ours without having to do any of the work to maintain that
14:02 torturous, like, hard to maintain JavaScript library. Yeah. Yeah. We're still bullish on doing that stuff.
14:09 Neck I'm right now, I'm working on, like, a Tailwind course,
14:11 like, an official paid Tailwind course that, you know, where the
14:17 the, you know, unique value proposition is like, it's taught
14:21 by me, you know? So if you wanna know how I do things, like this
14:25 is where you're gonna learn how to do that. And it's a nice thing that we can sell to our existing customers because it's
14:31 a separate product and it sits on the same site that still gets
14:33 like 10,000,000 visits a month or whatever. So, you know I love
14:38 it. That's a good idea. Brian Casel: I think that makes total sense for for
14:41 Adam Wathan: And then yeah. And then on the side, me and Steve
14:43 are designing video editing software right now because that's, like, a pain point that we have, like, when we make this
14:48 content. You know? No way. That I'm getting pretty excited about
14:51 that. So Talk about something that's outside Justin Jackson: of your normal, like, what you know. You must be
14:57 having to Adam Wathan: learn a lot there. Technology wise, yes. But it's
15:01 kinda funny because it's, like, it's a very full circle moment
15:03 for me because when I was in, like, college and, you know, committing to becoming, like, a professional programmer, I went
15:10 in with, like, the thing I wanna be able to do is build my own,
15:14 like, audio editing tool to use instead of Pro Tools or Logic or
15:17 whatever. Because I just have so many workflow opinions about all
15:20 this stuff. Because that's what I was doing before is recording
15:23 bands for a living and there's just so many things that could
15:25 have been, like, faster. You know? And Steve has the same background. Like, he went to
15:30 school for audio engineering. It's fun to be like Brian Casel: I got I got a degree in that too. And and we
15:36 all ended up being, like, web developers. Adam Wathan: To get into, like, yeah, designing like software
15:40 for compositing video tracks and doing edits. Brian Casel: I do wanna hear about that. I I go back and
15:46 forth between ScreenFlow, Descript, and a little bit of Screen Studio whenever I'm doing videos. And I'm just always on
15:54 the lookout for, like, how can I remove the grunt work? Adam Wathan: Yeah. Like, how can I just get it to delete
15:58 everything except the good takes, which is always the last take, and you can tell if
16:01 Brian Casel: it's a transcriber? Yeah. Adam Wathan: I've seen
16:03 Brian Casel: Descript has that feature, but it, like, only works, like, 80% of of the time.
16:08 Adam Wathan: Yeah. Yeah. And even, like, captions. Like, do you remember, like, Justin I mean, both of you guys will
16:12 remember this, but, like, there was a time back in, like, 2015
16:16 where, like, we all felt guilty about our podcasts not having
16:19 transcriptions. And we were all debating what service we're asking people at MicroConf, like, who do you pay to, like,
16:25 transcribe your podcast and what's their turnaround time and
16:28 whatever. Like, that is such, like, ancient history, like, that
16:31 whole idea. It's just like an instant free thing that's perfect now. You know?
16:35 Justin Jackson: Yeah. It's I just did Brian Casel: a voice note an hour ago and it, like, my iPhone
16:39 transcribed it. Adam Wathan: It's insane. Justin Jackson: Yeah. So Okay. So here's Anyways To wrap to
16:45 wrap this up, I I'm hearing some things. AI Adam Wathan: is cool. Like AI. Threat threatened by AI.
16:52 Empowered by AI. We're really just, like, trying Brian Casel: to lay lay, like, the groundwork so that, like,
16:58 the when the LLMs read the transcript of this episode, like, we're We're spared. We're on their good side.
17:04 Justin Jackson: Everybody's grappling this with this in real
17:07 time. It's clearly paradigm shifting. It's clearly like the tectonic plates of the industry are moving. Everyone's trying to
17:17 figure it out. The the other thread I'm getting and this has
17:21 come up before, but, you know, Adam's working on this new video
17:24 editor. It it feels like looking at what AI could enable for your team
17:31 for you know, as a as a another bet during this time is another
17:36 thing that's happening. Like, people are saying, okay. Well,
17:39 what can we chew what can we take our unique expertise in into that we wouldn't normally be able to do and start building
17:47 some shit that might be the next act of our of our stuff? Adam Wathan: You know? Yeah. I think the important thing is
17:55 just do stuff and Yeah. Make sure you're not being, a lot of
18:04 Not swimming, like, yeah, up against the current. You know what I mean?
18:07 Like, make sure you're informed before you take that stance or
18:12 you're gonna find yourself a little, behind the curve, I think. And I think the best way to learn the AI stuff, like, I I
18:20 really learned a lot very quickly by just, like, doing that one project where I just Do
18:27 Brian Casel: a quick project. Adam Wathan: Let myself write any code.
18:30 Justin Jackson: Yeah. Adam Wathan: And I don't think that should be like the goal.
18:33 Again, like to me, I like writing code. I I think a a lot of the criticism that you hear, a lot of like the concern that
18:38 you hear from developers like, well, I don't wanna be a full
18:41 time code reviewer. Like, that's the part of my job I hate the
18:43 most. Like, why would I design things in a way to be doing that
18:49 all the time? But it's up to you, like, how you use it. That's one way to do
18:54 it. You can have things like asynchronously submitting PRs for you and reviewing those PRs and stuff like that. But that's
19:00 not how it feels when I use it. It just feels like it's doing
19:05 what I tell it to do. And I see exactly on the screen what I expect to see. You know?
19:10 I'm not really, like, reviewing the approach. I'm giving it the
19:14 approach, and it's just producing the characters Like you know?
19:18 Brian Casel: The the creativity and the craft, that that happens
19:21 up front. Adam Wathan: Yeah. Brian Casel: And at the prompting stage, but even before
19:25 that, it's like the the PRD, like the project plan stage. Like, that's where we as the experience, look, this is how,
19:32 this is what we need to build because this is what we want to
19:34 deliver for customers. Like, that's where all the creativity
19:37 is happening. Yeah. You know? Adam Wathan: Yeah. It's cool, man. I feel like I could talk
19:42 about it a long time. Did you Justin Jackson: have one more thing you were gonna say, Adam,
19:45 before we close here? Adam Wathan: Yeah. There's two things I'll say quickly. On the
19:49 vibe coding thing, I'm not interested in that for, like, professional work, but it is extremely cool that you can
19:57 build custom tools for yourself to solve little problems extremely quickly. So, like, the I I built, like, a workout
20:05 tracking thing for myself just so that, like, it would tell me,
20:08 like, what plates to put on the bar and, like, what my next set
20:11 would be and calculate a couple things and stuff like that. And
20:13 that I entirely I've never read the code. I just told it to make it for me I just gave it feedback based on
20:19 the UI. And I don't care because I don't care if it has bugs that
20:23 would be security issues or it's just like a private tool for me
20:27 that I was never gonna build, but now that I can just ask the
20:30 computer to do it and it magically does it, now I have it and it exists. I did the same thing for like an interval
20:36 timer. So when I'm doing things in like the gym and I just wanted an app that would just beep every fifteen seconds. So I
20:41 knew to like switch movements and I couldn't find one on the
20:44 app store, but like, AI built it for me in two seconds. So that's cool. Yeah. The other thing I'm excited about that I'm
20:49 starting to get into now is just building custom AI, like automation flows. There's this really great article by Torsten
20:59 Ball who works at Sourcegraph. It's called like how to build an
21:02 agent or something. And as you're reading it, whole time you're just expecting to
21:07 get to the hard part where it's like, oh yeah, and then like the
21:10 super hard thing that I can't do. But like you never get there. It's just you get through the whole article and you're
21:15 like, oh shit, actually, yeah, I guess I could do that. And I
21:19 don't know. It just gives me all sorts of ideas for like, Jack
21:22 McDade messaged me the other day and he was like, does Tailwind
21:25 have are you guys adding this new CSS feature to Tailwind? And I'm like, well, one day. It's like on my calendar to go
21:33 and do it right now. And, you know, he's a bit paralyzed because he's gonna add it himself, but he wants to make
21:39 sure he uses the same APIs that we're gonna use that when we
21:42 eventually do add it, he can just delete his stuff and it'll
21:45 just keep working. And it just had me thinking, man, like, could totally build a tool that runs on a server on a cron job
21:54 and kind of just like watches the same sources for information
21:58 that I watch, like these like super CSS nerd blogs and notices
22:03 new CSS features. And I could give it some context and instructions and do everything I can to make it easy for it to
22:12 design new Tailwind features the way that I would do. And they just wake up to a pull request once in a while that's
22:17 like, oh, hey. Here's a pull request for this new CSS feature, and it's named the same way you would name it, and it
22:22 works the same way you would make it work. And then I didn't
22:25 have to do it. And the Brian Casel: But that's that is where we're going and it's not
22:29 even Wow. It it like, that is possible today. Adam Wathan: That's a % possible today. And the best part about
22:34 it is if it sucks, I can just, like, say no and close it. Whereas right now, if a human being does that, I should be
22:40 like, dude, I'm so sorry. I know you spent like fifteen hours on
22:43 this. I don't have time to really review this right now. It's just like, it's a cold machine. If I let it sit there
22:52 for nine months, the AI isn't gonna get mad at me or think I'm
22:57 ignoring it or Yeah. You know, it's there's so many elements to
23:00 having like a soulless machine do work for you instead of human
23:04 beings. Alright. I gotta run. I'm gonna be late, drop my kids off at dance. But Yeah.
23:08 Justin Jackson: Thanks for joining us. Adam Wathan: It's fun.
23:10 Justin Jackson: Yeah. Thanks, man. Adam Wathan: Alright. See you guys.
$

"I've changed my mind on AI coding" – Adam Wathan (creator of Tailwind)

@JustinJackson 23:13 19 chapters
[developer tools and coding][AI agents and automation][content creation and YouTube]
// chapters
// description

"I used to be more skeptical than I am now. Especially when everyone was talking about 'one-shotting' and 'vibe coding.' But then I forced myself to download Cursor and build an entire project without typing any code, making the computer type the code for me. I had to throw myself in the deep end." All three of us (@AdamWathan , @JustinJackson , and @briancasel) give our honest takes on AI and software development, and how it's going to affect our businesses (@TailwindLabs, @TransistorPodcastin

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