Episode 2
· 28:15
So probably the thing I spend the most time thinking about lately is how we can take advantage of AI in our business and especially in the products we sell and it sounds sort of like cliche or like eye roll y or something but I use it every day for so many other things and I do genuinely feel sometimes like I'm still building a product that I could have built in 2014 instead of building something that couldn't have been built before the stuff that we have available to us now. So I don't know, think maybe the best place to start is just to talk about sort of my own journey with using AI in my own work and where I found it to be useful, where I found it to be not useful. So I don't remember when it first showed up like twenty twenty three, early twenty twenty three I think is when ChatGPT came out. Around that same time we were seeing things like mid journey and stuff. And ChatGPT to me seemed like immediately useful as a Google replacement, just like getting answers to questions and stuff like that.
Speaker 1:And we've all seen it get better and better and better every single month ever since then to the point where, I mean, I use it a ton every single day and mostly not for work, honestly. Honestly, like I use it a lot just as like a place to think, you know, and to get feedback on the things that I'm thinking about or yeah, it's almost like having a little journal that can like talk back to you. So if I'm trying to figure something out, can just like write it down and go back and forth and get some ideas. And I don't know, I use it all day. I think it's a wonderful tool.
Speaker 1:But I was very late on stuff like Copilot. Like I literally have never even used GitHub Copilot. I'd never really experienced proper AI programming until the age of agents, you know, but right at the beginning too, in 2023, when mid journey came out, that clearly was a very valuable product to me because you could just generate things that you otherwise would have had to like commission people to create for you. And you can generate so many iterations of things and try things out so quickly. Whereas if you have to have a human being do it, you know, can only do stuff so fast.
Speaker 1:You gotta pay someone for every single new idea, and you might not even know what you want. It just opens up the possibilities of how you work in ways that just didn't seem, I don't know, it really changed things for me at least. We use it for silly things like generating like fake album covers for fake bands, for fake UIs that we're making, you know, stuff that just wouldn't make any sense to invest our own time into. So that's been super interesting, but back to the coding stuff, I was very, I don't know, a little, I wouldn't say anti AI, but I was a little surprised how many people were excited about using it to program at first because I like programming and sitting in front of a computer and writing code and solving problems and thinking about, API design and stuff like that. That to me is like fulfilling, satisfying flow state work.
Speaker 1:Like I want to do more of that. And I think if you follow like DHH, like I think this is sort of the take that he has on AI too, like he likes to program, you know. And that's how I felt for a long time. Remember even tweeting at least once just saying how surprised I was that people wanna use AI to automate the stuff that they actually enjoy doing, you know? But fast forward a little bit, and eventually I did sort of figure out how to start incorporating AI into my work, my actual programming work.
Speaker 1:And the way that I use it is really just to deal with tedious monotonous crap that I don't wanna do, myself. So man, I don't even know what I can think of as an example. There was a situation I remember where I needed to reformat a JavaScript object into JSON, you know, and I could have done that with like multiple cursors and fucking around for two minutes, or I can just ask the agent to do it and it's done in thirty seconds and it's perfect. And, we move on. But once I sort of realized that like, wait a minute, like there's all these little things that I do when I'm programming that aren't really the programming part.
Speaker 1:They're just like tedious sort of mechanical things that I need to do to get to the next point where I can start thinking and making a decision again. That's when I really realized that AI was like a game changer for my workflow and productivity. And after that, I really started just like really paying attention to every single thing I'm doing in my editor and asking myself like, is this fun or is this just like tedious crap to get me to the next step where I can do something more fun? And now I use it all the time just to like op, I operate at like at a different level of abstraction when I'm programming a lot of time now, instead of typing out all the specific syntax or things, I work with an agent panel a lot and I'm asking it not to like one shot things, but to do like very, very specific things, like add a prop to this component for this, you know, and it'll add a prop to the component, it'll add the right types, all this little stuff that would have taken me a couple minutes to do. And all I had to do was sort of like express the actual change I wanted to make to the code.
Speaker 1:So in that way, it still feels like I'm the one programming, I'm just giving instructions that are much higher level than typing specific characters into the computer. Like one thing I've been doing lately, I'm working on some Tailwind UI template stuff. I'll take the designs from Steve and I'll just like build it out as static HTML without thinking about which parts should be props and componentization and stuff like that. And I'll just take the design and build it out, trying to match it exactly with just like raw static HTML. And then I'll do all the componentization as sort of like a second step.
Speaker 1:And that step I do like entirely with the agent. And I don't just say componentize this design. I go in there and I say, I highlight a section, add it to the context window and say, Hey, can you refactor these lists items into a loop and pass the items that are being looped over as a prop. And I'll just do these tiny little individual extractions and surgical changes, but I'm using the agent to do it. So I'm still doing all the work, thinking about the design, thinking about how things should work, but I'm not typing any of the code.
Speaker 1:And I use like the audio or sort of like the dictation stuff a lot for this too, which is really cool. But anyways, that sort of discovery of that workflow really sort of changed how I thought about AI because for a long time I felt like, okay, AI is this thing that's gonna like take away all the fun from my life, because that's sort of what it looks like when it comes to something like mid journey. And I'm sure artists have figured out how to incorporate this stuff into their workflow and allow them to still feel like artists, but superficially from the outside, it's easy to feel how like someone who spent all this time building up these artistic skills of being able to create art by hand and enjoys sort of like the meditative process of sitting there with a canvas and painting something or something would be pretty depressed by the idea of just being able to prompt a painting into existence instead of, you know, this tactile process that they're used to. But yeah, now I kind of just see AI in these situations more just like an infinitely patient assistant who will just sit there with you and do all the stuff you don't wanna do.
Speaker 1:And I think that's the most important thing is making sure for me anyways, making sure that I'm using AI to do the things I don't want to do, not the things I do want to do. And if you use it that way, it actually ends up feeling like you have even more time to spend on the things you do want to do than you had before. So instead of AI sort of taking all the fun out of your job, it's actually creating space for you to sort of like increase the amount of fun that you're having and sort of skew the ratio to more towards fun work and less away from tedious monotonous work. So bringing that back, I guess, to like how I have been thinking about using AI in our business. We sell like design to developers, right?
Speaker 1:What website templates and UI blocks and UI kits and stuff like that. When I see a lot of the things that are on the market right now that are trying to use AI to help non designers design things, for some reason, it just, it doesn't really do it for me. Like you see tools, like Lovable, for example, there's a million of these tools now where you land on a homepage and there's a text area, what would you like to build? And you just describe the thing that you want and it designs a website for you. And last week, especially I was seeing lots of interesting demos with like Gemini three, where people were creating some like pretty good looking stuff, honestly.
Speaker 1:But for whatever reason, just cannot get excited about this workflow of trying to describe what I want in a text area. And when I think about our target customer, which is like non developers who want their things to look really good, but aren't designers, I don't know that workflow doesn't feel like the right workflow to me. Like how can someone who's not a designer effectively sit in front of a text area and explain what they want in enough detail that the AI is gonna produce something good. You know, the reality is like, you're not a designer, so you don't know what you want and you don't know what you're asking for and you don't know what level of detail to provide and you don't know what details make something look good versus if they were missing, it wouldn't look good. And I've seen some cool videos of designers using AI to build stuff that feels sort of way out of reach to me in terms of what a developer could do, because they're sitting there with the LLM and prompting something into existence and then going back and forth with it, back and forth saying, change this, try this, change this, try this.
Speaker 1:And it looks like a really cool workflow if you're a designer because they can iterate really rapidly, but they are still designing. They know what changes they wanna make to make something look better. And they're just using AI as sort of like a means to try out those ideas faster. But if someone's not a designer, I just don't see how they can sit in front of the computer and have that experience because they don't know what they're asking for. And that includes me, someone who's like invested an insane amount of time into getting better at design and sort of trying to break down design into something that a non designer can understand.
Speaker 1:I don't have a lot of luck sitting down designing things, with LLMs. So when I see the products that are out there, you know, and that is the obvious one, that is the one that everyone is building. It doesn't really do anything for me. And I don't know what we can do instead, you know? So the thing I've been thinking about more than anything is trying to sort of figure out what is the magical workflow that I think a developer would want to have, when they're trying to produce a design, especially, you know, as a non designer.
Speaker 1:And I really don't think it's try to describe what you want in a text area. I could be wrong. I might be the only one who feels this way. To me that just doesn't feel right. It doesn't feel like I'm ever going to get the types of results that I want.
Speaker 1:So when I think about the way that I actually do work, it's often more by like looking for inspiration than it is by just sitting and like imagining things, you know? And I think this is maybe an interesting place for us to explore. But I think the problem is every time, like I think of a way that we could do something here, it ultimately just ends up being something that you don't even need AI to build. So I don't know, I'm kind of rambling here, but let me give you like a little taste, for example, of something that we're working on right now, which is we've sort of like made this bet that, especially with like the popularity of something like chat CN that competing in like the application UI component space is probably not the best place to compete because a button or a text input can only look so good. You know, like the ceiling for those components is at a certain point and eventually the free stuff and the open source stuff is gonna hit that ceiling and there's not gonna be anything that we can offer.
Speaker 1:So what we're working on more now is more of like the landing page marketing site side of things where there's a lot more designing that happens and the ceiling for design is a lot higher. So obviously we had tons of that stuff in Tailwind UI, but I had this thought that, well, man, I'm all over the place today, but the, the thing that's sort of always held us back with ToneOne plus and ToneJuly is just feeling like everything had to be this consistent style because it meant when that style went out of date, we had to go and update like 600 components. And it also meant that it's like hard to come up with new ideas because eventually you sort of exhausted all the patterns you can think of in a particular style. So when I was like analyzing our workflow for designing things, a lot of it starts with like dribble, you know, and poking around dribble, trying to find something that sort of just like gives you a spark of an idea that you can go and, and run with. And nothing on Dribbble is consistent.
Speaker 1:Like Dribbble is all over the place. Right. And that's like, by design, that's a good thing. So I started thinking, well, what if Tailwind plus could be more like Dribbble where instead of feeling like one big UI kit, it feels like sort of a place for us to put anything and sort of a place of inspiration. But the problem with trying to build like a Dribbble as like in a company is Dribbble works because it's like a marketplace.
Speaker 1:That's not a marketplace story, but it's, you know, it's a community of thousands of people submitting their stuff. How do we get to that volume of work without having a community where we're just doing everything ourselves? And this is when my mind's got sort of racing with like, okay, well maybe the interesting thing here is not using AI in our end product. You know, like maybe the thing people are buying from us is not the ability to prompt a design into existence. But if I really believe in this browse a giant gallery of stuff and find something that you love that's close to what you want and then remix it.
Speaker 1:And we need to somehow produce thousands of these things. Maybe the real thing we should be doing with AI is just figuring out how to use it in our own design workflow, where we actually do have designers at the company to just produce a lot more stuff than we were producing before. So, I mean, I think in another podcast, I'm gonna talk in more detail about the specific idea that we're working on now and sort of like the vision that I have for it. But that sort of realization was a bit of an unlock for me. I kind of think there's like, there's three ways to sort of like use AI in your business.
Speaker 1:Okay. So one is building an AI enabled tool, you know, so that's like building something that couldn't have been done before you had access to these LLM APIs where you could send off these fuzzy data and get back real answers, you know? So I think like AI customer support agents, you know, stuff like that is one category of like using AI in your business. And for a long time, that's where my head was. And honestly, it still is if I can find that idea, then of course I'm gonna get excited about it and explore it.
Speaker 1:But that's like one way to do it. Another way is what I was just saying, which is, okay, well, how can you find ways to use AI in your business that give you a sort of competitive edge or that you build something better that you can sell to people? Even if that thing itself isn't like a wrapper around AI, It was made possible because you had access to AI. So that's kind of what I'm talking about now. Whereas before we were sitting down designing a bunch of stuff by hand, and that's honestly still what we're doing because we're still trying to figure out what the hell is this workflow where we can even produce designs that we're happy with at a faster pace.
Speaker 1:To be clear, I'm not that close to cracking that. And it's something that we're actively investing in figuring out, but at least feels more possible. But that's sort of like the second way I see using AI is how can we build a better business sort of on, by leveraging AI ourselves internally rather than selling AI to our customers, if that makes sense. And then the third way I think is like, can you build, how can you take advantage of the fact that other people are using AI in their workflows to build things and sell to them that is, sort of like designed for those AI optimized workflows. And this is another area that we're thinking in.
Speaker 1:So instead of being like, okay, let's build like a tool that lets people create tailwind components with AI. Why don't we create the tailwind components for people who use AI? What does that product look like? How can we make something that works way better than anything we've made before for people by really just like leaning into the fact that people are using cursor and stuff like that all day to build things? How can we build something that's optimized for those workflows?
Speaker 1:People feel like they can produce things even faster. And I don't know, that's just an interesting mindset to get into as well. So sort of like realizing these like sort of three categories of like building with or for AI has been really helpful for me because I, for a long time, was really stuck in that mindset of like, how do we build something that uses AI under the hood and sell it to people, you know? How can we wrap AI in some interesting way? And I just, I haven't figured that out.
Speaker 1:And it's easy to feel like, okay, well, I can't figure that out, then we're just stuck building things the old fashioned way. And hopefully we don't get leapfrogged by technology and get lost in the pack here. But realizing that, well, you know what, like we could just, even if we just leverage it internally to produce more better stuff than we were doing before, that makes our product better than it was. And we couldn't have done that without AI. So maybe I don't have to feel so stupid as long as we can figure that out, and same for the category of building things that are sort of optimized for people using AI.
Speaker 1:So that's sort of like where we're investing right now, and I'm pretty excited about that stuff, just sort of trying to use it more and more in our own work, forcing ourselves to do it honestly, and building little internal tools. Like for example, something simple I was working on last week. Right now in Tailwind UI, like we use the same sort of placeholder avatar images as everyone else on earth. And I think people stole them from us for the most part, to be honest, because we went through Unsplash for a long time curating the ones that we liked and then cropping them in certain ways and using them in our products. And I noticed that even the stuff that Gemini produces is using like the exact same URLs to Unsplash that we are using with this very specific Imgix parameters and stuff that we're using, which is ridiculous, but whatever.
Speaker 1:Sorry, trying to think what I'm saying because I'm running into a lot of other people with dogs here and trying to avoid getting into a chaotic situation. Okay. But anyways, what I built last week with Philip on our team is I thought, you know what, like, I think that's one of the things that's making our product sort of look a little cheaper than it could look as people just see all these avatars that they've seen in a million places. So what if we use AI to like generate our own set of avatar images that are only ours that no one else has and no one else is using and our stuff will look a lot more custom. So that was like an interesting thing.
Speaker 1:And I started building a tool with Philip for generating these avatars where, you know, we are using AI to like define like 32 personas of people and then using AI to sort of define a ton of different profile photo styles, like whether it's a black and white profile photo, whether someone, a candid photo taken out a restaurant where they've got a beer in their hand, whether it's them on their couch with their dog next to them, or whether it's them at a baseball game, or it's a professional headshot taken against a backdrop and everyone on their team had the same style of photo taken. And we can generate hundreds of these avatars and then sort of go through and regenerate the ones that turn out ugly and turn this into like a little asset serving service that we can use for our tailwind plus stuff. So that's just like, I guess one simple example of a place that we're starting to use AI more to sort of elevate our own product. And I'm looking forward to finding other ways to do it too, like by generating interesting background patterns and graphics and stuff to use in Heroes.
Speaker 1:And I think my hope is that if we just keep being very deliberate about building with AI ourselves and trying to use it to make better products in their current form that maybe, just maybe we will stumble across something that feels like a product sort of just by chance. Because to be honest, like that's where I've had the most success in, in all of business. It's never been sitting down and like imagining a product and building the product. It's you're building something, you run into a problem, you build something that solves that problem. And you realize that there's a market for that thing.
Speaker 1:Know, honestly, that is what tailwind itself is. Never set out to create a CSS framework. I set out to build a payments SaaS, that handled sales tax, you know, but it needed a UI and I wrote some CSS to style the UI and people were more excited about the CSS I was writing than the app I was building. And now, you know tailwind is used by tens of millions of people basically like literally we have over 10,000,000 active versus code intellisense users on our intellisense extension And, none of that would happen if I didn't just build something else first. You know?
Speaker 1:So, yeah, I think that's the right thing for us to do is just be consumers and users of AI aggressively and just see what comes of it. But I do have some other ideas around how we might actually use it in a real product too. And maybe I'll talk about that later this week. I'll do another episode where I'll spill some of the beans on some of the specific new stuff that we're working on, how we're building it and where I wanna go with it. And yeah, see, see you guys think always open to feedback on this stuff.
Speaker 1:It's really nice to get this stuff out there and got a lot of good feedback on the last episode. So yeah, I've really enjoyed doing these and do another one a couple of days probably. And yeah, so catch you then.
Listen to Adam's Morning Walk using one of many popular podcasting apps or directories.