Cognition CEO Scott Wu on acquiring Windsurf, AI replacing engineers, and the Moneyball-ification of everything

Scott Wu joins John Collison to talk about Cognition’s AI software engineer, the Moneyball-ification of everything, math competitions with Alexandr Wang in 6th grade, acquiring Windsurf over a weekend, whether coding tools will be replaced by the labs, and why he thinks we already have AGI. Full transcript on Substack: https://open.substack.com/pub/cheekypint/p/cognition-ceo-scott-wu-on-acquiring Timestamps (00:00) Intro (01:13) Early life and maths competitions (03:47) Addepar job as a high schooler (05:43) Where are all the young founders? (08:45) Moneyball-ification of everything (11:42) Cognition’s AI software engineer, Devin (15:46) Essential and accidental complexity (17:59) How Devin works with enterprises (19:48) IDE productivity (21:56) Nihilist computer use argument (25:55) Benchmarking Devin (27:15) Market structure (30:32) Agent economy (37:21) Cognition’s team of founders (39:31) Jevons paradox and software (42:00) When will we see AI UIs? (45:52) “I think we have AGI” (47:03) Windsurf deal (52:37) M&A in AI (54:21) Cognition’s culture (55:48) Learning as a CEO (57:12) Scott’s information diet

• 0:00Speaker 00

  Have you had a gin before?

• 0:02Speaker 01

  I have actually never had a beer in my entire life.

• 0:05Speaker 00

  Alright, well, you're starting with the best beer, so that's good. You order your Amazon packages with Devon?

• 0:10Speaker 01

  Yeah.

• 0:10Speaker 00

  So you're just in Slack and you ask him to buy something for you?

• 0:13Speaker 01

  Yeah, like just @devon can you go buy some more whiteboards for us or something like that. But I really enjoyed math competitions and going and competing and doing these things.

• 0:20Speaker 00

  And this is stuff like if I ask you what 694 squared?

• 0:26Speaker 01

  It is 481636.

• 0:28Speaker 00

  I have shuffled the cards. I am not collaborating. We give them to Scott.

• 0:32Speaker 01

  So now you have six cards and you're trying to make 163, right? And one way that you could do that here is 2 times 8 is 16, 9 divided by 3 is 3, 3 plus 16 is 19, 12 times 12 is 144, 144 plus 19 is 163. And so almost all combinations can be done.

• 0:47Speaker 00

  But you're probably thinking like, I could have done that. That's too easy. So.

• 0:52Speaker 00

  And for this guy, you can just put it upside down like that.

• 0:57Speaker 01

  Very good.

• 0:58Speaker 00

  Scott Wu is the co-founder and CEO of Cognition, which makes Devon the AI coding agent. Scott is a triple IOI gold medal winner and kind of famous for being a math whiz. And now he's at the cutting edge of agentic software development. Cheers.

• 1:09Speaker 01

  Alright, cheers.

• 1:13Speaker 00

  Tell me about your upbringing and all the math stuff. Like, I feel like you're known for the math stuff these days.

• 1:17Speaker 01

  Yeah, yeah. So I grew up, I'm from Baton Rouge. My parents were both chemical engineers. And so they immigrated from China for grad school. And then naturally, when they were looking for jobs, they were doing like air emissions permitting and things like that. And Louisiana has a lot of oil and gas, and so that's kind of how we ended up.

• 1:37Speaker 00

  A lot of air emissions too, I check.

• 1:38Speaker 01

  Yeah, yeah, yeah. And so that's how we ended up there. I always loved math as a kid. I had an older brother named Neil, super, super close, the whole way through. And Neil was about five years older than me. Neil started doing math competitions when he was in middle school. And so he would have been in like 6th grade and I was in 1st grade at the time. And naturally, I, as a little brother, would go and just watch what he was doing and try to learn some of the same math too. And that's kind of how I first got into math. And then, you know, I found that I really enjoyed math competitions and going and competing and doing these things.

• 2:12Speaker 00

  And this is stuff like if I ask you what 694 squared?

• 2:17Speaker 01

  I think it's probably not quite things of that nature. It is 481636. But it's things like, you know, yeah, like math puzzles, things like, you know, the frog that's like going up and then every night falls down the well and how many nights, you know, these kinds of things where you

• 2:33Speaker 00

  The ants on the log.

• 2:34Speaker 01

  Yeah, yeah, yeah, yeah, like where you kind of get to do the critical thinking and come up with interesting ideas and stuff like that. So I started doing math competitions in second grade. I remember it. There was a contest at the local college that I went to, which was for like middle schoolers and high schoolers. And so I competed in the seventh grade math division as a second grader. And I did the competition. It was like my first time doing any of these. I just really liked math and stuff. And then they were calling out like third place, second place, first place, and none of them were me. And I still just remember, I was just, I was so upset.

• 3:06Speaker 00

  That's your supervillain origin story?

• 3:08Speaker 01

  Yeah, yeah, exactly. That's how it all began, basically. And so then I trained a bunch. The next year I was in third grade and I competed in like Algebra I or something. And like I won that year. And then I basically kept doing math competitions from there. My last year of high school, which would have been my junior year. I left a year early, but I did IOI, the programming.

• 3:25Speaker 00

  Yeah, yeah, yeah.

• 3:26Speaker 01

  I did IOI three times and I got gold, yeah.

• 3:29Speaker 00

  Heard you go to school.

• 3:30Speaker 01

  So I went, I took a year off, actually. So I left high school a year early. I wasn't that good at school, I guess. I left high school a year early.

• 3:38Speaker 00

  Sorry, that's, obviously, that's surprising. You weren't that good at school.

• 3:41Speaker 01

  Well, I just, you know, I wasn't that good at finishing school, you know. I have a middle school degree, but you know, I didn't really make it through high school or college. So I left high school a year early. I spent a year actually in the Bay, working at a company called Adipar.

• 3:53Speaker 00

  Sure.

• 3:53Speaker 01

  And I did that as a software engineer. That was back in 2014.

• 3:56Speaker 00

  Yeah, wow, it was a blast.

• 3:57Speaker 01

  Yeah, yeah, it was a while ago. And then after that, I decided, okay, I will go try out college after all and see what that's like. I went to Harvard for two years and then I dropped out.

• 4:05Speaker 00

  How did you end up at Adipar? And that's very forward thinking of them, obviously, if they took on a

• 4:08Speaker 01

  Yeah.

• 4:09Speaker 00

  High school aged, high school dropout?

• 4:10Speaker 01

  Yeah, yeah, it was a fun group. You know, funnily enough, there were four of us who started the same, at the same time as high schoolers. And it was myself, Alexander Wang is actually another one. We started on the same day. Eugene Chen, who's now running Phoenix Dex. And then Srinath RA, who's most recently at Sandbar as the CEO.

• 4:35Speaker 00

  Wait, sorry, this is a real small group theory moment. So you and Alex were in the same.

• 4:38Speaker 01

  That's right. So we knew each other, we met in middle school.

• 4:41Speaker 00

  Alex now of Meta.

• 4:42Speaker 01

  Yeah, now of Meta, that's right.

• 4:44Speaker 00

  MSL, I guess.

• 4:45Speaker 01

  Yeah. And so we met in sixth grade. He was from New Mexico. I was from Louisiana, but we met in this math competition called Math Counts. We were both at the national competition. And then we started talking. Google Hangouts was the thing at the time.

• 5:01Speaker 00

  It turns out there's some math in AI. Yeah, this may be an.

• 5:03Speaker 01

  Yeah, it's a fun thing. Well, a lot of the folks, as it turns out, from our vintage ended up being. I think there's like a real infectiousness of, you know, being entrepreneurial too. I think Alex deserves a lot of credit for, I'd say, being the first of our group.

• 5:16Speaker 00

  Alex Wang got you into the idea of starting a company.

• 5:19Speaker 01

  Yeah, or you know, somehow I think there's definitely a bunch of that involved for sure. Yeah. But also, you know, a lot of folks, Johnny Ho, who's one of the co-founders of perplexity, for example, Demigol, who started Pika. You know, a lot of these, Jesse Zhang, who started Decagon. You know, a lot of us were actually competing in these math and programming competitions in the same year and we all knew each other.

• 5:43Speaker 00

  Okay, so this gets to something I was wondering. You know, there's this topic that people talked about a while back of where are the young founders? There always used to be kind of people in their early twenties working on breakout companies. You know, Michael Dell was 19 when he started Dell, 23 when he took it public. Obviously, you know, Mark Zuckerberg was very young when he started working on Facebook. And when it was like a real breakout, you know, he was still very young. And there was a period where there was no young founders. And now there's many, many more, like a whole bunch of people that you mentioned. You're 28, running cognition. Is the presence of young people as founders of leading companies a? Biomarker for industry vibrancy where, you know,

• 6:26Speaker 00

  Michael Dell was young during the takeoff of the PC era and, you know, Mark Zuckerberg was young during the takeoff of social networking. And now we're in the takeoff of kind of AI coding tools.

• 6:36Speaker 01

  Yeah, I was gonna say, I appreciate you calling me young. I mean, I think relative to being 18 or 19, you know, it's still a long way.

• 6:42Speaker 00

  The test is like in your twenties.

• 6:43Speaker 01

  So I have a, I have a take on this, actually. And I'm curious to hear yours on this. I, I, I've been thinking about this question as well. And my take is actually just that overall, being a founder has just gotten harder. And that's probably like the biggest, like the highest order bit. I think the reason that young founders who were just really sharp and really determined, like did very well is because at the end of the day, being a good first principles thinker does beat experience, you know? And just a lot of being a founder is doing something that has never existed before and coming, coming to your own conclusions. The thing is, now there's a lot of people who have both, you know, the first principles thinking and the experience. And I, I think things have gotten a lot more, you know, call it mature as a space. And so it's like, you know, basically it's gotten harder, you know?

• 7:27Speaker 01

  And so, so there are fewer that are literally coming out of college. I think now they're.

• 7:30Speaker 00

  It feels hard to make the claim that, you know, it was easy to start a leading business in prior eras. You know, Facebook, Facebook, lots of competition. It's not like Dell was the only PC maker. And so I don't think they had it easy by any stretch of the imagination. However, I think you are getting at something where clearly all the large companies these days, they're very aware, they're very connected with the ecosystem. If you look at Satya or Mark Zuckerberg, they are very aware of everything that's going on AI and they're paying a lot of attention to it. And so, yeah, maybe there aren't giant opportunities that are just being left on the ground by the biggest established companies.

• 8:12Speaker 01

  Yeah, and maybe harder is not the right word. It's more just that the space is a bit more mature and there's more of a playbook and like more existing knowledge. You know, there's, there's obviously something unique with every business, but a lot of the details of, you know. Here's how you should structure equity. Here's how you should figure out fundraising. Here's how you should hire your initial team. Many of these things, I think, do carry over a lot with experience, where I think in previous eras where the book wasn't written at all, almost. And so it really just came down to how sharp you were and how good you were at making your own decisions. I think now there's a lot more experience to draw from. Maybe that's part of that. I also do kind of just have a theory of like, I guess I would call it like the money-balification of everything. So to give a few examples,

• 8:53Speaker 01

  like, one of the things that I do casually for fun is like playing poker. And poker is a very fun game. It's actually much more mathematical than a lot of people realize. It's very, you know, of course, people kind of think of it as that.

• 9:03Speaker 00

  People know that, like, the poker solvers and the odds tables.

• 9:06Speaker 01

  Yeah, solvers and so on.

• 9:06Speaker 00

  Or is it more mathematical than that?

• 9:08Speaker 01

  No, no, I think that's right. I think that's right. Well, I think there's like a first-order impression of, you know, it's all about just knowing what you got.

• 9:13Speaker 00

  I'm all in, yeah, exactly. Play the person on the other, yeah.

• 9:15Speaker 01

  Yeah, yeah. And it obviously is much more mathematical than that. But the one thing that's kind of interesting is you see it in the evolution of the top players in space as well. That, you know, back in the day, in the 80s or 90s, you know, the top pros, again, I don't think the idea is that it's less competitive, but like the skills that made someone a really great poker player were just like, you know, really great intuition. Like, I think they understood a lot of the mathematical concepts, but just at a very system-one level of just being able to kind of think about them. And obviously, they had just like a good feel for the game and a good sense of how they should be able to kind of improve their own play. And now, it's just all math nerds. You know, it's basically like, at some point, when the space gets mature enough, that, you know, you know what I mean? Where it's like, I think for a less mature space,

• 9:59Speaker 01

  when people don't know what the right questions to ask are or how to even kind of think about it, like, what is the right frame of reference? Then I think there's something about just having a really sharp intuition and coming to your own conclusions. And then at some point, as these things get more mature, you know, the conclusion of it kind of is math. You know, and I feel like that's been the case in a lot of different fields. And I feel like it's happening a little bit for startups as well.

• 10:21Speaker 00

  I see. More spaces have kind of resolved to their underlying, like a chess engine just deciding that the position is, you know, Mason 41 or something.

• 10:28Speaker 01

  Yeah, and chess is totally the same way, by the way, which is like, you know, back in the 1800s, like, people.

• 10:33Speaker 00

  The romantic style of play is gone.

• 10:34Speaker 01

  Yeah, yeah, exactly. The romantic style of play. And now it's kind of like, yeah, like, there is a right sequence of moves and, you know, just seeing how how close you are to to that optimum.

• 10:43Speaker 00

  Yeah. What are other domains where the monobalification of everything is shown?

• 10:47Speaker 01

  Yeah. One of my other hobbies, which I played at least before, you know, the advent of cognition, was a game called Super Smash Brothers. I used to play tournaments for Smash. And you saw very much the same pattern where the, there's a game called Melee in particular. I don't know if you played Smash. Okay, okay. It's for the GameCube, which came out in 2001. So it's a very old game, but, you know, people just still keep playing the same game.

• 11:08Speaker 00

  Yes.

• 11:08Speaker 01

  And, you know, for the first like six to eight years of the game, is like, the personality was very much really wily, you know, sharp thinkers, people who are just like, quick on their feet and coming up with these ideas. And now it's just like, it's just, it's just all math, you know? And then the people who play and do really well are.

• 11:23Speaker 00

  I think some of the RTSs are a little bit that way as well. The player has gotten less creative as people have gotten better at them.

• 11:28Speaker 01

  Yeah, yeah. And it's a funny thing where it's like, like, you know, there's a lot of beauty in the nerd side of it, too. It's just like, it's like a difference in what skills get most selected for, is maybe the way I'd describe it.

• 11:41Speaker 00

  Yeah. Okay, I'm getting distracted from asking you about cognition. What is cognition? What does it do?

• 11:46Speaker 01

  Yeah, so we're building the AI software engineering. We've been building Devon for, you know, we've been going for the last year and a half, and most recently just acquired Windsurf. And so, you know, Devon, the agent, and Windsurf, the IDE, but at a high level, you know, we really want to build the future of software engineering.

• 12:01Speaker 00

  Is it confusing for people that you have two brands, you have cognition, the company, and then Devon, the slightly anthropomorphized instantiation of it?

• 12:08Speaker 01

  We've been talking about, I mean, now there's Windsurf as well, and so now there's a third thing, but I think some consolidation is probably good.

• 12:16Speaker 00

  Okay, and so people are maybe familiar with the GitHub Copilot or the IDE style paradigm, where you're there writing code in your IDE and it helps you autocomplete it, or you can give some instructions in the IDE. That is not the cognition Devon paradigm. Instead, with Devon, you're in a Slack channel with Devon, and you're prompting it to like, go off and build me an X or a Y, but you're talking to it as you would a coworker in Slack.

• 12:44Speaker 01

  That's right, yeah, and so you can call it from Slack, or Linear, or Jira, or you can call it from your IDE as well, but you don't have to, right? But yeah, I think that's exactly right. There's been this paradigm in the past, I would say GitHub Copilot was really the biggest, kind of like the most well-known originator of it, of IDEs. And I would describe it as basically when you are typing at the keyboard as an engineer, making you a little bit faster at it, and giving you the tools and the shortcuts and everything to do that faster. And Devon is a very different paradigm of what I would call like an async experience, where you have an agent and you delegate a task. And so Devon naturally operates a little bit more like at a ticket level, or a project level, or something like that. You have some issue in GitHub or something, and you tag Devon, and then Devon gets to work on it.

• 13:27Speaker 00

  Yep, yep. And what level of task is Devon doing a good job of today?

• 13:34Speaker 01

  Yeah, we like to call Devon a junior engineer today. There are some things that in AI, of course, is way, way better than all of us at, especially encyclopedic knowledge and just pulling facts and things like that. There are some things that it still makes terrible decisions on. But I think that's the right average overall. And what we see folks typically using it for are things like bugs, for example, or like simple, kind of like feature requests and fixes and so on, where you're talking about an issue and you and your team are figuring out what you should do, and you're just like, hey, add Devon, go do this. Or on the other hand, a lot of the more, I'll call it like the repetitive tedious tasks that come up often in engineering work. And so that's often migrations, or modernizations, or refactors, or version upgrades. It's crazy how much testing and documentation, it's crazy how much of

• 14:19Speaker 01

  Software engineers in the world's time is more like things like going and fixing your Kubernetes deploy than it is things like building and coming up with really

• 14:28Speaker 00

  Dependency management, yeah, all that kind of stuff. What metrics can you share on where the business is at?

• 14:32Speaker 01

  Yeah, so Devon is deployed in thousands of companies all over the world. We work with some of the biggest banks in the world, like Goldman and Citibank, all the way down to startups with two or three people. And in general, a lot of how we look at things is in terms of merge pull requests and getting Devon to the point where it is a significant percentage of the merge pull requests in an org. Typically, in a successful org, Devon is merging something in the range of like 30 to 40% of all the pull requests that come through.

• 14:58Speaker 00

  And you talked about this async model, but isn't it the case that, as I look at other, you know, the GitHub, the cursors and everything like that, I mean, they are cloud codes, they are not, they're not fully synchronous because now you prompt them and they go off and do something. And so are these distinctions a moment in time thing? Do they kind of go away where everyone is synchronous in the cases when they can do it instantly and asynchronous in the cases where they don't? But is this a durable distinction?

• 15:26Speaker 01

  It's a good question. I think the two experiences continue to exist for the next while. And then I actually think that figuring out, you know, the shared experience between them actually is the really interesting thing, right? And that's a lot of recently with WinSerp and things like that. It's something that we've already been thinking about and now are pretty excited to ship some things in the near future on. Do you know the concepts of like essential complexity and accidental complexity? Have you heard about this? Yeah. And I think there's a real thing where, maybe one way to describe it is, the ethos of a software engineer, what it means to be a software engineer in my mind is basically just somebody who solves problems in the context of code, right? It is somebody who tells the computer what to do and makes all these decisions of, you know, it can be big decisions like what is the right architecture that we want to use for all of this?

• 16:11Speaker 01

  Or it can be like a lot of these micro decisions like, oh, like by the way, there's like a case where this balance is less than zero and what do we want to do here? Should we show, you know, an error or should we, you know, request this or whatever, right? And all these decisions, you know, Are what people typically call the essential complexity of like, what is all of the actual underlying logic of the decisions of what the software is doing? Right? And the accidental complexity is basically everything else, you know, like all the things that you have to do to support things as they scale, you know, or all of your standard, for example, anytime you have a class, you probably have all the standard crowd features along with that as well, where, you know, everyone knows that you need to have that in your class, but there is no real decision that needs to be made in terms of going and doing that. Right? And it's an interesting thing, which is, you know,

• 16:54Speaker 01

  up until, you know, AI coding has come along. I feel like the meat of software engineering has been in making the decisions, and yet you spend 80 or 90% of your time doing more of the latter, you know, of just going and doing the routine implementation and so on. Right? And so I think this merged experience that comes up is basically something where, for anything that actually needs you in loop, where you can go and make the decision and you're looking at the high level strategy or deciding what you want to build, you're involved and you're doing that synchronously. Then for all the parts that are hero execution, you are able to hand that off asynchronously. Right? And so the interesting thing is that, obviously, for an individual project, there are typically long stretches that actually are one or the other, and it alternates between both of them. Right? And I think what that will effectively look like is, you know,

• 17:38Speaker 01

  the synchronous experience is the IDE where you are looking at the code directly and you see each of these things. The asynchronous experience is the agent that will go off and do each of these things. But to be able to go back and forth between your IDE.

• 17:49Speaker 00

  So you want the engineer to be interactive with the agent as it's going and working, but on the high impact moments of important choices as opposed to all the grunt work. How do you get large enterprises comfortable with giving Devon's sufficient permissions to be effective?

• 18:05Speaker 01

  Yeah.

• 18:06Speaker 00

  Like you talked about the migration at USK, super boring. And so you change the table and get it talking to the new table, and then eventually you delete the old table.

• 18:15Speaker 01

  Yeah.

• 18:15Speaker 00

  And that last step is kind of scary.

• 18:17Speaker 01

  Yeah.

• 18:17Speaker 00

  And I think people still have a, you know, models hallucinate way less than they did, but people still have fear of the model just making something up and

• 18:24Speaker 01

  For sure.

• 18:25Speaker 00

  doing it. And so, yeah, how do you get people comfortable with giving it enough power to be effective?

• 18:29Speaker 01

  So we pretty strongly recommend that people using Devon don't give it, you know, prod database access, for example.

• 18:36Speaker 00

  That's one way.

• 18:37Speaker 01

  I don't know of any instances where it has been an issue or things like that, but obviously, you know, you'd just rather not take that chance. And the framing that I would give, honestly, is,

• 18:47Speaker 01

  You know, we have processes for these things because humans make mistakes too. And that's why we have pull requests and review.

• 18:52Speaker 00

  Yes.

• 18:53Speaker 01

  And that's why we have CI. And that's why we have all these things already, right? And so Devin naturally slots neatly into all of these things. And so typically, the way that folks will work with Devin is, you know, they're doing some big code migration and they'll break up the task or maybe they have 50,000 files that all need to go, you know, upgrade from this version of Angular to that version or something like that. And Devin will go and do each one and it'll make pull requests, right? And so you will go and review the code and make sure that things look correct, but there's still this human.

• 19:18Speaker 00

  It's back to your point of incidental complexity where the reason a migration is time consuming is not the actual single deletion step, like all the time cost comes in other places.

• 19:27Speaker 01

  Yeah, yeah, yeah, exactly. I think in practice, what we see with folks, especially in these kind of like enterprise migrations is, you know, when folks measure internally, they see something like an eight to 15x gain for a lot of these use cases with Devin. Because, yeah, as you're saying, you're just reviewing the code. You know, you're not going and writing every single line or going through every single reference or things like that.

• 19:48Speaker 00

  So let's talk about that because I think all organizations around the world are trying to figure out the productivity impact of AI coding. And I think what everyone sees is engineers for sure want to have access to AI tools for coding. It's not totally obvious on the like PRs per dev type metrics or what's happening. Generally, you see some increase there, but of course, it's not clear how good even a, you know, pull request per dev metric is. And then maybe you can say that there is some, you know, ongoing maintenance cost of if you're shipping low quality code or something like that.

• 20:28Speaker 01

  Yeah.

• 20:29Speaker 00

  And so I feel like everyone right now is looking for some slam dunk productivity data on what is the impact of, you know, there's probably some CTOs looking for the slam dunk data to justify, you know, the spend of their CTO. What's your view on how big is the productivity impact? Is it actually measurable?

• 20:47Speaker 01

  Yeah, for sure. Yeah, so I think this is something where actually this gradual shift towards agents actually will help a lot, as it turns out. If anything, I think, to be honest, I think IDE productivity is often underrated because, you know, how do you state it to your point, right? Like you look at the numbers and it's, you know, of our engineering org, on average, people. Took the tab completion 238 times this week. It seems quite clear that that should be worth something and it should make you faster, but how much faster does it make you? It's a bit harder to say. On the other hand, with agents, a lot of the workflow, obviously, is going and doing the task for you, right? And so if it's a jury ticket or something, or a migration, or things like that, where you typically do have a good sense of how many engineering hours are going to be needed for this and what's going on. And because it's doing the whole thing end to end, it's a lot more clear of like, yeah,

• 21:32Speaker 01

  you didn't have to do this migration anymore. You reviewed the PR in five minutes and like, that's all done.

• 21:37Speaker 00

  Yeah.

• 21:38Speaker 01

  And I think as time goes on, I think these things will become more and more and more clear.

• 21:42Speaker 00

  There is a view that some people have out there that coding tools are a moment in time thing. That get run over by increasing model performance, GPT-6 or GPT-7. Presumably you do not hold this view.

• 22:01Speaker 01

  Yeah.

• 22:01Speaker 00

  How do you avoid getting run over by the labs?

• 22:04Speaker 01

  Yeah, yeah, for sure. So, look, I think the labs are obviously, like, I think they're incredible business. As best as I understand it, I would kind of describe this view as like a, call it like the nihilist computer use take, which is just like, of course, all of these different things that we do in the world, in knowledge work, just involve using a computer. And the AI is going to get better and better and better at using the computer until someday there is nothing left except just the AI going and using your computer and doing your work for you. To the best of my understanding is kind of the argument there. I see the wisdom of it. This is the kind of thing that's very hard to disprove. But I think that the, in practice, what we've seen in the space is naturally, there is

• 22:48Speaker 01

  a lot of contextual knowledge. There's a lot of like industry details, there's a lot of, and so, as we were saying, like going and doing some angular migration or doing some, you know, it's not to say that these things can't get better. In fact, I think they will continue to get much better. But I think that the way that we make models better and better at them is by giving it the right data of like, you know, how good can you be at angular migrations if you've never seen angular, you know, and you've never done an angular migration yourself, right? And there's this kind of a cap on that. And obviously, there are all sorts of these things of, you know, using your data dog to go and debug errors or... I think the biggest thing I would just say here is... Software engineering in the real world is so messy. You know, and there's all sorts of these things that come up. And I think in practice, you know, most disciplines look like this.

• 23:33Speaker 01

  And I would say the same thing about law or medicine or and so on. And so while the general intelligence will continue to get smarter and smarter, I think there is still a lot of work to do in making something, both, you know, on the capability side, really good for your particular use cases, but also in actually going and delivering a product experience and bringing that to customers of how that actually happens in the real world.

• 23:54Speaker 00

  So it's not a general intelligence task. It's a specific intelligence of, you know, working in the Stripe codebase requires some general intelligence, but requires a bunch of context, requires working within the workflows we have and everything like that. And you think that persists as an area where you need to specialize?

• 24:09Speaker 01

  Yeah, exactly. Maybe one way to put it is, I think the argument is something like a superintelligence. And I think in some sense, yes, I think we are, you could consider us short superintelligence. I think what we're getting to with RL, as this thing is improving and improving, like, and we see more and more of the gains and people are developing the techniques. You know, I think of RL and this paradigm of AI as basically, the platonic ideal of it is the ability to solve any benchmark. Right? You have exactly a data set of here are the things that you want, and here's how we measure success, and here's how we do that. And whatever that benchmark is, it can be the hardest thing ever. You know, it can be like unsolved math problems or whatever. Someday, we want to get to the point where we can just take that and train a model that will just get 100% on it.

• 24:51Speaker 01

  And I think, frankly, we're moving towards that ideal a lot faster than most folks would have expected. I think we're really, I mean, there's been some pretty crazy, you know, developments, like the IMO gold medal, or like, you know, the scores on Sweet Venture, or things like that. The thing is, when that happens, I don't think what we end up with is just pure ASI, end of humanity, human knowledge work or whatever. I think the thing that we end up in is basically a point where the hard question is, all right, now what is the benchmark, right? And I think defining the benchmark in all of these spaces is kind of like a lot of the practical, real messiness of the world, right? And so for a software engineer, obviously, you know, it's like, yeah, like, what are all the tools that you interact with on a day-to-day basis? How do you use those tools? You know, what does it mean to build a representation of the codebase over time?

• 25:34Speaker 01

  How do you decide whether you shipping the feature was successful or not successful? You know, all of these various things and creating the right environments around them.

• 25:41Speaker 00

  And so can there be a good benchmark for? A model's performance on the kinds of things that Devon wants to do? Or is that just, is like Devon's business model and Devon's revenue is the benchmark essentially?

• 25:53Speaker 01

  Yeah, yeah, it is a good question. From our perspective, we have a lot of benchmarks internally. The biggest is one that we call junior dev, which we might need to upgrade to senior dev pretty soon. But it's basically the ability to do a variety of just random, real-world junior dev tasks. And so we've shared some of the examples. Obviously, we don't publish the whole benchmark because then it would get obviated. But a lot of the tasks are things like, hey, you need to go and fix this Grafana dashboard and get this going and then pull up the results. This is a very common thing that a software engineer does, right? And the thing that's hard about it is perhaps not some algorithmic coding thing itself, but it's like, turns out on the setup, actually, the server that's hosting this is running the wrong version of some package.

• 26:34Speaker 01

  And so you have to go through the errors and figure out what happened and then say, okay, I need to downgrade the package to this other one, which is actually the right dependency for this thing. And then I need to run it and pull this up and make sure the numbers look correct. Things like that, which are basically as close as we can make them to what real software engineers spend their time on.

• 26:50Speaker 00

  And so how have the newly released Claude 4.1 and GPT-5 done this benchmark?

• 26:55Speaker 01

  Yeah, I mean, both of them are, the two of them are better at this benchmark than any of the models that we've seen before this week.

• 27:02Speaker 00

  As you think about the AI business and industry over the next five to 10 years, like you can think about all the different layers of the stack. You have the data centers, then you have labs, and then you have the application layers, such as yourself. Who benefits? Like, what gets more competitive? What gets less competitive? Are all these just classic competitive allegopolies? Yeah, what's the market structure?

• 27:26Speaker 01

  So everyone always makes fun of me whenever I say this, but I think all the layers are going to do very well. Like, I think all the...

• 27:32Speaker 00

  There's just going to be a lot of AI.

• 27:33Speaker 01

  And I think the prices are cheap everywhere. I've been saying this at least for the last six to 12 months. And I think we've seen prices go up a decent bit across all of these. But no, at a high level, yeah, first of all, there's going to be a lot of AI. It can't be understated in the sense that like, I think we're kind of coming off of a decade of... A lot of various, you know, B2B SaaS and so on. You know, I think there was the Internet, obviously, in like the 90s and early 2000s. And then there was the mobile phone and cloud, which were kind of like late 2000s, early 2010s. Right. And those were some of the biggest things in the last 30 years. Over the last 10 years or so, I think there was a real time where most of the stuff that was being built was a lot more incremental, basically, right? Like each, each next thing and building for a particular niche or for a small part of the workflow and making that more efficient.

• 28:17Speaker 01

  And AI now, I think, is the total opposite of that in the sense that, you know, now we're talking about the entirety of knowledge work and perhaps the entirety of physical work as well, depending on what happens with robotics. Right. And so, so first thing is there's just going to be a lot of AI.

• 28:31Speaker 00

  Yes.

• 28:32Speaker 01

  I think the second thing about where does the value accrue? My honest answer on that is, the simple thing is, value accrues wherever there's meaningful differentiation in the layer. Right. You know, simple, like if there's Nvidia and there's TSMC and there, you know, like, for as long as Nvidia needs to work with TSMC and for as long as TSMC needs to work with Nvidia, of course, there will be some rubbing up on, on, on each other's shoulders, but like they will continue to do great, right? And you, you kind of see this down the stack as well, right? I would argue that the problems that are being solved in all these different layers are very, very different problems that, that have pretty meaningful different differentiation, right?

• 29:06Speaker 00

  You're saying this prevents too much vertical integration, basically, where you get the layers kind of keep each doing their own thing?

• 29:11Speaker 01

  Exactly. Yeah. Yeah. And I, I think there's a real diff where, um, yeah, you know, as soon as you go from hardware to, to, obviously, foundation model training is its whole own can of worms and, and very much like the DNA of the companies is finding exceptionally strong researchers, giving them as many GPUs as you can afford to give them and, and setting up a, a culture that kind of like orients around that, right? And then the application layer, I would say, is really focused, I would say, obviously, it has a lot of the elements of research as well, but, but I think in particular is really, really focused on just figuring out how to make one use case work. For us, for example, like, the only thing that we care about is, is making, you know, is building the future of software engineering. And maybe one thing I would call out is like, you know, People often talk about AI code abstractly in a vacuum.

• 29:54Speaker 01

  I think there are a lot of companies that think about code in the foundation model layer or things like that. Like, I think we uniquely really think about software engineering, right? And all of the messiness that that comes with and all the product interface and all of the delivery and the usage model. And of course, like a lot of these particular capabilities that come with that. So I think there's like a real, you know, it's everyone has their own DNA and everyone has their own things that they do best.

• 30:19Speaker 00

  And we at Stripe have been thinking a lot about building the economic infrastructure for AI and what is required. You can have an agent acting on behalf of a person. And you want to be able to just be prompting or doing stuff in your app. And part of the tool use that your AI can engage in is going off and conducting commerce in the real world. And so we're building infrastructure for that. And then we noticed that because of the economics of AI, everyone has usage-based models, right? Per token, per what have you. And so we're building our, you know, usage-based billing infrastructure. And again, we find the billing systems people are building on Stripe, they're very different from the like classic SaaS is per-seat pricing.

• 31:04Speaker 00

  Whereas again, everything in AI is per-unit consumed. And you can get into how the agents engage in commerce with each other, where there's no human in the loop. So there are all these ways in which our product roadmap is being formed. But I'm curious what you think the economic infrastructure for AI needs to look like. Are there things that we should be keeping in mind?

• 31:23Speaker 01

  Yeah, yeah, for sure. Yeah, seat-based to usage-based, big, big, big, big one for sure. I think on both sides, right? From the perspective of one, seats don't really make sense when it is like the AI themselves are arguably seats as well. You know, like they're doing a lot of the labor too. And then on the other side, I think, you know, usage obviously just goes so naturally with the cogs themselves because a lot of this is, you know, effectively GPU spend on how much you're spinning the models, basically. And so I think that makes a ton of sense. The other big one which comes to mind, obviously, is just for there to be an entire agent economy as well, right? And so I think today, I would say is, you know, Still probably more of a talking point than than reality, but but I think things are pretty rapidly changing and and getting to the point where your agents are. You know, funnily enough, we use Devin. Devin is obviously entirely focused towards software engineering.

• 32:08Speaker 01

  But like, we order our DoorDash on Devin. You know, we order our Amazon packages with Devin. And it's like, there are pieces of that that turn out to work nicely anyway.

• 32:16Speaker 00

  So you order your Amazon packages with Devin?

• 32:18Speaker 01

  Yeah.

• 32:19Speaker 00

  So you're just in Slack and you ask it to buy something for you?

• 32:21Speaker 01

  Yeah, yeah, like just at Devin, can you go buy some more whiteboards for us or something like that?

• 32:25Speaker 00

  At a certain point, do the real world things you ask Devin to do run into just blockers with sites trying to block bot activity?

• 32:34Speaker 01

  You know, a lot of, a lot of Devin working really well, obviously, you know, relies on Devin being able to do these things and get through it. But but some of these things, you know, I think are are quite natural with the model, which is, you know, you often have API keys or secrets or things like that that you want Devin to be able to hold on to. And so that works for credit card numbers as well. And obviously, there's a lot of work of, you know, real world software engineering doesn't involve a lot of just going and browsing the web and finding different sites and clicking around on them. You know, even if you're just testing your own front end or putting in documentation or something. And so so good browser use, I think, is is an important piece of that as well. And I think it's just kind of something that's.

• 33:09Speaker 00

  So shouldn't you build a consumer app? Like, doesn't everyone want this Magic Wand app? Or you can just have your virtual assistants? Like, there's a million virtual assistant startups. It seems like none of them have really gotten to any scale.

• 33:19Speaker 01

  Yeah, it's a fun question. I think from our perspective, like, I think, on the one hand, like, it's it's it's fun seeing Devin go and do these DoorDash things. At the same time, we also just know that, you know, our our team is so small. We just don't have the the kind of, you know, focus to be able to do that in addition to doing software engineering. You're pulling up Devin and you're seeing this. And then on the other side, there's like the IDE there. But like, you know, Devin's just going on DoorDash or something, you know? It's a very like fish out of water experience, and I think it's fine for us to keep.

• 33:47Speaker 00

  But you know the way a lot of product development follows from people noticing? How a product is being used.

• 33:54Speaker 01

  Like emergent patterns.

• 33:54Speaker 00

  Exactly, and these emergent patterns, like Twitter especially.

• 33:57Speaker 01

  Yeah.

• 33:57Speaker 00

  People started linking to photos offsite, so they built, you know, in native image support. Or the hashtag was invented by the community. So similarly, you know, you're checking the Devon logs and you notice people are buying a lot of DoorDash.

• 34:08Speaker 01

  Yeah.

• 34:09Speaker 00

  Maybe that's a suggestion on the product side of things.

• 34:11Speaker 01

  Yeah, it's funny. Well, to be fair, it's mostly just ourselves.

• 34:14Speaker 00

  I know, it's still emergent product usage.

• 34:16Speaker 01

  I agree, I agree. It's a fun one, yeah.

• 34:18Speaker 00

  That's funny. I love that.

• 34:20Speaker 01

  Yeah, we had a fun one where Devon was, Walden had a flight that got cancelled and was trying to, you know, use Devon to go and like negotiate with the airline to get the refund for it. And Devon went to the site and naturally the site forwards you to their agent to have the conversation. And then Devon was kind of like explaining these things and like wasn't making progress. And then at some point Devon said, like, "This is not working. I need to speak to a human right now."

• 34:43Speaker 00

  And did it?

• 34:43Speaker 01

  It did, it did. Yeah, so it got to the human and then, you know, the human got on the line and then it sent some, like, the link to like the airline contract of like, "Oh, section 22 says this, this and that." And then Walden actually did get it.

• 34:53Speaker 00

  But sorry, Devon was speaking?

• 34:54Speaker 01

  Devon was chatting with the human.

• 34:56Speaker 00

  I see.

• 34:56Speaker 01

  Basically made it past the robot agent.

• 34:58Speaker 00

  Oh, that's funny.

• 34:59Speaker 01

  Equivalent and then got to a human.

• 35:02Speaker 00

  And did it successfully get the flight refund?

• 35:03Speaker 01

  It got the refund, yeah.

• 35:04Speaker 00

  Okay. Well, again, the people want this. And going back to the economic infrastructure for AI, the other thing that, you know, we think about is, it feels like trust is going to become a bigger deal online.

• 35:19Speaker 01

  Yeah.

• 35:19Speaker 00

  I don't quite know what form that takes, because obviously, you know, it's been a big bad internet for a long time.

• 35:25Speaker 01

  Yeah.

• 35:25Speaker 00

  A lot of scams out there, there's a lot of hacking. But I don't know, the hacking attempts become more sophisticated, the deep fakes and everything. And so having a good sense of who is a trusted individual, who is a trusted business.

• 35:38Speaker 01

  Yeah.

• 35:39Speaker 00

  Just seems to become much more important in this world.

• 35:41Speaker 01

  Yeah, yeah, like related to that too, I also think one of these things, you know, I feel the cloud flare with agents and everything is a hot topic and.

• 35:48Speaker 00

  Explain the cloud flare issue.

• 35:49Speaker 01

  Oh, yeah, yeah, of course. So, you know, there's a lot more agents browsing the web these days. And there's been certain things, you know, protections set up to not give agents access to websites. And I think the paradigm, you know, up until now, the paradigm for a lot of this stuff, I mean, there's robots.txt and all these things, has often been basically almost like a, you know. There are tons of things which you are not allowed to do.

• 36:11Speaker 00

  Yes.

• 36:11Speaker 01

  As a non-human. And I think what we will probably need to see a lot more of over time is basically like delegating access, if that makes sense. Like making it more clear that an agent can do something on your behalf. And in some sense, you're attaching some of your reputation to it, too. There's a monetary question of how this works out, but there's also just like actions that the agent takes are attributable to you.

• 36:35Speaker 00

  Yeah.

• 36:36Speaker 01

  And on your behalf.

• 36:37Speaker 00

  That's a great point. Right now, we have like bots versus no bots, you know, clankers versus clankers not allowed. Whereas instead, it needs to be bots allowed if you sign for them.

• 36:47Speaker 01

  Yeah, as I was just saying, like, simple version is just like if you're signed into your Google Chrome, like email account, and you have a verified address. Then you can have an agent run in that browser window and do things. But all of the, you know, you're responsible for the work that it does.

• 36:59Speaker 00

  Yes. Yeah, it's sort of like API key permissions, but at a mass consumer scale across everything and all websites and everything. Hmm, I like that. And how does the existence of Devon affect your own hiring of engineers?

• 37:13Speaker 01

  Yeah, I mean, from our perspective, we've always, you know, loved keeping the core engineering team very tight and very elite.

• 37:20Speaker 00

  What's tight, like thirty people?

• 37:22Speaker 01

  Yeah, so up until a few weeks ago, our whole team is about thirty-five people of whom...

• 37:26Speaker 00

  Across all roles?

• 37:27Speaker 01

  Across all roles, yeah. Of whom, I mean, almost everyone actually is an engineer by background, funnily enough. But what we call core engineering is, was about nineteen.

• 37:36Speaker 00

  Yep.

• 37:36Speaker 01

  With Windsurf, obviously, the team count has grown a lot, but actually, you know, with core engineering itself, it hasn't actually gotten all that much bigger. It's gone from nineteen to something in the range of like thirty to thirty-five.

• 37:45Speaker 00

  Okay, so you keep the engineering team smaller. And are the engineers, how are the engineers themselves different versus a company being built twenty years ago?

• 37:54Speaker 01

  Yeah, so it's a pretty different profile of the work that we have to do in the sense that like, there is a lot of execution and implementation that has to be done, but Devon does that so that humans don't need to. And so what we typically look for, our whole interview process, for example, for a lot of these, is basically just having people build their own Devon in eight hours and seeing how far they get with it.

• 38:15Speaker 00

  Sorry, build their own version of Devon or build stuff with Devon?

• 38:18Speaker 01

  Build their own version. Their own agent, their own full end-to-end agent in like eight hours or six hours or whatever. Yeah, I think what we find is, and I think this, you know, we'll see this trend generally in software engineering, which is knowing all the little, memorizing all the facts or knowing all the little details or being really good at syntax of some language or things like that are going to be less important. And what's going to be more important are a lot of the high-level decision makings or understanding the technical concepts really well.

• 38:42Speaker 00

  Yes.

• 38:42Speaker 01

  You know, having a good sense of products and just having a good intuitive sense of like what to build and what to do and being like, you know, like a self-owner that way too.

• 38:51Speaker 00

  Yes.

• 38:51Speaker 01

  And so, yeah, a lot of our team actually are specifically former founders, which is kind of a fun one, like of our initial kind of 35, I think 21 of us have founded a company before. And so it's been a, it's been a very high density of that.

• 39:04Speaker 00

  Wow. When will you hire your last engineer?

• 39:08Speaker 01

  It's a good question. I'll make a distinction here, which is I think that there will come a point, and my guess on this point is probably in the neighborhood of, let's say, two, three, four years from now, where we stop using code as the main interface.

• 39:25Speaker 01

  And basically, being a software engineer really is just instructing your computer and telling your computer what to do and saying, oh, like, you know, you're looking at your own product and you're saying, hey.

• 39:32Speaker 00

  You think two to four years from now, software engineers are not really looking at code in their day-to-day, just like they don't look at assembly, you know, today.

• 39:39Speaker 01

  Exactly. Yeah, yeah. And so that's going and looking at your own product and deciding, oh yeah, like, we need to make a new page here. By the way, like all this data, let's save this this way. And, you know, let's index this according to X, Y, and Z, you know, because here are the things that lookups that we need to do or whatever. You know, making a lot of these architectural decisions, but not looking at the code themselves, you know, at least in the majority of circumstances. I think at that point, obviously, the jobs change a lot. Funnily enough, I mean, I think if anything, we will have way more software engineers, not fewer. And I think just because the interface is not code anymore doesn't mean that the core skills of software.

• 40:14Speaker 00

  Yes.

• 40:14Speaker 01

  People often ask us, like, my son or daughter is in high school or is just starting, like, should they even be studying computer science? And my answer is always absolutely yes. If anything, you know, funnily enough, I feel like university computer science always had the opposites in of

• 40:28Speaker 00

  It was too theoretical.

• 40:29Speaker 01

  teaching you the concepts of what, what, what programming was about and what computer science was about, and not enough of like, all right, here's like syntax that you need to use. And like, here's what it means to get a React app set up and whatever. I think we'll get to a point where those theoretical concepts and that high-level understanding of, you know, Maybe in one line, like the model of a computer and how to make decisions, problem solve with the computer as a tool. That is what programming will be.

• 40:53Speaker 00

  Yeah.

• 40:55Speaker 01

  And if anything, there should be a lot more software engineers. Yeah, I think one of the nice things is everyone talks about Jevon's paradox and how it relates to AI. You know, I think there's nowhere that it's more true than software because, you know, we really never seem to run out of demand for more code.

• 41:07Speaker 00

  You can just write a lot of software.

• 41:09Speaker 01

  Yeah, the half joking way to say is despite how many software engineers in the world, you know, we all know this, there's so many products out there that are still so bad.

• 41:16Speaker 00

  Yeah, yeah, yeah, yeah.

• 41:18Speaker 01

  You know, you're logging into your bank or you're dealing with your like, you know, checkout at retail or whatever, and then there's all these things that are still like, super outdated, super buggy, you're logging into your healthcare platform or whatever, and you're trying to click around and find your thing, and it's like.

• 41:31Speaker 00

  We haven't finished writing all the software yet, yeah. Isn't it shocking that the UIs haven't changed at all? So we still, we talked to Siri, which is the same, I mean, button placement and the same brand on the iPhone as pre-transformer models. We, you prompt Devon via Slack.

• 41:47Speaker 01

  Yeah.

• 41:47Speaker 00

  You know, we use our AI tools in, you know, in a web browser.

• 41:52Speaker 01

  Yeah.

• 41:52Speaker 00

  And we enter them into a text box like, you know, we're playing Zork in the 1980s or whenever that came out. And so, 70s maybe, I don't know how old Zork is. Do you know what Zork is?

• 42:02Speaker 01

  I don't.

• 42:03Speaker 00

  Oh, you're too young. And it was like the original text-based adventure game.

• 42:07Speaker 01

  Oh, I see, I see.

• 42:07Speaker 00

  Yeah, yeah, yeah. But, yeah, when are we gonna see AI UIs? Because it's very retro right now.

• 42:14Speaker 01

  Yeah, my, my high level thought on this is, you know, you always see this with, with new waves of technology. Like, I think mobile phone is a great example where, you know, the initial apps kind of just look like basically websites, but in a spoiler box, you know? And over time, you know, there's, you can still get a lot of value out of those. Your core value prop of the phone was already there. But of course, over time, we built a lot of cool touch interfaces or we, you know, developed a lot of the science of, of what makes a good app UX.

• 42:41Speaker 00

  Yeah, but we've no multi-touch, we've no rubber banding.

• 42:44Speaker 01

  Yeah. Yeah, I think we are, you know, I think we are entering that phase now where, you know, for a few years, it was just kind of like replacing existing flows and just using AI to do that better.

• 42:54Speaker 00

  Yes.

• 42:54Speaker 01

  And now we're starting to think about a bit more of these, these kind of like various generative flows.

• 42:58Speaker 00

  Yes.

• 42:58Speaker 01

  Maybe the simplest example that comes to mind is a lot more products now have the little chat box at the bottom where, you know, rather than having to click through all the menus yourself, you can just kind of ask the chat box and find that, which is one very, very simple version of that. But, you know, I think there's, there's way more innovation to do, yeah.

• 43:14Speaker 00

  Yeah. One framing I was thinking about with this is, you know, it became clear shortly after the invention of the transistor and the microchip that everything would have a microchip in it, right? You know, everything could benefit from having a small computer in it and, you know, your car would have a small computer in it and your dishwasher would have a small computer in it and, you know, everything. And there's some equivalent where everything will pass through a transformer model before it's consumed.

• 43:39Speaker 01

  Yeah, one of my thoughts on this too is I think AI is, I'd say uniquely different from some of these previous ways in an important way, which is, you know, personal computer or internet or mobile phone. All of these had a, one of two things or often both. One was a big hardware component of like, yeah, you had to just go ship modems to everybody and you had to get people on the internet and you had to give everyone a phone first, right? And then two was like a very core critical mass effect or like a, you know, empty room effect or whatever, you know, network effect, whatever you want to call it, where the internet was great and all, obviously, but like it doesn't really get that useful until all your friends are on the internet too and like the restaurant that you're looking up is on the internet too and, and, you know, various other things as well, right? AI actually has neither of those problems.

• 44:22Speaker 00

  Hmm.

• 44:23Speaker 01

  And as a result, what you kind of see is like as soon as the tech works for somebody, you know, it's pure software, it can work single player and give you a ton of value directly. It kind of works for everyone. I think there's been a few things that we've seen as a result of that. One is, you know, there's, there's a new person posting that they're the fastest company from one million to a hundred million every, you know, every couple of weeks because, you know, AI is just so much faster at, as soon as it works, it works for everyone.

• 44:46Speaker 00

  Yes.

• 44:47Speaker 01

  But, but I think the other part of that is, I think to your point, I think there's, there's actually a bit of lag with products, I would say, where, you know, I think you could freeze all the capabilities today and have no new models and no new research come out and there would still be like a whole decade of product progress to make. Whereas I think before, you know, the product progress kind of tracked alongside the, the distribution itself. Now it's been much more sudden where it's like, you know, Two years total where everyone's been thinking about it. And honestly, if we factor in a lot of the more recent capabilities, agentic capabilities, things like that, it's like arguably less than one year for a lot of these. And we are all kind of grappling with that all of a sudden and trying to figure out what the right new product experiences are, right? And so it's just taking a bit more time.

• 45:26Speaker 00

  What are your AGI timelines?

• 45:27Speaker 01

  Yeah, I think we have AGI.

• 45:30Speaker 00

  Okay, now.

• 45:32Speaker 01

  Well, so I was gonna say, you know, there's this joke that people talk about, which is, you know, back in 2017, if you ask, you know, do we have AGI? The answer is no. And today, obviously, if you ask if we have AGI, you know, the first thing everyone always says, "Well, you have to go define AGI."

• 45:45Speaker 00

  Yeah, yeah, just hemming and hawing.

• 45:46Speaker 01

  Yeah, yeah, and I think it's kind of true in some sense of...

• 45:49Speaker 00

  Devin will order your DoorDash for you. Sounds like AGI to me.

• 45:52Speaker 01

  Yeah, yeah, and so, so, so obviously a bit of a facetious answer, but my honest opinion is, I think there is some, you know, rapid singularity, superintelligence thing that people kind of talk about. I would guess, it's very hard to say, you know, nothing's impossible, but I would guess that that's not something that happens in the immediate, immediate future, especially because, you know, as we said, a lot of the work to do is going and collecting all the real world. What are the problems that you want to solve? What are the, how do you define success for all these things? With that said, I think, yeah, I mean, we're gonna just keep, like, I think it's not so binary, basically. I think we're just gonna keep rolling out more and more improvements, and these things are gonna be more and more capable. But I don't know that we have some sudden shift, at least for the next few years.

• 46:35Speaker 00

  Yeah, no, that makes a lot of sense. We gotta talk about Windsurf.

• 46:39Speaker 01

  Oh, yeah.

• 46:40Speaker 00

  It played out so quickly, so give us the, give us the play-by-play.

• 46:43Speaker 01

  So we heard the news that it was gonna be Google buying Windsurf, or, you know, I guess not technically buying, or this whole, this whole deal that was happening, that Friday, at the same time everyone else did.

• 46:53Speaker 00

  Okay, so this means that's something that played out in advance, the Friday where the news came out, yeah.

• 46:57Speaker 01

  It was basically just as sudden for us. You know, we heard some rumors, maybe the night before.

• 47:01Speaker 00

  Devin was scrolling Twitter for you.

• 47:03Speaker 01

  Yeah, yeah, exactly, yeah. Devin came back and said, "Hey, you guys should check this out. We probably should look at this." And so, so we heard the news then, and naturally that afternoon, we're kind of talking about it and thinking about, like, is there something that we should do off of this?

• 47:14Speaker 00

  Yes.

• 47:15Speaker 01

  It's, it's not uncommon that there is some crazy news that happens in AI, you know? But, but this is especially, I think, you know, in our space. We talked about this idea. We reached out to them cold that evening and got to meet the, the, the new Windsurf leadership, you know, Jeff and Graham and Devin, that evening. And as we were kind of both talking about it, I think we kind of came to this conclusion together, which is if there is something to do here at all,

• 47:40Speaker 01

  then it has to be ready to go by Monday morning. You know, because everyone, all the customers were realigned, the whole team was like,

• 47:46Speaker 00

  Yes.

• 47:47Speaker 01

  Do I have a job, do I not have a job?

• 47:49Speaker 00

  It was a melting ice cube.

• 47:50Speaker 01

  Exactly. And so it's like, if it even waited until Thursday, you know, instead of Monday, like it was, people weren't going to cancel their contracts, people were going to get, you know, be interviewing at other places. And so, and so we said, okay, this is, what this means is like, if we want to explore this, we have to just spend the entire weekend on this nonstop.

• 48:09Speaker 01

  A lot of fun moments there. I mean, we, we, we got to kind of the handshake agreement that Saturday. And then obviously there's all the legal and everything to figure out, you know, we all pulled an all nighter that Sunday night. We had a very optimistic plan that we were going to get it signed.

• 48:23Speaker 00

  You also pulled an all nighter the Saturday night? Or did you get some sleep?

• 48:25Speaker 01

  We got a couple hours of sleep on Saturday. It was especially, I mean, I have the huge shout out to Jeff and Graham and Kevin because they had had a pretty rough few days before as well, actually. And so they were already pretty sleep deprived coming into it. We were going through it, we had this optimistic view that we were going to get it signed on Sunday night. And so then we could go and, you know, focus on filming and figuring out how we address the team and everything. Obviously that did not happen. And, you know, we got it signed on Monday at like 9am because us and the lawyers were up all night, basically just sorting out all these things. We luckily filmed the, the kind of windsurf video in the windsurf studio. We said, okay, we should just film it anyway.

• 49:00Speaker 00

  You realize you're going to announce acquisitions without a video?

• 49:02Speaker 01

  Yeah, yeah, yeah. It's always nice to have one, yeah. And then as soon as we got things signed, we, we were up in front of the whole team and giving them the update and then kind of sharing that publicly pretty soon after. It was a lot of, it's fun. I live for these moments, honestly.

• 49:17Speaker 00

  So you read the news on Friday?

• 49:19Speaker 01

  Yeah.

• 49:19Speaker 00

  And you signed the deal on, signed and announced the deal on Monday. But that means that you decided more or less instantaneously that you wanted to buy the remaining part of Windsurf.

• 49:30Speaker 01

  Yeah. So I think we, we talked it through on Friday evening. And I think from our perspective, there were a few things that were nice about this. First of all, obviously, you know, we, we know the space very well.

• 49:38Speaker 00

  Yes.

• 49:39Speaker 01

  And that sense, we didn't really have to diligence the product or the customers because we knew that, right? But as we were kind of understanding the pieces of like, what happened exactly with the team? How many of the folks are still there and who left? You know, we found that there was a very nice synergy in the sense that, you know, there was a core kind of like research and product engineering team that went to Google, and all of the other functions were entirely intact, which includes enterprise engineering, infra, deployed engineering, go-to-market, marketing, finance, operations, you know, all these various things. And funnily enough, I think with cognition, for better or for worse, I think we had done a good job of building out this core research and product engineering team, but we're, you know, I think a little bit behind on growing all the other functions. And so we found a very natural fit there as well. And as we were kind of just talking, you know, it's like they had JP Morgan,

• 50:24Speaker 01

  and we had Goldman Sachs, and they had, you know, and there were all of these kind of just like very natural ways to fit in. And so I think from our perspective, yeah, we knew there was something really interesting there, and we wanted to do it, and a lot of the rest was just figuring out the details.

• 50:35Speaker 00

  So you got to acquire a bunch of people who have lots of familiarity with the space. They have a product offering that is in an adjacent but not identical place to Devon, and so you get accelerated, it sounds like, the good market efforts, and broaden out the product portfolio, that's what you think about it?

• 50:53Speaker 01

  Yeah, yeah, yeah, absolutely. And then, of course, the products themselves, I think, are, funnily enough, we were thinking about what does the interaction of an async product like Devon look like with a more sync product? And we had some ideas for certain synchronous things that we wanted to build. We weren't going to build an IDE entirely because it felt like there were a couple of players in town already. But as it turns out, you know, having the idea, there actually were a lot of natural synergies with a lot of the synchronous stuff that we thought about. And, you know, very simple thing, we shipped Wave 11, you know, a few days later, after we closed that deal. And there are a lot of these basic things like, yeah, like being able to access your DeepWiki in your IDE, or being able to use all of the like Devon code-based representation in search, or, you know, spinning up the agent there, right? And all of these things, I think, we just felt a lot of natural complements.

• 51:35Speaker 01

  And so from there, it kind of felt like, you know, if there was a right person to work with and do this with, you know, it would be...

• 51:41Speaker 00

  So in six months, do I buy Devon's and I get Windsurf bundle? Do I separately buy Windsurf and I can buy Devon?

• 51:48Speaker 01

  Yeah.

• 51:48Speaker 00

  How will it work?

• 51:49Speaker 01

  Yeah, a lot to figure out still. We certainly want to keep each of the product philosophies the same. Like I mentioned, like I think there will still continue to be both sync and async products. But I think making the integration between them much stronger and much easier, and I think is going to be really nice. And so certainly a lot that'll be much easier from the customer perspective. But if for some reason they really wanted to use one of the two, I'd imagine that they would still be able to do that.

• 52:10Speaker 00

  It's obviously been an interesting aspect of the AI space. But there's been a number of these 49% licensing type deals to avoid the risk of an acquisition being blocked. Companies buy the license to the IP and then the talent that they want to be able to be sure comes with the company. Do you think that stays as a thing in the AI space? Like it's a funny moment in time thing, right?

• 52:35Speaker 01

  Yeah, I certainly don't feel like I'm the expert on this one. It's the thing that I find funny. There's one new bell or whistle each time, you know? I feel like there's a

• 52:44Speaker 00

  Unlike all the legal and contractual stuff.

• 52:47Speaker 01

  Character scale, you know, like you see like this one, like and now we do, you know, this licensing deal and now. And so I think the meta game around that is certainly developing. There is some amount of polarity at the top level of AI as a space in the sense that like there is a point at which you want to just have like, you know, these things do scale with resources and they scale. And so I think basically the games get bigger, I guess is one way to put it. And I think for most companies, the question is basically whether they think they will get there themselves or whether they want to work with another company.

• 53:20Speaker 00

  You're saying you would expect more M and A, whether it be like classical M and A or this new model of M and A because there are scale benefits in this game?

• 53:28Speaker 01

  Yeah, like maybe one of my hot takes is like, I think for a lot of the big, of course, there will be, you know, many medium sized outcomes in AI, but I think in this space, a little bit more so than previous ones. It's a little bit more polarized towards like you become a hyperscaler or bust. And so, you know, for some companies that feel like that is like, you know, that is the trajectory and the moonshot that they want to go for. And that's one thing where for others, like, you know, working with someone is something that people do.

• 53:54Speaker 00

  And so now as you're bringing the Wind Surf team on board, Cognition has this Very intense culture, you know, you guys work, you know, you work on the weekends, you

• 54:05Speaker 00

  all work out of this house and so you're doing this buyout offer.

• 54:10Speaker 01

  Yeah, yeah, yeah, I think for us, it's, you know, and most folks have been really excited to come in and do it. And only a small fraction have taken the buyout, but I think from our perspective,

• 54:22Speaker 01

  we just want to make sure it's an opt-in, you know, situation for everyone because, you know, let's be honest, it isn't for everyone. And I think it is a, you know, a very intentional thing there.

• 54:31Speaker 00

  Wasn't the intensity you wanted people to, what did you want people to opt into?

• 54:34Speaker 01

  Opt into the intensity and the new culture and yeah, we're going to be going after some very ambitious goals. You know, I think, you know, by revenue standards or by, you know, whatever you want to call it, you know,

• 54:46Speaker 01

  there are, you know, folks might call us a mid or later stage company, but from our perspective, you know, we're still very much early stage in terms of the profile of what happens next and how much more there is to build and how much more there is to do. And obviously at an early stage, yeah, you know, we do all have to be, you know, signing up for the uncertainty and the willingness to just go and take on a different challenge every week and to, you know, to put in a lot of hours and to have that culture. That was a big piece of it. Obviously, you know, regardless of what happens, we wanted to make sure people were well taken care of, but.

• 55:20Speaker 00

  Yeah.

• 55:22Speaker 00

  Every day, Cognition is the largest company you've ever run. Cause you're speedrunning, coming up to, it was true of me with Stripe as well, to be clear. And was, you're speedrunning, learning how to run a company. I'm curious, how you, how do you learn this stuff? How do you say I, but how do you learn more broadly?

• 55:40Speaker 01

  Yeah, yeah, no, I mean, it's a, I've got a lot to learn still for sure. I think many of these functions are, if anything, you know, we, like I mentioned, we have under invested in a lot of functions, maybe because they're not as top of mind for us as they should be. And now that's something that we're pretty actively working to do more of. I don't believe in like. Professional coach or career coach in the literal sense, but I think obviously you learn a lot from your peers and your friends who are doing similar things. So having a lot of close friends who are

• 56:10Speaker 00

  People you went to math camp with, apparently.

• 56:12Speaker 01

  Getting so, yeah, and then learning from these, all these different folks. And I do think as an entrepreneur, you know, it helps a lot to have, you know, a close group of friends and you can just be very honest and say, this thing is totally messed up and I have no idea what we're going to do. And please tell me if you have done anything like this before, you know, or things like that, which has been really helpful. You know, I think Eric and Kareem from Ramp, for example, or all these various folks from math competitions or, you know, my previous co-founder Vlad from Lunch Club. You know, a lot of different folks that I talk to for advice and I think it really does help a lot.

• 56:46Speaker 00

  Last question, I'm curious, what is your information diet in terms of how you learn about the world?

• 56:53Speaker 01

  Yeah, a lot of, I think Twitter is really, for tech news, I think is really the place to be. We share a lot of things.

• 57:01Speaker 00

  Do you find there's too much video in the algorithm these days?

• 57:03Speaker 01

  I think there are.

• 57:05Speaker 00

  Like it's kind of become TikTok.

• 57:06Speaker 01

  There is a lot of video, but then I just don't watch the videos, you know, for the most part, or you see the first few seconds, which is an interesting thing to think about as people who are making videos too. Of like, make sure you can convey your point with no sound and with the first three seconds. Like as much as you can do that, I think there are still like another like 5X of users who reach that are in that camp. The Twitter algorithm is the extent of how AI affects my information diet.

• 57:28Speaker 00

  But that's you on the receiving end of AI as opposed to you using AI as a tool.

• 57:32Speaker 01

  It's a good point. It's a good point. I mean, I should have Devin, you know, just GitHub Actions.

• 57:35Speaker 00

  The morning report like Zazu.

• 57:37Speaker 01

  Job, basically where Devin just goes and does the morning report and gets that. There's a lot of optimization to do still.

• 57:42Speaker 00

  The President's Daily Briefing. Yeah. Well, Scott, thank you. This is awesome.

• 57:47Speaker 01

  Thank you so much for having me.