We Were Going to Shut Down Our Company…Now We’re Worth $700M | Eric Simons (Bolt)

So we were, you know, we were at the end of the year, we were going to start figuring out how to spin things down. And so we had a handful of ideas and like I'd mentioned, like, we had tried the idea for Bolt in February of last year and the models weren't good enough. And so in May we got an early preview of Sonnet and we're like this. Whoa. Like this changes potentially everything. This is really good. And so July 1st we greenlit it and then in October we launched it. And in the first two months alone, we went from 0 to 20 million to be RR in the first two months. It's just kept growing since then. And to buy knowledge, I mean, it's like by traffic and revenue, whatever. I mean, this is like the number one web based code gen thing in the world. And I mean in developer stuff, I think it's second only to like Cursor, which is crazy because it's been out four months or something. Like the speed here is wild. But yeah, I mean, we went from, you know, being ready to shut down the company, you know, to now something that's huge.

All right, welcome everyone. My guest today is Eric, the CEO of Boat. Boat is an AI coding platform that's exploded from 0 to 3 million users. And I'm excited to watch Eric demo building a mobile app from scratch with AI and also talk about how his overnight success was actually seven years in the making. So welcome Eric.

Awesome. Yeah, thank you for having me. I'm stoked to be here. Yeah. So shall we dig in? Should I start with just to dive right in, show a demo of this thing?

Yeah, let's do it, man. I mean, you just launched this product, so I'm super excited to take a look.

Heck yeah. Yeah. So for those who haven't like heard of Bolt before, it's like, you know, a text to app product. And so I'm going to go pull my screen here. And so it's really the simplest interface in the world. What do you want to build? And there's just an input box, you can type anything. And so we started with just full stack web applications. We still do that insanely well. And this week we just announced that we also now support building native mobile apps for iOS and Android. So you can go from idea to app store just by prompting. It's like pretty mind blowing. So I'll show you. Just. Maybe I'll start with one example, you know, make me. We have this like, build a mobile app with Expo. So we partnered with Expo. If folks are Familiar, but they've got like the best react native tooling and ecosystem in the world. So this, it's really cool. So there'll be a mo. I say is build me a mobile app that and I'll just say I'll start with like a simple one is a clone of Spotify, see if this works. And so this, you know the, this is the core experience of Bold here. So it kind of at this point kind of looks like chat GPT or something until this right side of the screen trying to comes in, we have a full developer environment that's like booted inside of the browser. This is what we spent the past seven years building was like this technology to like boot an OS inside of your browser tab. Right. But anyway, so it's getting to work. It's installing all the dependencies we need, it's writing all the code. It's going to go ahead and quickly spin up a clone of Spotify here. And one of the main, I think one of the main things people really love about Bolt is that we've done a lot of work to make the, the actual user interface it builds to be really beautiful. And so once this thing boots up you'll kind of get to see that. But that's, this is kind of the, the core of the experience and it took this very seriously. It's making a very sophisticated clone here of Spotify.

I, I mean I. One of the things as a beginner programmer is just like installing all the libraries. Such a pain ass. So it's nice that it's doing all that for you already.

Yeah, yep, exactly, exactly. That's like what I think, you know, especially like with both the bidirectional, you know, communication between the operating system layer and the AI agent. The AI agent has complete access to all of the processes and the runtime and all the errors that are happening. The state of the file system or whatever have you. So it can, when things go wrong you can really dig in and drill in. And so as you can see here, this is a native iPhone app here.

That's great, man.

Yeah, this looks like Spotify, right? It's pretty insane. How. Well, so you kind of get the idea. And this is just the ui, right? So we can keep prompting and building on top of this. And so this is like, I think someone that, you know, people use Spotify or music asking you kind of get an idea. I mean that, that's zero shot, right? I mean like that's just, it's just off the dome. Just whip that thing up so let's maybe start another app. And we were kind of talking vivo for this, like, what would be kind of a cool app to build. And so, you know, I was mind blown to find out that, like, calm the meditation app is worth like $3 billion, you know. Yeah, it's kind of wild. So it's like, you know, you want to start a $3 billion company right now?

Let's do it. Let's do it, man. Yeah.

All right, so I'll build you a app that is, you know, a simple meditation app. Have it be, you know, beautiful and minimalistic. And so this is like, you can, you know, obviously you can like, you know, create like a PRD type dog where you like, really outline what you want, if you want to be specific. But a lot of times it's just, it's fun and you actually get really great results just by kind of giving your general idea and then throwing vibes at it. You know, like, what does it mean to be beautiful and minimalistic? What did I find out? Yeah, you know, this is like, I guess like harpothe is. You got the vibe coding thing. It's. Yeah, I mean, it's kind of true. I mean, you can really go to this thing and, you know, give it, you know, what the sort of feel you're looking for.

If like a super genius like Kabarthy is doing it, then, you know, like, I have no shame in just like vibe coding too, like, totally, if you want.

But it went out and grabbed MP3s for these things and, you know, if some of them work or not. Okay, well, it tried its best to find MP3s that would work, but yeah, that's the general idea.

This episode is brought to you by Vanta. Whether you're a startup founder navigating your first audit or a security professional scaling your GRC program, security has never been more critical or more complex. That's where Vanta comes in. Businesses use Vanta to establish trust by automating compliance needs across over 35 frameworks like SoC2 and ISO 27001. Vanta can help you start or scale your security program by connecting you with auditors to conduct your audit and set up your program fast. Plus, with automation and AI throughout the platform, Vanta gives you time back so you can focus on building company. Join over 9,000 global companies like Atlassian, Quora and Factory who use Vanta to manage risk and prove security in real time. Get $1,000 off@vanta.com Peter that's V-A N T A.com Peter for $1,000 off. Now back to the episode.

Let's try to redesign the UI a little bit. Let's try to maybe like just do like, just like a Google image search or something.

Yeah, let's do it. Okay, so maybe like, like get a screenshot of calm.

Yeah, like the calm timer thing or something.

Yeah, yeah, let's do it. Let's see what it can do here. There we go.

This, that.

I like that ui. That looks cool.

Yeah, yeah, yeah.

Okay, let me, let me go ahead and save this and then I will go ahead and over here and I will drag her in. I'll say, you know, make the timer look like this ui. Let's see what it does. It'll get to work. It's doing it.

I mean, I feel like these AI cogent tools, like both are like really exceptional at front end stuff, right? To do like. Yeah, look at that. Oh, nice. It knows to expand it. That's funny.

Let me go to like response. If you get a little bit bigger. But yeah, so it's like going in and out. I can then prompt it to, hey, make that circle. You know, make the breathing smaller and go and do that. But yeah, I mean, you know, it took that because before there wasn't even a circle. I mean it was like. There's like a timer countdown, right? Like this thing, it just kind of ascertained. Okay, yeah, like this is. You got to start the breathing thing and that's going to like. It should probably go like that, right? So it got that from like a screenshot. I didn't, we didn't say anything about that.

Yeah, you need to expand the circle.

Yeah, exactly.

And, and how would I actually deploy this to the App Store? Or maybe like, does it already show that?

Yeah, so the way that you deploy something, I said now it's like fitting properly within the viewport and it's doing its thing. It's got the timer countdown out there. Pretty. I mean, to make a mobile app like this, I mean, I. This would, you know, getting X code and all the stuff. Stuff would be insane. I don't know how half this stuff works. I don't know. Most of it works. Right. And yeah, so if you wanted like, you're like, hey, this looks really great. I want to get this to the App Store. So we actually. Up in the top right here, you can actually, you can build a web version of this if you want and then you can do instantly. That'll be online when I go to the App Store. You can click deploy to app Store and that takes you over to Expo. So what Expo does they. All the tooling that you're seeing there, like running this app, etc. That's all their stuff. Right. They also have this thing called EAS that is basically like their fast track to actually like building these things and getting into the App Store. And so for both Apple, Google, you have to sign up for individual accounts, right. To go and actually do that. But then EAS you can run from like command line, whether it's like locally or in Bolt and you can actually say take this Expo react native project, they will bundle it up and then using your App Store credentials, they will publish that to test flight or actually into App Store review. But it's like it takes like you know, out of five or ten minutes maybe if you already have all this stuff set up like you're, you register on Apple or Google and then you just link EAS with those like it's wild how fast it is to actually get deployed. I think the Expo folks actually made a video going from like idea to App Store with the launch of this. I think we're doing a live stream with them I think like next week or something. But yeah, they've done a phenomenal job, you know, of making this crazy easy to actually take these things to the App Store for.

Yeah, I love how you kind of partner these other platforms like Expo for deployment to App Store and also Supabase for databases. Right. To extend both functionality.

Yep, absolutely. Yeah. So up here if we wanted to add a suit based database, we can click. So if we want to have like user login and let people track their meditations or share it with Frazzer, I don't know. Right. Yeah, you can build full stack web and mobile apps with this. And yeah, like there's, you know, Expo is great at mobile, Supabase is great at database, the hosting. Like they do a web build of this thing that's using netlify to do those things. So we've got great integrations with the best companies in the world that specialize in this stuff just built in out of the box.

And how about if I like, if I want to refer a change or like can you like walk through some of that?

Yeah, yeah. So like this one, right? So you know I said make the breathing circle smaller and so right here on every message there's a, the most recent one there's an undo button and on all the other ones you can say roll back to checkpoint.

Right.

So if I, if, let's say we didn't like this change. I could just, you know, hit revert and what'll happen is it's now we have the, the big circle back again.

Oh, that's fast. That's like almost instant.

Instant. Yeah, because it's. We're checkpointing every single message turn and so then it would basically just calculate the diff, apply it and then you know, the dev server picks that up and recompiles it, you know, usually within milliseconds.

That's great, man. Because yeah, I was using some other ID the other day and like, you know, I had to use like GitHub and stuff and like, you know, do all these commands. Yeah, it's just kind of a pain too.

Yeah, it's like a huge pain. Right. And often with these, with, you know, the reality of using these AI, you know, prompting based, you know, text to app things is like the AI is just going to do stuff sometimes it's.

These AI cogent tools, like both are like really exceptional at front end stuff. Right? That they're like, yeah, look at that. Oh, nice. It knows to expand it.

Like response if you get a little bit bigger. But yeah, so it's like going in and out. I, I can then prompt it, say make that circle, you know, make the breathing smaller and go and do that. But yeah, I mean, you know, it took that because before there wasn't even a circle. I mean it was like there's like a timer countdown. Right. Like this thing, it just kind of ascertained. Okay. Yeah, like this is. You're gonna start the breathing thing and that's gonna like. It should probably go like that. Right. So it got that from like a screenshot. I didn't, we didn't say anything about that.

Yeah. And need to expand the circle.

Yeah, exactly.

And how would I actually deploy this to the app store or maybe like is there already show that or.

Yeah, so the way that you deploy something, I said now it's like fitting properly within the viewport and it's doing its thing. It's got the timer content out there. But yeah, I mean to make a mobile app like this, I mean I, this would, you know, getting X code and all the stuff uploaded would be insane. I don't even know how half this stuff works. I don't know. Most of it works. Right. And yeah, so you want like, if you're like, hey, this looks really great, I want to get this to the app store. So we actually up in the top right here, you can actually, you can build a web version of this if you want and then you can do instantly that'll be online. When I go to the App Store, you can click deploy to App Store and that takes you over to Expo. So what Expo does they. All the tooling kind of you're seeing there, like running this app, et cetera, that's all their stuff. Right. They also have this thing called EAS that is basically like they're fast tracked actually like building these things and getting it into the App Store. And so for both Apple and Google, you have to sign up for individual accounts, right. To go and actually do that. But then EAS you can run from like the command line, whether it's like locally or in Bolt and you can actually say take this Expo, react, native, native project. They will bundle it up and then using your App Store credentials, they will publish that to test flight or actually into App Store review. But it's like, it takes like you know, out of five or ten minutes maybe if you already have all this stuff set up like you're, you register on Apple or Google and then you just link EAS with those like it's wild how fast it is to actually get deployed. I think the Expo folks actually made a video going from like idea to App Store with the launch of this. I think we're doing a live stream with them, I think like next week or something. But yeah, they've done a phenomenal job, you know, of making this crazy easy to actually, actually taking these things to the App Store.

Yeah, I love how you kind of partner these other platforms like you know, Expo for deployment to App Store and also Supabase for databases. Right. To extend both functionality.

Yeah, yep, absolutely. Yeah. So up here, if we wanted to add a, you know, suabase database, we can click. So if we want to have like user login and let people, you know, track their meditations or share it with Fraser, I don't know.

Right.

Yeah, you can build full stack web and mobile apps with this, you know, and yeah, like there's, you know, Expo is great at mobile, Supabase is great at database, the hosting. Like they do a web build of this thing that's using netlify to do those things. So we've got great integrations with the best companies in the world that specialize in this stuff just built in out of the box.

And how about if I like, if I want to refer a change or like, can you like walk through some of that?

Yeah, yeah. So like this one, right? So you know, I said make the breathing circle smaller and so right here on every message there's a, the most recent one there's an undo button. And on all the other ones you can say rollback checkpoint. Right. So if I, if, let's say we didn't like this change, I could just, you know, hit revert and what'll happen is it's now we have the, the big circle back again.

Oh, that's fast. That's like almost instant.

Instant. Yeah, because it's, we're checkpointing every single message turn. And so then you would basically just calculate the diff, apply it and then, you know, the dev server picks that up and recompiles it, you know, usually within milliseconds.

That's great, man. Because, yeah, I, I was using some other IDE the other day and like, you know, I had to use like GitHub and stuff and like you do all these commands. Yeah, it's just kind of a pain too.

Yeah, it's like a huge pain. Right. And often with these, with, you know, the reality of using these AI, you know, prompting based, you know, text to app things is like the AI is just going to do stuff. Sometimes it's going to take what you say. It's going to like kind of be, you know, do something that, you know, is following your instructions but like, isn't. And you're like, oh crap, I need to roll that back and like change what I, what I actually want. And like, you know, and so what you need to have this sort of, you know, having that be a fast process where you have to like dig around how to like check out a specific command or, you know, whatever. Right. Is a pretty key of, you know, the experience, you know, to allow to actually use these things.

Well, okay, so do you have any, like, can you give me like your top, you know, three to five tips for actually getting the most out of both, like beyond just like viable coding. Like, what are some tips to actually.

Yeah, if you're like really building something, if you're really, if you're building like a, an app or you're not just like, if you like what, we're just having fun here. Right. But if I really wanted to like take this to production, I mean, a lot of this stuff you're sitting down and kind of planning out like, hey, what am I looking for? Right? Like having a clear idea of, okay, what should the different, different pages be? How should they like, like what should generally be on them? How should it work? Like, what exactly I'm looking for. Because when you come into the prompt here, it's like coming in with that information and you can, like, you actually list out the different things you need. Has a huge, it has a huge swing on this thing's ability to actually make an app that's going to really fit your, your requirements. And then with that, you want to break it into concrete. You don't want to like, just throw an entire PRD at this thing or whatever. You want to like, break it into concrete tasks where like, for those who are an engineer, right, it's like the same way you would start out. If we wanted to build this app, what we do, we would build out just building the ui. We would be like, okay, does the breathe functionality work? Okay? Yes, it does. And this, like, generally looks good. Okay, let's go to this tab. Does this. They look and function, you know, like, without that core business logic, does the skeleton of this thing actually, like work? Yes or no, Right. Once you kind of get that done, then you go, okay, now let me actually add, Let me add my database. Let me add authentication. Like, let me go, like, let me go get the actual MP3s to play this thing from. You know, all those sorts of things. Like, you want to start adding the logic in once the UI is complete because you're kind of just delineating then like what the AI is focusing on doing. It's not focusing then on like UI stuff, it's just focusing on, hey, when that button gets clicked, you're going and you know, making. You're going and like registering the user in Supabase, right? Versus if you were saying, okay, make me a login page that has Supabase fired up. And like it's. Sometimes it can do that, right? But it's when things go wrong, you have to like completely revert. But, you know, some things are working for that one prompt. It's like you can't granularly kind of roll back that. Okay, no, no, that just that one part wasn't quite right, you know, because all this stuff is combined in a mega, you know, commit basically from the AI, you know, so that, that would be kind of my recommendation. This is like anyone, you know, building products in general. It's like, have a plan and then, you know, kind of like really think through what you want. Break down, you know, those into specific chunks. Start with the UI first. Get the UI to really be great, and then break out the chunks of. Okay, now like, let's make the login page actually log you in. Let's make the registration page actually register you for account, you know, and then like that sort of thing.

So you want to, you do want to run like a PRD or maybe help get AI to help you write a prd. But you don't want to. You don't want to feed the whole thing into at once. It would get confused. If you want to do it one by one, that's what I say.

That would be my recommendations. And again, you can actually get good results by just feeding the entire thing in and zero shot. It'll like, get it all right or most of it right, you know, but it's. It, you know, it's. You can try that. And if it doesn't work, then you can just say, okay, let me just start from scratch and just kind of break it out piece by piece then. Right? But, yeah, so, but generally it's. If you do the work of having a prd, which is like just a fancy way of saying, like having a plan, how each page works, how the functionality will work, it'll allow. It puts kind of, you know, bumpers on what the agent is going to actually go and do, which. Which is just really helpful for you to actually get to the end zone and have a great app really quickly, you know. And how about.

I think you probably stop sharing now. Let's chat about stuff. Yeah, but how about, like, when I get stuck in this, like, you know, doom loop, right, where like, he was making bugs, like, so, so obviously you can just hit the button, like, fix the bug. But what if that doesn't work?

Yeah, good question. I mean, this is like, this is kind of the problem with all of these things is that inevitably you will. You'll end up. And I think if you, like, try and fix a bug, hit the fix bug button like more than three times in a row with us, we go, okay, like, let's take a step back. Like, you know, can you describe what, you know, what you're seeing? Can you go and, you know, do some ancillary research on what is happening with this, you know, et cetera? I think so today that that's like kind of the, you know, the best way to get unjammed is like to, you know, go like in our discord and say, hey, I'm running into this thing, you know, or Google and see if there's information, you know, that you can find that'll be helpful. The kind of the next stuff we're rolling out. That's cool. We announced this one already, which is. It's called like the Bolt Builders, but basically, you know, oftentimes you just kind of need Someone who's you know, like an expert, you know, on development stuff to like step in for like anywhere from like 10 minutes to like 30 minutes, right. To just come in, help figure out what's going wrong, you know, and then, and get you on your way. And so we're actually rolling out a program like in bold. There's going to be a button where you can say, hey, like connect me with someone live that can like go and punch. It's like, you know, there's all this stuff, people are like AI is going to replace developers. Like no, it's not like, it's not like developers are more important than ever, but they just get to focus on solving hard problems instead of just, you know, slamming out HTML and css. Right. And that's kind of one, one aspect of something we're rolling out and then the second one is just there's been a lot of work being done on reasoning models that are, you know, making a huge difference in how effective debugging can be. And we were beginning to roll this out now by default in Bolt, where when you say like when an error occurs and you say, hey, fix this thing, it's we're going and doing, you know, multiple different passes across different models, I mean different reasoning models to get like, you have sort of like mixture of experts sort of opinion on okay, what is.

Oh thanks.

Yeah, what's actually happening here. Right. And solving it. And that stuff is pretty effective. So I think over as new models come out and they keep improving, et cetera, it's like, you know, the errors that you run into, where you need to raise your hand for developers is, you know, going to keep going down and approach zero. Right. But yeah, that's kind of the, it's a two pronged approach with like technology and people, you know, this is how it's done.

So if I keep running the bugs on boat, you actually tell me to calm down, like maybe show the calm screen. It's like.

Sit here and meditate for a minute. Just, yeah. Quick and then you can keep prompting. Yeah, yeah, it does. I think in the mirror messages, hey, like it says like, hey, like listen, like you've hit, you've hit the fix button a couple times. You know, you might want to go and here's our discord, like see what's going on or like you like blah, blah. Yeah, I mean, because it's like people just kind of sit there and just keep hammering it, which is really frustrating. And for people that haven't done a lot of development before, understandably it's like there's a button that says fix it so why wouldn't I click it right? Or like attempt fix or you know, and so it's just, you know, you keep hammering that thing but the reality is like, you know, often there's some context that's missing and you know, especially as if it's like context that's specific to how your code base is set up, which these things end up, you know, all specializing. At some point or another you need to have that additional context either come in from another human, from a Google search you did or something, right? Or like whatever have you but. Or you have a reasoning model that's like going and you know, thinking hard about it and it's like, oh, maybe that's it, you know.

So yeah, I think it's really smart to build the reasoning stuff into Bolt because like normally what I do is I copy the code to like ChatGPT or something, try to get to solve it, but like, you know, it doesn't have all the context of all my files.

Exactly. Yeah, exactly, exactly. So the early stuff we've been doing on that, it's very effective at really improving the debugging experience because it has full access to the context.

That's a key differentiation. Man, if you guys saw that. Because I find with these Cogen tools it's kind of like the startup curve. You start like, oh this is so amazing. And then you get stuck in this loop. It's like, oh, this thing is not working. And then eventually, hopefully people figure it out and get back on the path.

Totally. I mean that, that is, that, that is the challenge for us. And all the other stuff in this is like, you know, and I think the thing with the Frontier model, they're, they just keep getting better and better, right? So I think like one year ago we tried building Bold and it wasn't possible because the models at the time just weren't good enough. And now here we are, right? It's like now you can actually like, you know, zero shot web and mobile apps, you know, where like generally like you can build, you know, meaningful stuff with these things and you will run into errors with any of these tools. But there are errors where it's like today you just have to like go Google stuff and whatever have you. Tomorrow it's like it, these things are just not going to make those errors. Right? Period. These entire classes of problems are just not going to be problems. And when I say tomorrow, I mean like literally like in the next couple of months. I think we're going to see some new models potentially coming out that just, you know, really our step function improvement over this stuff. You know, it's nice to kind of.

Work on a startup where like, you know, you have your engineers, but you also have this like these giant companies spending billions of dollars making your product better, essentially. Right, Totally.

Yeah. So I mean that's, I mean now that you know that for us to, you know. Yeah. I mean, spoil wasn't, you know, when we launched in October and which could you start taking off like rocket. It was like, you know, we were, you know, it took us about a month to kind of like really get our bearings straight of, of kind of what was going on here and. But it became pretty clear like, okay, this is. Yeah, Cogen is. It's gonna. It makes sense because software is deterministic. Like it either when you write code and it executes, either runs without errors or it doesn't. And so I compare that to like if you had to make an LLM that's like good at law is subjective. Like, it's like, what how did the judge feel about it? Like, what was society's general feeling at that time about that? It's like, that's not deterministic. Right. And so it's hard to do reinforcement learning, you know, on, on that type of pride that's like most problems, right. Kind of are those sorts of things. Software is not. Software is not that. And that's what anthropic, I think really identified. They were like, we can make the best coding models by just doing a ton of reinforcement learning on these things. And so I think we're going to. It makes sense that they are getting better faster than kind of any of these other areas. Right. That the LLMs could be used for and that it's probably only going to accelerate.

I want to mention one other use case that most people might not know about with Bolt. So it's actually pretty good for like when my designer makes a mock up and then I want to explore some variations.

Right.

And then before I had to be like, hey, you know, please, can you please explore this variation for me? Here's why. But now with Bowen, these other tools, I can just put their design into Bowen and be like, hey, can you make this like, you know, two columns or can you try this? And then I can show them what it looks like almost right off the bat.

Yep, totally.

So like that's probably powerful, I think, for a pm.

Oh, huge. Yeah, it's huge. It's huge, right? And even for like, you know, what you're saying is actually one of the main use cases that we see companies, PMs and designers and entrepreneurs using us for is like, this isn't like, hey, you have to build your entire company using this app sort of thing. We have startups that are launched to make full stack apps and making money doing it right. But when you look at existing companies, especially where they have existing code bases, the value of this is instead of today, if you want to make a high fidelity prototype, right, typically you're going to do that entirely in Figma, where you can click between these static screens and that's like a lot of work to like lay all that stuff out, right? Whereas if you can just like take your design system, your different components, you dump that into Bolt and then you just say, okay, made me a web app or a mobile app that has this page with these requires this page that hit enter zero shot, probably comes back with something that's pretty usable. You can refine over a couple of prompts. Boom. And that's like a real app. Like you say, here's an iPhone app. Take a screenshot of this QR code, try it end user, what do you think? And actually feel it, right? Actually like, actually not like, hey, how do you think this would feel in this Figma screen? It's like touch the mobile app, you know, it just like, how does it. And that's like, you know, way faster for the product development, product discovery process, you know. And so I think that we're, I think we're, you know, we're already seeing a lot of companies start to do this where they're just doing all of that stuff in Bull, you know, like the high level designs, like the design system, whatever, in Figma. And then they're taking that and just putting it in Bull and that's where they're actually making their prototypes of stuff, you know.

That's amazing, dude. Yeah. It's easy for me to generate a real app with both and to try to fit it with auto layout or whatever. Figure components.

Yeah, exactly. And that's. But before, it's unthinkable to go to the code prototyping stage first. Because that's expensive, right? To have developers spending time just whipping up, it's like, that's unthinkable. But the equation is completely changed now where it's cheaper and faster to have a real coded prototype. And that's crazy. Yeah, it's amazing. It's amazing.

So Eric, I want to switch gears a little bit and talk about Your company building journey. I think most people don't realize you spent seven years building this company before Bolt came out of nowhere. Can you talk about basically right before you guys launched Bolt, what was going on and what happened afterwards? This whole thing?

Yeah, I started this with my co founder Albert. He and I have been childhood best friends since we were 13. We grew up down the street from each other in Chicago and he and I learned how to write software when we were 13. And we've been building web apps and stuff ever since, products and startups and whatever. And the core insight that we had of starting Stackless, which is like the company behind Bolt, and it was the first product of this company, stackless.com was like a web based IDE. The insight we had was actually it kind of, it came from like when we first came to Silicon Valley, we had the good fortune to bump into Dylan Field and Evan Wallace when they were starting Figma. And this is back in like 2012. And so we got to see kind of the story play out for those guys with what they were doing. And back in 2012, like their first like seed round or see that they raised or whatever. They didn't even have, they didn't have a design tool. They had a demo of a 3D ball dropping into water inside of a browser tab. And the pitch was basically, browsers have gotten really powerful. They can now do complex 3D graphics rendering with WebGL and the predecessor to WebAssembly. Because of this, we think that we can make a design tool. We're gonna have to create a new rendering engine for WebGL or whatever. But like if we do that, we can bring design to the web just like Google brought docs and we'll change everything. And obviously we know how that story, you know, played out for them now. And it was, but it was this classic, like deep technology, you know, browser play. And so in 2017, 2016, 2017, we saw the same deep technology browse plate emerge, but for development. And so specifically There were some APIs that landed, like WebAssembly and Shared Memory and service workers. And we were like, oh, well, like for the first time ever, you could actually write an operating system in webassembly, like booted inside of the browser tab. So it boots in like super fast. Like a hundred milliseconds. Could be like a couple of megabytes in size. You'd have to write it from scratch. But if you could do that, you could run full web development tool chains in the browser. So you like use the web to like build the web and. Which is kind of crazy, you can't do that. Right? Like, that seems like something that the browser should be able to like have built in capability and like build. Windows has Visual Studio, Mac has xcode, the web has nothing. So that was the origins and really it was like we wanted to build the tool that we wish we had when we were 13, you know, learning how to code. And heck, for everything we built since then, where it's just, you go to, you know, it's going to spend, you know, days, weeks or months, like at big companies it takes like a month just to get onboarded to the code base and have your environment set up locally. And like, it would be great if you just open a browser tab and it just works. You have to even install like thing. Right?

Yeah, makes sense.

Yeah. So that was kind of the core, the core bed of the company. We got that technology to work. It took us like four or five years to really nail that. And but the big challenge we ran into then is like developers had no interest in leaving their local environments. And so a lot of developers were using the product for free and it was like, you know, widely adopted open source and stuff. We couldn't figure out the monetization aspect. And so, you know, last year we were, you know, the company's kind of on the rocks because we were like, we got to figure out how this like, like commercially make sense to make money. Yeah, yeah, to make it, to be a company, to be a. When you raise venture capital, you have to, you know, you can't just sit around and, you know, just make cool technology. It has to actually have real tangible business value that can grow quickly. And so we were, you know, we were, at the end of the year we were going to start figuring out how to spin things down. And so we had a handful of ideas. We're like, hey, here's different things we could try that would be, you know, really interesting. And like I'd mentioned, like, we had tried the idea for Bolt in February of last year and it. The models weren't good enough. And so in May we got an early preview of Sonnet and we're like this, whoa. Like this changes. This changes potentially everything. This is really good. And so July 1st we greenlit it and then in October we launched it. And in the first two months alone, we went from 0 to 20 million of ARR in the first two months. It's just kept growing since then. And to my knowledge, I mean, it's like we buy traffic and revenue, whatever I mean this is like the number one web based code gen thing in the world. And I mean in developer stuff I think it's second only to like cursor, but which is crazy because it's been out four months or something. Like the speed here is wild. But yeah, I mean we went from, you know, being ready to shut down the company, you know, to now something that's now something that's huge.

So what's the secret? Because there's a lot of competitors, man, there's like V0 and like Replit. Like what was the secret between, you know, scaling to from 0 to 3 million in like four, four months?

Yeah, yeah. 0 to 3 million users you mean? Yeah, like ye. I think, I mean all the technology we've been working on, like booting the operating system in the browser thing, that's called web container, what we made. And that is really the, that's the unique differentiator because all the other tools in the market, when you go to them and you say, hey, make me an app, they have to provision a cloud VM to do that. And so often the free tiers of these things are very limited because otherwise people are going to be mining Bitcoin or hosting phishing sites or whatever because someone's got to pick up the bill for that cloud vm, right? You can hit all the networking that happens there. So if they're ddosing people with it, like that's on you as the operator, right? And the experience as an end user, just the latency is sucks. Like every keystroke has to be synced to a server, synced back. And so it's really hard to manage these things because like as we saw like, you know, like on the demo I did, it's like sometimes these things will just like get into a weird state and you need to like just reboot it, you know. How do you do that when it's like it's a cloud VM that the end user doesn't control and you have like, you know, millions of people using this. How do you do that? How do you do that reliably? And I think that's the challenge a lot of people are running into. And whereas with Bolt, all the compute is done locally on the user's device. Just like how Figma works, just like how Google Docs work. Like when you go to figma, they don't spin up a cloud VM for every user that's using the product. Oh they just, it's running locally on your device. And that's the same thing that's happening with Bolt, like we're not having to spin up a VM for every single user. So we have like a very permissive free tier. It's a great experience because there's no latency. It even works offline. If you disconnect your Internet, you can keep like editing code and stuff and it still works. So that's like the big. It's kind of like an under the hood difference that you wouldn't like, you wouldn't see it. You know, I mean like, even if you're technical, a lot of people are like, what this is like not using the cloud V. I mean, no, but when you. It's that the things that you notice in the experience is that it's faster, it is more reliable, like you can hit refresh and get back to a clean state. Right. Things are wonky or whatever. Right. You know, there's not, there's. Perhaps it freights here because, you know, there's. You don't have to worry about people mining bitcoin. Some people have come, like we've seen in our analytics, they're installing bitcoin miners and they're sitting there and we're like, I don't think you know what's happening. Like you're using your electricity to mine the bitcoin.

That's funny. Anyways, so, I mean, yeah, from an end user perspective, they don't know anything about cloud VMs. But like the, the speed and the simplicity I think is like, for me at least that is like the differentiator. Right?

Yeah. And the reliability too, because a lot of these other things, it's just, you'll get disconnected. There's some like kind of cryptic error mirror message that you go digging and be like, oh, this thing ran out of memory. And then. Yeah, I mean, some weird thing like that.

Right.

And then and that. And anyone who's worked in dev environments locally, this stuff breaks all the time. It just. They're for a million different reasons. And so that is really the challenge is like if you want to provide a very reliable service, you know, and that's. If you want to grow big, like have lots of people using your products, you have to scale reliability in a very, you know, a very direct way. It's like, why is Chipotle so. So great? It's like you can go anywhere in the world with a Chipotle and that burrito is going to taste like the same. You know, it's like, it's Right.

Okay, so Chipotle.

Absolutely. Yeah. Heck, yeah. Yeah. You know, like that's, you know, I've been. Or insert your favorite, like you know, large, you know, like great, you know, sort of company of choice. Right? Yeah.

I also like how you made the system prompt open source. Like what was the decision behind doing that? You know.

Yeah, unconventional. It was like a very uncomfortable one. But we were, you know, but we were like, as we were building Bolt, we were looking out just, you know, like in the open source world, I mean typically the open source was like a ton of great stuff. You know, where it's like, okay, this has got framework reference on this, blah, blah. But for this stuff there was really not a lot. And so, and the more we thought about, we were just like this just doesn't seem, it doesn't seem like there's a real strategic advantage here. Like, you know, for us. Like we were like if there is going to be something that differentiates us, it's like web container. It's like that in browser. The experience of that. Right? Yeah, the system prompts, you know, every, everyone figures out how to get around the system prompt. So you like protect yourself. Everyone's going to find out anyways. And on top of that like we're, we're going to open, let's just open source like the core of Bolt. Like let's just open source the first version of this thing so people can play around with it and like, like, you know, let's see if you know, we've been doing a lot of work in open source for the past, you know, five, seven years here. So it's like there's these tangible benefits that are kind of ineffable but like, but are very real. Where people adopting your open source products really is, that is a tide that raises whatever boats your company has. And that is exactly what we've seen happen with Bolt. We're like, we're the only folks in like the web based AI cartridge and the, that have anything open source meaningfully, much less like the actual product. Like you can run both like locally or host it yourself, whatever with whatever model. And that's been amazing because you know, for our system prospect everyone's. The open source world is like tweaking these things. They're like finding out stuff we didn't know. I mean it's basically the world is getting to collaborate and do R and D on what are the best models for Cogent. Like what are the best, what's the best way to tune this system prompt? You know, like what are the different things we can do? And like even Product functionality, let's integrate Supabase or this. Like that stuff landed in our open source version before we did it. Yeah. And so like, and so with the developers there, we've got a great relationship with all the people that have showed up to work on the thing because we share our knowledge of the stuff that's working too. And we're like, hey, this is, you know, this is kind of what we found. And then, you know, it ends up in the open source product. Yeah. When these new models come out, it's like, because our team is like 20 people. You know, it's like we have a, you know, 25 as of this week. But like, you know, it's not a lot of people and, and we're busy just keeping our servers online and just, you know, that's that we're hiring as fast as we can. But I mean, you know, to really deeply evaluate all the new models and stuff that are coming out is gargantuan, like challenge. And so the open source world being able to actually do that and, you know, glean those results, not just for us, but for everyone, is amazing. It's a degree where like, you know, the Frontier AI labs are also using Bold, it's called Bolt diy. Like they're using Bolt DIY to actually test out their new models and say, okay, because how else are they going to get this into a real production app where they can, you know, rapidly test it? There's not many.

So that's brilliant. Dude. When I build products, I have a discord too, and community and people give me feedback on the product and my engineers try to build stuff, but the open source is basically, people are actually building the product for you to a certain extent. They're kind of building the product with you.

Yeah, totally. And it's cool. And people will take this back. They take parts of Bolt DIY and they put in their own products and stuff. And that's the thing, right? It's a situation where the pie is growing for everyone, you know, and I think that a lot of other, I think the natural tendency, if you're starting like an AI company or whatever, it's like we gotta like kind of control our thing. It's like our secret sauce and stuff. And you know, like, we need to increase our valuation. You know, it's just like all these like, things, it's like, ah, you know, but, but the reality is, you know, you know, if you actually believe like, hey, this is not a winner takes everything scenario. And in fact more people contributing is just going to increase the speed in which this pie is increasing for everyone. Then it's like okay, well and let's be the best place that gives out free pies. Like just to let everyone actually contribute and play around with this stuff in a real way.

I mean the pie is still very smart. Right. The number of engineers, coders out of everyone in the world is like tiny.

Yeah. Especially this is enabling people that have never been able to code themselves to actually build. And the people that have, they're like non developers taking Bolt DIY and trying out the other. You don't have to be a crazy developer to swap the models in and out and they get the stuff running locally and so people are doing that and then, and they're these people the, they understand like a PM or a designer. Like they're going to have a better perspective on how good is this model at making a product zero shot than like an engineer typical one, you know. And so they're actually pulling the stuff down too and trying it out and be able to be part of the discourse that's happening. Which is really important because they're, you know, this is that their knowledge set and their skill set is really key to how products like Bolterman and the AI models actually are getting trained right. On you know, good data. So anyway, so it's very cool to see the ecosystem that spawned here and I'm glad that we, I'm glad we made that decision because it was like in the Ale butter. We're like we may regret this, we might regret this, but it seems like it would be a good idea.

You know, it's just like how FIGMA has, has like a Figma community. Right. Like there are people making plugins and stuff. It's like amazing ecosystem. Yeah. I also like the reading your system prompt because it kind of confirmed because you have stuff like very important or like never do this. This is exactly how I try to prompt you to do this stuff. It's like never.

It's true. Yeah. It's surprising. Our folks have chatted a lot with the AI. Yeah. And this is blessed recommendations from that. Yeah. I was like wow. Okay.

Yeah.

Because I mean when you see it, it comes from a programming perspective. It's like that seems like like when you go like Z index 999-9999 to get something to show up above it's like it feels like that sort of solution to it. But that's like actually, you know, when you're doing prompt engineering they having these kind of in exclamation in Caps like, you know, very important, critical that it actually makes. It has a difference, you know, in the words and the capitalization, whatever. So.

So let's wrap up with how you guys are building Bolt internally, how you guys are building this product. So you have 25 people. Do you have any product principles or values or beliefs in building Making Boat great or building these AI products?

Yeah, good question. I think for us, we want to build a product that is great for developers, but is also great for people that love building products and websites and web apps and so like PM designers, entrepreneurs, et cetera. And so I think it's kind of interesting where balancing those, like, there. There's not a lot of. There's not a lot of prior art or products that have been able to balance those because it's all prior to AI Cogen. Right? I mean, it's not. This is new territory and so it's kind of tricky. You have to like, you know, for us, we want to take bets because, like, there's only so many bets we can take with the size of team we have, we're scaling. But even then, like, we still. It's still a finite number of things we can sink our time into. And so it's like one of the things that are going to be really great for both audiences. You know, it's like I. The analogy that I often like, kind of lean on is like, it's like a power drill, you know, like my mom uses a power drill, you know, to install stuff at home. Right. Most homes in America, right. Have a power drill. Right. But so do most construction sites, you know, like, so do most professionals who build, you know, structures. Right? And so I think that the trick is like, how do you build a tool that's like going to be really. That's gonna be able to really cater to different types of people and different types of use cases. And, you know, that's powerful, but it's bounded and is, you know, gives you like, reliable results. And yeah, I think that's how we think of the tool and the sort of bets that we're making. And I think the other thing is like, we, we expect the AI code gen, like the AI models to get better quickly. And if you look at kind of what happened before Sonic came out, I mean, there's a lot of, like a lot of the other guys in the space for text to app or whatever they existed, and they had all this stuff built out around rag pipelines and like, there's all this info they'd spent a year building or Whatever. And then Sonic came out and then it just obliterated all that. Like Sonic zero shot was better than all the stuff that they had spent, you know, however much time and money making. And so we kind of look at that. It's like, so there's like a tsunami wave that hit, but it, but we expect more to happen. And so how do you like make, how do you make bets right in product and engineering when you're on the beach? Tsunami just came. There's probably another one. You probably don't want to like build a whole village, right? You don't want to like completely, you know, set up a whole bunch of infra. That's going to get wiped out. Like, you probably want to like get a really good surfboard, you know, and kind of prepare for the wave that's coming in and set up structures that some can be durable, some are designed to like be like, hey, we, this can and will get wiped out. And some things you just don't build because you're like, it's too expensive to invest or we think it's that it, that this is going to happen. So that's kind of the, the multi dimensional, you know, sort of analysis and like kind of bet placing that, that we're kind of looking at at least. Right. And which I don't hear a lot of other people kind of talking about because it's, you know, I think there's kind of a hunger to just like laser in and do, you know, build the best things now. But you know, I think that's actually.

A really important point because I, I talked to the Granola founder too about this and yeah, he made the same point. It's like, you know, maybe your customers are screaming at you to improve this feature, but you know that the next model is probably going to improve it for you. It's like you don't want to, you don't want to do something that the next model. Which is deep.

Exactly, exactly. Exactly. And because all of that is just lost time and effort and like, yeah, you'll get to market a little before, you know, maybe other people will or something. But then it's also going to be probably worse than whatever the model, the next model has anyways. And so you're kind of better off, you know, but that's the trick. Things you just don't know. You don't know. You don't know what things necessarily are going to improve. So you have to like, it's just.

Hanging out outside the opening eye anthropic offices, you know, what's going on guys?

Yeah, it helps me in like the early access programs or whatever I think for these. But like, but even then it's like, you know, it's when you're doing like a target, how do you do roadmap planning six beyond six months, heck even six months. You know, it's like that's, that's the stuff where it's, it's fun because it's so dynamic. Everything can change, you know and, and, and we've seen it happen already and so we're, we're acutely aware and paranoid about okay we startups are an energy conservation and application project. And so it's like caseably like the granola founder but they said is like dead on. It's like, you know, not a good investment of time if you think there's high probability that this will be fixed upstream effectively, you know, maybe in short order, you know.

You know what I really like about all this? I feel like there's a lot of companies of bullshit work to a certain extent with PM has to write about your documents and then developers have to set up all these environments and download stuff and the core feedback loop is just you build something, you talk to customers, show the customer, you build something like that. So that's what I really like about what you're building because you're getting rid of those bullshit man. Just getting stuff into the hands of customers as quickly as possible and then getting that feedback, that's what actually matters, right?

You know, I mean I, I totally agree and like, and for me it's like I, you, my, my co founder and I, we learned how to code because it was the only way to get stuff in front of customers. Right. Like, and it's like if this stuff existed then it's like I don't know if I would have gone as deep as I have, you know, on understanding how software works now to you know, build and whatever have you because is especially if you're like a product person or an entrepreneur, it's like, like of course like great co batters are making sure the software is going to be reliable performance, all that matters, right? But it's actually like, like the rote task of writing that is often ancillary to the higher order task that you. That actually really matters that determines whether any of this code is even going to be used by anyone ever. It's like does this thing solve your problem? You know, a prospective customer? Yes or yeah. I mean and as soon as the fast you get answers on that. Oh my God. Amazing, right? Like, that's, that's, I think the hardest part about building products or services or startups or whatever is just, it is, you know, it can take a very long time to actually get real data and insights on from customers, you know, and like if something is great or not, you know, and this is, this accelerates that, you know, I don't know, order of magnitude more, you know, and at a fraction of the cost.

I love it, man. I love it. All you already need is boat and you gotta like do some shit posting on Twitter and then you have to get the customers and then that's all you need.

I loved your meme with the Jeff Bezos overlooking the. I retweeted that from our company account. He was like, yeah, this. That is a dank meme.

So, so where can people find Bolt? And also where can people fall follow you?

Yeah, yeah, yeah, yeah. So for Bolt, we have the simplest domain in the world. It's Bolt new. You can just go there. I think our Twitter link and stuff is on that page. But if on Twitter we are at Bolt New and not the period, but B O L T D O T N E W. So like Bolt new spelled out. And then my Twitter is. I think I'm Eric Simons 40. So like, yeah, that's where you find me. And I tweet. Yeah, I. I tweet every day or two or something, I don't know. But I post stuff that I find interesting, like when we're releasing stuff. And so, yeah, hope. Hope you follow me there.

Awesome. Eric, I really appreciate your conversation and hope to see great things for both coming soon.

Yeah, awesome. Yeah. Thank you for having me. Appreciate it.