Brandon, Scott, thanks for being here. And thank you all for sticking around. It's the end of the day. So I want to set the scene, set the stage first. It's two years ago. People are freaking out about ChatGPT, but the term vibe coding hasn't been coined yet. Like we're still kind of figuring out like what is all this stuff? All of a sudden there's this article in Bloomberg about your company and Devin, this coding agent, and you're saying, you know, you could just ask it to make you an app and it just, you know, it'll just do it. And people just absolutely lost their minds. They're like, this is crazy. The technology though, it was early, you were early, and now you have, you know, you can only really recently can you actually do this, and you have a lot of competition now — Claude Code, Codex, you know, Cursor. Do you think you came out too early?
Oh cool, okay, fun starting question. So look, the thesis of our company from the very beginning, you know, back in the day, to your point, ChatGPT had just come out. The way that folks were interacting with AI was very question-answer, right? You ask a question, you get an autocompleted answer. You ask a question and it was already useful, right? But our thesis from the very beginning was that reasoning was really going to improve. And because of that, you would actually have AI that could actually interact and make decisions in the real world. And it would go and, you know, take on tasks and do things step by step and go through all that good stuff. And now we call that agents, but, you know, there wasn't a term for agents at the time, but that's always been our thesis, you know, from the very beginning. And that was kind of the highlight of our launch.
It was interesting, you know, I mean, I obviously remember that launch two years ago and all the things that we got. And basically everyone's response to that launch was one of two things. It was either, oh, there's no way this works, or this is just not going to work at all, or, oh my god, all the software engineers are going to lose their jobs. And it was kind of, you know, it's like, you know, those can't both be true, you know, right? And it was kind of like, actually there is a path in the middle. And there's a path which basically involves all of us being able to build much more and do much more because we have, you know, your whole army of AI software engineers to go and do things. And that was what we had always believed in. And I mean, I think to your point, it was certainly early at the time for the capabilities, but what we found is that by really focusing on those specific problems and what it meant to go and do agentic work in the real world, and you know, big enterprises like a Goldman Sachs or like a Santander or even like the US government, you know, it was the right time to start working on those problems as the technology kept getting better and better.
Do you agree with that, Brandon? You're an investor in this company, an early investor. So what do you think?
You know, Scott and I have talked about this a lot over the, I guess, couple years now — maybe two years we've been invested. In the fullness of time, I think he was entirely right. I think the agent form factor was the right form factor. I think whether you call it vibe coding or like the tab-complete moment, that was like a fairly transitory period to go after like what looks like the — I don't want to say the final form factor, but pretty close to it. So it's either like right time or maybe even a couple months too late, one could posit. Earlier maybe always better, but no, I think in the fullness of time, Scott and Devin and the Cognition team were entirely right about the form factor.
This is how it always is with your venture investors. No, you actually weren't fast enough. He let something slip backstage where he said you have said that the expected value of Cognition is $8 trillion. Is that true?
Well, so here's my honest take on it, which is I think about the entire market of software. And so there's about 30 million software engineers in the world. Okay. And in aggregate, you know, their total salaries, for example, are about three to four trillion. Obviously it ranges across wages in different countries. Then you think about all the cloud spend, all the SaaS, all the tools that companies pay for and so on. Total software spend around the world is around six trillion, six to seven trillion, right? And the way I always think about it is like, well, first of all, we are going to become much, much more efficient at producing software. Do we think that number is going to go up or down? And I think at this point, you know, most folks have the Jevons paradox, right? Where it's, you know, things get cheaper, but you actually buy so much more of it or produce so much more of it, right? People see that now. And I think it's clear in software. I mean, there's really never been a bottleneck of how much software is demanded, right? And so I think that's going to increase and I think that whole pie is going to grow a lot, right? And I think what happens honestly is that even as a market, you know, this idea of AI for software engineering — I mean, it sounds crazy to say it, but I think it will do, you know, on the order of trillions of revenue per year, you know, if you give it 10 years or something like that, and you kind of see that horizon. Because, you know, we see this everywhere around the world, it's the technology works like an engineer that goes and uses this across the board, they can build three times faster, five times faster. And so, of course, they should be using this and working with this technology. And I think from there, obviously, you know, it's more a claim that the whole space is a massive space rather than a claim that, you know, we're going to go and win all of it by ourselves.
If I just give an anecdote to that, and like in venture capital, people talk about like golden markets. And like, what is a golden market? It's something that was like a zero-dollar market maybe a year or two ago and it's a multi-billion dollar market today. When Copilot first came out, Microsoft, I think, made all these announcements that it got to like $200, $300 million of revenue in the first year. And that was like a feat that's almost never been done before in software. It gone from like zero to a couple hundred million even for somebody with the scale of like a Microsoft. I think codegen today, you know, a couple years ago it was zero. This year it will be somewhere between a 10 and 20 billion like revenue spend market. I think most people would argue we're still in the early days and that's grown like multiple hundreds of percents per year.
Wow. Yeah. And I think a lot of us are starting to see the power of this now with the harnesses, right? Whether it's OpenClaw or Claude Code, like we're seeing this on the consumer level now, right? Is that basically like what you've been seeing, but it looked like that on your end at the enterprise level now for like a year or something?
I think that's exactly right. Yeah, I think over the last bit, so where we started helping folks was largely in, you know, a lot of these big migrations, modernizations and so on, big repetitive tasks that enterprises have to go through. And you know, if you think about what it looks like at a big enterprise, and it's like, you know, you have a 50,000-file codebase, it's been maintained for decades, it's, you know, super, super stress-tested, you want to make sure everything is exactly right, right, and working in that environment, and solving for that obviously is like a very specific thing to get to. But for us, what we've already seen is that for a lot of these tasks, you now just want to be able to just delegate that to a full agent that can test its own code, that can go and click around on a website, that can go and like, you know, have access to all the same tools that a human can.
Those are the kinds of things that you used to hire like consulting firms, like big consulting firms for. Have you like taken business away from them, do you think?
Look, it's again, I mean, there's just so much more software to build. And I think there's a certain — it's an interesting topic for what it's worth because I think a lot of folks forecasted this naturally, right? If you're going to make every software engineer three times more effective, well surely you're just going to go cut two-thirds of your software engineers. And naturally that's not what happened, right? And the reason that's not what's happened is because the reality is every bank and every private equity company and every healthcare company, every big engineering org in the world honestly, if you ask them, I've never met one that's told me, all right, we're shipping this project this month, we're shipping this other project next month, and then we're all done, no more software, I'm glad to be finally done with this. Right? Like everyone has a hundred projects and they have to pick eight. Right? And so getting to do 24 projects instead of eight is really a lot more of what you see typically.
Scott, I'm not gonna ask you to do math like you had to do on that podcast the other day. But you are this math, you know, competitive math champion and there's these great photos of you as, you know, like in second grade winning math competitions. The thing I'm curious about is does that actually come in handy? Like do you use that on a daily basis or not?
It's a good question. So there was an article that came out this week about one of my good friends, Jeffrey Yan, who's building a decentralized exchange called Hyperliquid, which is, you know, one of the biggest. And in it, it kind of mentioned that there was this group of us who all grew up doing these math and programming competitions. And among them, you know, Alexander Wang, who ran Scale AI and now is running the AI team at Meta. You know, Johnny Ho is one of the founders of Perplexity. Jesse, who's the CEO of a company called Decagon, which is also doing really well and growing. And it's kind of a funny thing because, yeah, I mean, we all knew each other in middle school. We grew up — I'm from Louisiana. A lot of them, you know, Jesse was from Colorado, Alexander Wang's from New Mexico, but we were all super, super tight. We were all the same year. And we all knew each other because we met at these like national competitions. And so, no, I mean, it's an interesting one to call out exactly. And maybe what I'd say is I think in AI you really see a big difference in the ceiling of what's possible on a technical level. And here's kind of one way I would describe it, which is, you know, if you think about some of the businesses of the previous waves or the previous decades, if you think about like a DoorDash or an Uber or a Facebook or something like that, obviously it's like, you know, you need to have good engineering, right? You can't have your servers crashing. You need the web app to run quickly. It has to, you know, it has to be a smooth experience. But you know, a lot of the detail beyond a certain point is obviously less so about the engineering, you know, once you've gotten it basically enough of the way there, and it's more so about things like, you know, just being really, really scrappy or figuring out the right exact product loop and so on, right? I think in AI what you kind of see is so much of the difference of the entire product experience that you can serve your customers or that you can build in your product is of course just a very pure technical question, right? And if you have an agent that's smarter or can work faster or can make better decisions, then obviously that makes a big difference. And I think that's maybe one of the reasons that a lot of these companies in the last few years have swung so technical.
