Could OpenAI fill Microsoft’s shoes?

OpenAI recently announced a $200 million deal with the U. S. Department of Defense, which has us wondering: Could this further strain the company’s relationship with its biggest backer, Microsoft? After all, there have been numerous reports about growing tensions between the two companies, particularly as they become more competitive over enterprise deals. Today, on TechCrunch’s Equity podcast, hosts Anthony Ha and Max Zeff discuss how the OpenAI/DoD deal reflects Silicon Valley’s increasingly cozy relationship with the military and why industry leaders are calling for an AI “arms race.” Listen to the full episode to hear more highlights from the week, including: Whether it’s a good thing that Vice President JD Vance joined Bluesky (and was briefly suspended) What it means that Wix acquired a six-month-old “vibe coding” startup for $80 million (and why Anthony hates the phrase “vibe coding”) A panel in which investor Ali Partovi and Cognition President Russell Kaplan discuss what technical talent means in the age of AI Equity will be back next week, so stay tuned! Equity is TechCrunch’s flagship podcast, produced by Theresa Loconsolo, and posts every Wednesday and Friday. Subscribe to us on Apple Podcasts, Overcast, Spotify and all the casts. You also can follow Equity on X and Threads, at @EquityPod. Credits: Equity is produced by Theresa Loconsolo with editing by Kell. We’d also like to thank TechCrunch’s audience development team. Thank you so much for listening, and we'll talk to you next time. Learn more about your ad choices. Visit megaphone.fm/adchoices

Speaker names marked ? are inferred from speaking time and the show’s known hosts — they are not voice-verified.

• 0:00Anthony Ha?

  This episode of Equity is brought to you by Schneider Electric. Schneider Electric, your energy technology partner.

• 0:16Anthony Ha?

  So Max, did you see that JD Vance joined Blue Sky this week?

• 0:21Speaker 01

  I did, and it was clear right from the start that he was up to no good on there.

• 0:28Anthony Ha?

  Tell me more about his no good.

• 0:30Speaker 01

  So, I mean, right from the start, he started posting about transgender kids and he was suspended almost immediately upon arriving, perhaps setting a new record for a time spent on Blue Sky, T-SO-B, before getting suspended. We have to come up with a new acronym for all of that. But apparently it was just a fluke and Blue Sky said that they thought he might have been impersonating JD Vance, but it turned out, alas, it was the real one all along.

• 0:59Anthony Ha?

  Well, this is, I think, an interesting question in the sense of, you know, I think there's always this periodic thing where conservative and centrist kind of pundits are always like, "Blue Sky is dying. Blue Sky is irrelevant. Blue Sky is an echo chamber." And so from a certain perspective, JD Vance joining Blue Sky is a good thing for Blue Sky and them to ban him even accidentally seems like a huge mistake. But there is also something that seems wrong about, A, JD Vance joining Blue Sky and B, kind of just coming out guns blazing and like, "Let's talk about banning gender-affirming care," which is pretty antithetical to the ethos of Blue Sky.

• 1:36Speaker 01

  Yeah, he was clearly looking for a reaction right from the start there, and I think he got it. But, you know, look, maybe he'll foster some great conversation over there. We'll see. Maybe he'll subscribe to the moss feed. I'm a big fan of that on Blue Sky.

• 1:51Anthony Ha?

  I don't know what the moss feed is.

• 1:53Speaker 01

  The moss feed. It's a special feed where instead of just regular social media, it's just pictures of moss. It's the best. It's the best part of Blue Sky.

• 2:01Anthony Ha?

  Yeah, I don't understand how anyone can say blue sky is dying when there's something beautiful and strange like that. But we have other things to cover. So welcome back to Equity, TechCrunch's flagship podcast about the business of startups. Today is Friday, June 20th. I'm Anthony Ha, TechCrunch's Weekend Editor.

• 2:18Speaker 01

  I'm Max Zep, Senior AI Reporter with TechCrunch.

• 2:21Anthony Ha?

  And on the show today, we are going to have deals of the week from OpenAI and a vibe coder called Base44. And then we're gonna have a panel that Max did. So he'll tell us more about that before we get to it. But let's start with OpenAI, which announced a $200 million contract with the Department of Defense. Max, what did you make of this deal?

• 2:42Speaker 01

  Yeah, so you know, OpenAI landed a $200 million contract with the DOD. For OpenAI at this scale, that size of that contract is not nothing, but it is not a make or break contract for them exactly. I think the notable part here is just that OpenAI is coming out and saying, "We're gonna build AI models, you know, on the frontier of AI and technology for national security, to help kind of with cyber defense, to help with healthcare access for service members." And I think OpenAI for a long time had a part of their kind of charter that basically stopped them from being able to help militaries with AI workloads.

• 3:25Speaker 01

  And there was a point last year where OpenAI peeled back this provision from their charter. And slowly, but they've been moving towards working with the military, which, as we know, is a very important partner for Silicon Valley. And now I think OpenAI is just coming out saying, "Look, we're working directly with them, and you know, we're making $200 million from them." So I think that this is kind of a big moment for OpenAI's growth into a normal company.

• 3:55Anthony Ha?

  Right, and we should say that there is this tension where, on the one hand, if you are a big enterprise software provider that just has high expenses, high revenues, you're probably doing government and military work to some extent. At the same time, companies, not just OpenAI, but companies like Google and Meta in the past have kind of tried to stay away from that. I think their employees have tried to sort of push them through activism to stay away from defense work, and all of that really seems to be going away right now, those barriers.

• 4:23Speaker 01

  Yeah, OpenAI is kind of, you know, unique in the way that it has this mission to benefit humanity and to not, you know, make humanity worse off with their technology, which is a noble goal. But in some ways, it's not that different from Google's, you know, slogan when they were a young startup of, you know, "Don't be evil." They didn't want to work with the military when they first started out, and of course, they do a lot of work with the military now. So I think that this is really just part of OpenAI's evolution here. But look, I think that, you know, an interesting part of this is that Microsoft is a big provider to the military as well. And, you know, there's been a lot of reporting in recent weeks that Microsoft and OpenAI's

• 5:04Speaker 01

  relationship is a little bit touch and go, as OpenAI is kind of growing into this real competitor to Microsoft. And I'm curious to see what Microsoft's reaction is to something like this, and whether, you know, this was a deal that maybe Microsoft was in contention for. It certainly seems likely.

• 5:20Anthony Ha?

  It seems like, and again, almost all of this stuff is based on sort of anonymous sources, you know, like good reporting, but officially, OpenAI and Microsoft have been doing great, but they're still figuring out the terms of what their relationship looks like post, you know, transition. But my understanding is that it really is these kind of big enterprise contracts where there is increasing competition that in the past, Microsoft's maybe pictured this as, "OpenAI is going to be developing this core technology," but, you know, they're fundamentally this kind of nonprofit, and then Microsoft is really going to be the one packaging this technology for enterprise clients like the DoD. And, but the thing is, like, OpenAI is so big now, you know, needs to sort of keep growing, that that kind of stay in your lane mentality just doesn't make sense, right?

• 6:06Speaker 01

  Yeah, yeah, I mean, Microsoft is kind of famous for working with militaries and governments and kind of being this like, you know, maybe not the most exciting technology coming out of Microsoft like for the last 20 years, but they've been a very reliable person to handle all of your cloud services and data. I think that in recent years, though, Microsoft has really struggled with some of these government contracts. I mean, there were a lot of security concerns around Microsoft that got brought up in the last year or two. I remember there was kind of this bombshell report that said that Microsoft's cybersecurity was just insufficient. And I remember the CEO Satya Nadella came out and said, we're gonna, you know, really focus on this to revamp this. And now OpenAI is having this new push into, hey, we're gonna

• 6:49Speaker 01

  work with militaries. They have an OpenAI for government effort too, to try to work with the United States to handle their kind of AI services. I think it's really interesting to kind of think about a world where as AI plays more of a role in controlling, you know, your data and like AI being tied into your cloud services, whether OpenAI could kind of fill Microsoft's shoes. I think that's a real fear over at Microsoft these days.

• 7:14Anthony Ha?

  Right. And I mean, I think it creates a lot of tension, but I imagine that if you're Microsoft, you still want to have that relationship with OpenAI because if you don't have that relationship, you're probably in an even worse position. I think the other thing that's interesting about this news is also like the way it's being framed around this kind of idea of like Cold War arms race around AI, particularly kind of against China, which on the one hand, I do think that, you know, the government should be investing very seriously in AI because I do think it's going to be very important for the coming years. And it would be like a disadvantage for the United States if it didn't do that. I think on the flip side, sometimes it feels like this rhetoric is a way to sort of stifle

• 7:56Anthony Ha?

  criticism or skepticism about AI because it's sort of like, if you think that aspects of AI are not great, unethical, bad for the environment, and then you're sort of an AI skeptic, then suddenly you're unpatriotic because you're supporting China getting ahead in AI.

• 8:11Speaker 01

  Yeah, there is kind of this real rhetoric going, like you said, this is a means to protect our national security or to keep our AI cutting edge. I think that's a lot different from how people are using AI today. You know, a lot of it is around this kind of like vibe coding and whatnot. And we actually are going to talk about another kind of more common use of AI in this next deal with Base44.

• 8:36Anthony Ha?

  Yeah, so this is an acquisition, Wix, which is, you know, one of the kind of started out as like a build your own website. Sort of a similar idea of like build your own website without any coding. And so they've acquired this six-month-old startup, Base44, for $80 million. And I think it got a lot of attention because it was, to be clear, I mean, I think there's a little bit of hype around this where it's like, oh, my God, we're starting to get to like the solo entrepreneur vibe coding unicorns. It's not quite that yet. It's $80 million, which is a lot less than a billion dollars. And he, I think, was kind of the main owner of the company. But, you know, he had eight employees who are going to be getting, you know, $25 million out of the 80 million. But for a six-month-old startup, that is still very impressive.

• 9:21Speaker 01

  It's pretty good. It's pretty good. And I think that, look, there has been, yeah, a lot of hype about these AI startups that can, you know, do a lot more with a lot less. It's also, I think, this valuation that they reached so quickly is also just testament to the crazy valuations around AI startups themselves these days. And not to say that it's not worth $80 million, but just, I think that all kind of AI talent is just receiving these crazy, crazy prices these days. But Base44 is really interesting. I mean, so basically what they offer is they let users build full applications just from natural language prompts. So, you know, whereas before you might have to kind of use a website like Wix or something to drag and drop in application,

• 10:05Speaker 01

  like parts of your website, buttons and menus and whatnot. Now you can kind of just talk to it and just build something like that. This is really similar to what Vercel offers with V0, which is a popular tool to just kind of build front-end applications. I see this as a really compelling kind of use of AI because I think a lot of people want to have an app and they want to have a website. And they just really have no interest in building it out themselves. They're just like, sure, it doesn't have to look great and AI can do it. So I think this is honestly a really smart investment for Wix to make.

• 10:38Anthony Ha?

  Yeah, it feels like, I mean, I would say even that, again, that generation before of the kind of like no code, build your own website thing is just such a very compelling use case for now AI where like I have, you know, a personal website myself, like I have very rudimentary like HTML skills, but like I don't want to be using that stuff. I think just having something that feels very out of the box, I think makes a lot of sense to me. We've already used the phrase vibe coding a few times in this conversation, but just a confession, like I think that is an ugly phrase. It just sounds like, it just sounds so hypey, you know, like, like you're just like, come on, like, be serious. Like, that does not seem like a term we're going to be using a year from now.

• 11:21Speaker 01

  You know, yeah, this was a phrase coined by Andre Karpathy, who was an AI researcher, who was at OpenAI and Tesla before that, and he's seen as a visionary, and he has dubbed that like, you know, everything that you chat with, and then you don't have to actually like get into the code is a vibe coder. But this is of course like very different from like Cursor or Windsurf or these other quote unquote vibe coders, you know, this is a grossly overbroad term that that we've started using everywhere, myself included. But it's, this one is, you know, really just about building applications. It has nothing to do with the backend of websites, really. There's a lot of ways that AI can kind of help people build

• 12:02Speaker 01

  things without necessarily being like a vibe coder, you know, it's, there's a lot of room here for nuance and whatnot.

• 12:10Anthony Ha?

  I think it also maybe speaks to some of the anxiety around coding as a job and AI and this idea, question of like, somewhat, I think surprisingly, after this wave of just, if you're worried about your job, learn to code. The most secure thing you can do is learn to code and suddenly like, oh, maybe that's actually going to be one of, I mean, and it seems like in the current wave of like layoffs and cuts and things, coders are actually one of the groups that are the hardest hit. So it's like trying to preserve some sense of mystique around the idea of coding, even if it's natural language.

• 12:42Speaker 01

  A hundred percent. And yeah, I think it's really fascinating. I'm curious if we'll see more kind of. Startups like this with very small teams go for very high valuations here. This is something that people have been talking about a lot for a while. I think Sam Altman was kind of like one of the first to bring up this idea of, you know, we'll have the one person unicorn and whatnot. I don't know if we've quite seen that yet, but you know, we have certainly seen these AI startups with very few people going for massive valuations. So I'm curious if we'll see more of that.

• 13:15Anthony Ha?

  Yeah, and I think one thing that's worth emphasizing, again, I'm not a coder, a serious coder myself, but like, I think when people talk about that idea of the one person unicorn, there's a little bit of magical thinking at play where it's not that you couldn't automate a lot of these things, but I think it's important to just keep repeating that when you automate something, when you have some version of AI doing it, usually it's not doing a great job. So if like, you're really comfortable with the idea of like having slightly garbled marketing copy or, you know, probably an application that feels slightly more just out of the box rather than bespoke and customized, it's probably fine, but if you're automating something, that's probably something that you're not very good at. That's not going to be a distinguishing feature of your startup.

• 13:58Speaker 01

  Yeah, I remember I was at a conference the other day. It's just a funny note. It was Instagram's co-founder was speaking and he was talking about how, you know, Instagram sold for like a billion dollars back in the day and Instagram was only like 11 people at the time. So he was saying like the thought of it being like one or two people to make a billion dollar company, isn't that, you know, actually crazy when you think of like the fact that something like Instagram and like, I don't even know, the early like 2000s or 2010s like reached that crazy valuation. So it doesn't seem completely impossible.

• 14:33Anthony Ha?

  No, yeah. So I mean, I think we also have this panel that you did that we're going to be listening to after the break, right?

• 14:39Speaker 01

  Correct. So we're going to be listening to Ali Partovi, who has helped launch the careers of a generation of elite founders through Neo, just startup incubator and kind of founder incubator. He was an early investor in Facebook, Airbnb, and also Cursor and a lot of interesting AI startups, including Cognitiv. And we're going to hear from Russell Kaplan, as well as the president of Cognition. They're an AI company that's behind the AI software engineer, so to speak, called Devon. So together, they're going to kind of unpack what it means to build and back technical talent in the AI age. And we're also going to get into, in that talk, a little bit about the Meta and Scale deal. They are both close to both of those companies, so I thought their comments were enlightening around this.

• 15:24Anthony Ha?

  Great. Well, listen for that panel after the break.

• 15:31Anthony Ha?

  Every company today is looking for an edge. Better insights, faster decisions, greater resilience. But as AI drives new growth, it also creates new demands on energy and infrastructure. Schneider Electric is connecting energy, automation, and digital technologies to help organizations make smarter decisions and operate more efficiently. Because the future won't be shaped by AI alone, it will be shaped by energy and industrial intelligence. Schneider Electric, advancing energy tech.

• 15:57Speaker 01

  I wanted to start out here, just kind of focusing on how you two know each other. And maybe, Ali, if you could kick us off here and just talking about kind of what Neo is and maybe how you met Russell.

• 16:09Speaker 02

  Yeah, I started Neo eight years ago with the kind of crazy conceit that one can identify future tech leaders when they're still in college. And not only that I had a knack for this, but that I could build a, essentially build a machine, an institution that systematically identifies future tech leaders and helps them maximize their potential. So even before raising the first fund, I spent thousands of hours visiting college campuses around the country and interviewing college students and networking and asking people who the top students were at different universities.

• 16:49Speaker 02

  And so it was this effort in, I think, September 2017 or so. Multiple people said, the one person at Stanford you should totally talk to is Russell Kaplan. He was a co-term student at the time, so you were doing your masters. And he was, you were actually interning at Tesla during the fall of your, of your co-term year.

• 17:12Speaker 03

  I mean, yeah, I had the sort of the opposite experience where, you know, there's this kind of old guy who's like flying around the country, like interviewing a bunch of, you know, a bunch of underclassmen. And like, he's already sort of super post-economic and successful. And like, why is he just taking red eyes, giving, giving these like coding interviews to, you know, all these undergrads? And it was a very extensive, like breadth first search of, of the college network. He, he sort of was talking to all the folks I found interesting. And so I ended up talking with him and Ali graciously agreed to, to meet me at Tesla, where I was, where I was interning at the time. And we, we went on a hike and we ended up sort of talking about how he viewed, you

• 17:55Speaker 03

  know, young people and their potential and talent and the sort of philosophy behind Neo. And what I, what I appreciated was that he was really focused on sort of the technical fundamentals and building a very technical community of people at every phase of their career, which I think later resulted in, yeah, coding interviews that you gave for all the, for, for all the college students you're meeting with. And after I did my coding interview with Ali, I kind of wanted to mess with him a bit, and so I gave him a coding interview. And it was, it was a problem, like an algorithmic problem with, there's sort of many different levels for how to solve it. There's kind of an easy answer and a medium answer, a hard answer, and so on. And we, we spent most of our time together, him, you know, with the pencil, trying to,

• 18:40Speaker 03

  trying to solve this algorithmic coding question. And he, he got the medium, the easy and the medium in the, in the time of our, of our coffee chat. And that evening, I was on a retreat for Treehacks, which is the Stanford hackathon I, I started when I was in college. And we were like, having a great time, it's midnight, a bunch of, a bunch of us, kind of engineering students. And Ali calls me around midnight on Friday, saying, Russell, I've got the hard version. I've got the hard version. And I said, what, what do you mean? And he, he tries to explain it and I'm like, I don't know if that, if that works or not. He said, let me just send you the code. And so he sends me this handwritten C code. This is like ancient dinosaur, you know, legacy code. And, and he sends me the C code and we look at it and, oh, actually, yeah, this totally works. And so after that, Ali and I were very good friends.

• 19:25Speaker 02

  To be clear, so I do give coding tests to undergraduates. But I have not actually coded in years. I didn't even have like a coding environment. So I wrote this code in Gmail. Never ran it, but just sent it to him and it did work. But yeah, Russell was one of the first undergraduates I interviewed. But it was very clear from the first meeting that he was interviewing me. And it was, and it actually changed, meeting Russell changed the trajectory of what Neo grew into. Because it kind of made me realize that whether I'm twice the age or more successful than an undergraduate,

• 20:08Speaker 02

  there's still this element of equality and of breaking down boundaries, where we can still each challenge each other on something as simple as an algorithmic riddle or challenge. And that's an important part of the Neo overall ethos of bringing people together across different boundaries, whether it's different generations or different parts of the industry or just different backgrounds.

• 20:35Speaker 01

  Ali, I mean, you've been incredibly successful on your own. You were an early investor in Facebook and Airbnb and you've co-founded Code.org. You've done lots of impressive things. I'm just really curious about this kind of desire that's driving Neo to find young people that are, in your own words, smarter than you. And you know, like, what kind of drives you to search out these people and help them with their careers?

• 21:06Speaker 02

  Sure, yeah. Well, first of all, life is just more fun, I think, when you're surrounded by young people and surrounded by people smarter than yourself. And my dad was a university professor and I think he experienced that and perhaps instilled similar kind of feelings for me. So there's, I guess, a selfish element of I get to be surrounded by people much smarter than myself and younger than myself all day long. Which is how I stay young. But I'd say there's also really a goal to build something larger than myself and that is hopefully here long beyond all of our lifetimes.

• 21:46Speaker 02

  We viewed Neo as hopefully a hundred year institution that is like an educational institution, but focused on the, you know, the absolute best talent in computer science and helping them maximize their potential. And so, yeah, so for me, there's, for me and our whole team at Neo, there's a real motivation to build something much more than just an investment firm. We view the VC investing as merely the business model, but the mission and the drive that makes us, you know, excited to work on it each day is the idea of building something that's larger than ourselves.

• 22:24Speaker 01

  And Neo has just launched this program, Semester in SF, that I really want you to talk more about. It's kind of like an alternative to like the Thiel Fellowship, so to speak, like sort of, it's, you know, basically giving young people an opportunity to take a semester off college instead of drop out of college entirely to come be in San Francisco. What inspired this initiative?

• 22:48Speaker 02

  So I guess for starters, I've always, what I've shared with the Thiel Fellowship and various other programs that I admire a lot is, you know, this belief in the extraordinary potential of a single individual and a belief in helping a young person maximize their potential and helping them aim higher and have more courage to take more risks. And that's, you know, that's definitely an ethos that other programs share and, you know, sitting here, we have a, you know, living example in Russell of somebody who has absolutely crushed it in his career and there's, there's others. But I've also just always thought about, are there ways to kind of programmatically help, help people who are in that early stage of their career in college?

• 23:34Speaker 02

  And so the specific inspiration actually was about, I guess it was about a year ago, a year and a half ago, I was attending Free Hack. Ironically, because Treehacks is an organization that Russell started when he was in college. It's a, it's one of the largest hackathons on the west coast. And I was hosting a get together for a bunch of the Neo scholars. And I, I know all of them personally, but then there was somebody there who I didn't recognize. And I introduced myself and this young person introduced himself and I asked him, you know, what college does he go to? And he said, oh, I'm not in college. I'm a Thiel fellow. And I said, oh, okay. Tell me, tell me more about that. And he was describing how, you know, like he was working on some project and the terms

• 24:19Speaker 02

  of the Thiel fellowship gave him a hundred thousand dollars, you know, to do whatever he wants, except return to college. Because if he were to return to college, he would owe the money back. He would be in debt. And I kind of realized, we're standing on Stanford campus at the time, surrounded by college students at a college event that he's attending. Seems like this person maybe misses college. And, and I actually kind of felt bad for him that like he was, he had done a program that has some good elements, but it was also kind of, you know, putting him in a place where he wasn't able to do necessarily what he might want to do.

• 24:59Speaker 02

  And yeah, so this made me think, what if there was a program that was more like a, kind of a lighter commitment where you still have some sacrifice, leave college for a semester and do it with a group of other kids. So not, you know, not as a solo person wherever you live, but a, you know, a shared workspace in San Francisco with weekly mentorship, with activities and opportunities to expand your network. Because a lot of the best parts of college are, it's not the classes and the professors. It's the, it's the time in between. It's the random silly stuff you might do with your friends. And if you just drop out and get a hundred thousand dollars, you might miss out on a lot of that or have to recreate that environment.

• 25:44Speaker 02

  Whereas we decided to create a program that feels like college, but for entrepreneurship, where you can build whatever you want during that semester and It could lead to different directions. If if you want to keep going, then we'll fund it and make it into a startup. But for the vast majority, they'll go back to school, finish school. But I think they'll never be the same again after it, because anyone who has tasted building their own thing, even if it's just for three months or something, I think they'll want to become a founder and hopefully they'll want us to be their their first investor.

• 26:18Speaker 01

  Yeah, Russell, you've had a very successful early career. I mean, you're still early in your career and it's going well. So, I mean,

• 26:25Speaker 03

  Thank you.

• 26:26Speaker 01

  You're welcome. And I don't have to tell you, but I I think, you know, I'd love to hear from you kind of how much of a role Neo played in your early career and even college. I mean, and just while we're on the topic of it, like, what was, you know, the university's role in your development?

• 26:42Speaker 03

  Yeah, I I think I've checked every box that there is in terms of how you're engaged with Neo, because I started, I was in the first batch of Neo scholars. And then I I was at Tesla, I was working on the autopilot team as a machine learning scientist, working on the visionary network. And so I I worked on this project called Hydranet, where we took all of the different models and we turned them into one big model to predict everything that's going on on the road. And I was really excited about the technology future of of that work and this ability in computer vision to really understand the full physical world. And I ended up leaving Tesla and and starting a company doing that basically real time video understanding. And Ali and Neo were our our first investors.

• 27:25Speaker 03

  So then I I became a sort of founder backed by Neo. And then eventually our our company got acquired by Scale AI in 2020. And I ended up at at Scale, where I was leading the machine learning organization. And over the course of that period of time, we, you know, built a lot of different business units and different products. And I ended up helping out as a mentor in the Neo Accelerator. And now we're LPs in the Neo Fund. So I I check all the all the Neo boxes and it's been it's been a really special, really special part of my own sort of personal development, because I remember when I was in college and we went to the first Neo reunion, it was me and a bunch of friends my age. And then we saw all these

• 28:04Speaker 03

  These like super legit people in tech are like, wow, you know, the CTO of Meta is here and like the, you know, the CTO of Dropbox here and all these people who have who sort of completely made it. And what was surprising was they were actually all really excited to talk to us, too. It was like, and you know, I'm, I'm, I'm early in my career, but I can kind of see the, the value of connecting people at sort of different generations of, of where they are. And you, you learn a lot that way. And I think, as you said, it is, it is more fun. And so it's been a really, yeah, really special part of, of my own journey.

• 28:34Speaker 01

  It's fascinating. And you mentioned scale in there. We're going to get back to that because we do have a former scale employee and an early Facebook investor up here. So get ready for that.

• 28:44Speaker 02

  Investor as well.

• 28:46Speaker 01

  Say again, and scale.

• 28:46Speaker 02

  And scale investor.

• 28:47Speaker 01

  Okay, even more to talk about. But so I want to talk for a while here, actually, about cognition, which you're the president of right now. And I want you to just briefly kind of describe, I think you guys have launched about a year and a half ago. Is that right? So I want you to kind of describe just what Devin is and how it's different from Cursor and Windsurf and some of these other kind of vibe coding tools out there.

• 29:11Speaker 03

  Yeah. So cognition is the research lab behind Devin, which is an AI software engineer. And we, we launched this in March of last year. So, you know, you call it a year and three months ago. And at the time, I think there was a lot of sort of disbelief of, oh, wow, like, can, can you actually do autonomous software engineering end to end? And that's really the fundamental difference between, between Devin and a lot of the AI coding tools, of which there are many, is, you know, to what degree are you going in the loop yourself? Building this with some AI leverage or really delegating the work entirely?

• 29:51Speaker 03

  And there's not one right answer, I think, for, for this. You know, GitHub Copilot was arguably the first really successful, widely used AI developer tool. And even just this sort of tab complete while you're writing code is, is really helpful. And then you have this whole generation of AI IDEs, which are, are sort of an even more souped up version of that. It's a full development environment where you're in flow, you're, you're writing code, you're with AI. You're doing it, you're doing it together. And, you know, we view Devon as sort of the next stage of evolution, where now every engineer basically becomes a tech lead manager of an infinite army of AI junior engineers. And, you know, probably we're going to keep having all of these different form factors for a while.

• 30:33Speaker 03

  And there's so much more code to write than code we have. But, you know, we think that this sort of autonomous agent model is going to get increasingly popular. And part of the motivation for working on this was when I was at scale, there is a sort of very interesting side effect of leading the machine learning organization, which is that you you work on the data pipelines for every foundation model lab, more or less. And so, you know, every new model release is several months downstream of data work, because you're really trying to sort of shape these models to behave in certain ways. And it's really the data work that does that. And what became really clear from that position is that, you know, these models are getting better at everything, but they're in particular getting a lot better at code. And one of the reasons for that is that code

• 31:18Speaker 03

  is one of the few domains where you have automatic verifiability. You can have a model write some code, and then you can actually run the code and see, did that code work or not, did it crash? And so you have this automatic feedback loop that you can use for really high compute reinforcement learning versus in a lot of other domains, we would have to go find these human experts, which got really, really hard.

• 31:37Speaker 01

  Right. And I think one of the fascinating things about Devon is when you launched, I remember everyone kind of, you were, I think, really one of the first, you know, to take on this whole, like, we're building an agent. And you just delegate tasks, like in Slack or something, to this agent that just codes for you. And now, I think, a lot of, you know, the cursors of the world, the windsurfs, like, they're all kind of moving in this agentic coding direction, where they're, you know, coding co-pilots are now doing more and more tasks on their own, and they want you to leave them alone for 30 minutes or an hour or something. But I feel like you guys were early to the idea. But I'm curious of just how well it works in practice. And can you tell me about how people are using Devon today in the real world?

• 32:17Speaker 03

  Yeah. It's been insane to see the level of sort of adoption and traction and growth. The thing that was surprising early on was that even with junior engineer level capabilities, there are just huge amounts of tasks in software that don't get done, but that people would love to get done. And one way of framing like where is Devon? Most commonly used right now, it's for solving engineering toil. It's the whole backlog in Jira of P1s, P2s, things that people have to get done, but it's like no one's really favorite thing to do when they wake up in the morning. And whether it's you're an engineer at a 10,000 person company, doing a big refactor that was going to take two years, and you've got to break it up into many small pieces, and now you can delegate each of

• 33:02Speaker 03

  those pieces to Devon, rather than doing it all. Or you're at a small company, and you're facing a deluge of feedback and bug reports and requests, and you just can't handle it on, and you sort of tag Devon as the frontline defense. That's where it's today, I think, junior level engineer. And our main goal from a research standpoint is to get Devon promoted to senior engineer. And I think once we have that, all bets are off. It's going to be pretty wild.

• 33:29Speaker 01

  How long could Devon code without someone else intervening?

• 33:33Speaker 03

  The guidance we give our customers right now is each individual session with Devon, you should give a task that would take maybe a half day to one day's worth of work, max, for a human. And there's really interesting research from a lab called Meter that tries to measure how fast AI agents are progressing. And the Y-axis they use is the human time it would take to do the task. And they find that in coding specifically, the doubling time is 70 days. So every 70 days, you can kind of give agents these tasks that were twice as long as the previous task. And that trend has held for more than a year and three months. And so if that continues to hold for the next 12 months, that's another

• 34:14Speaker 03

  five doublings will be at 32 days of work per session, which would be insane.

• 34:19Speaker 01

  We will see if that really holds true. That would be something. I want to ask both of you here, just you're trying to kind of automate the work of a junior software engineer. What is the value of going to college right now and studying software engineering and trying to be a junior software engineer? Please enlighten us.

• 34:38Speaker 02

  Yeah, I'm sure happy to address this. So I will say that the value of studying computer science is not just to become a junior software engineer. So the part that I feel most confident about is

• 34:53Speaker 02

  Is learning computer science at an earlier age. So K through 12 is actually where code.org has really promoted computer science education. And I would argue that everybody, literally every child should have exposure to it because it's the best way to learn how to think. Because you could ask the exact same question about learning algebra or learning how to, you know, how to write. I mean, I don't mean just the ABCs, but like studying Shakespeare. There's so many things you do in high school that are not going to be your, you know, your occupation, but they do teach your brain how to think. And I think computer science is one of the most effective and structured and potent ways to help, to, to learn how to think.

• 35:36Speaker 02

  And it, it leads to all sorts of additional skills, not just software engineering. So a big part of why we love backing founders with a CS background and this whole search for neo-scholars focusing on CS, it's not because, you know, Russell is going to be coding all day long the rest of his life. It's because a computer science background helps you understand organizational challenges, how to divide a more complicated problem into parts, how it's comparable to how to divide an organization into parts. It helps you with pattern matching, which is relevant for business strategy. So I would say, I mean, if you look at the Stanford curriculum, you know, the, the page

• 36:20Speaker 02

  for the computer science degree, it says, this is a, this is a, majoring in computer science is a good preparation for careers in things like law, government, the, the list of careers, software engineering is one of them. It's, it's definitely not the only career. Now, having said that, I do think that occupations in software engineering are going to be increasingly, it's just going to be different. You're not going to get a job doing the same type of like. Easier work that you might get today as a junior engineer. I think it'll be higher level work. And that's also, I'd say, a trend that's been since I was a college student, you know,

• 37:01Speaker 02

  like, and since before, the entire trend of software engineering has been that the engineers bump up to a higher level and the level that they used to be at is now abstracted away or automated. And I think that just makes each engineer more and more valuable.

• 37:17Speaker 03

  I think that's a great point. It's really the entire history of computer science is just increasingly higher levels of abstraction. And when we automated writing assembly, you know, with higher level languages, we didn't reduce the number of programmers as a result. In fact, there's way more programmers now than there used to be in the age of assembly. And I think this next wave of software engineering, where it's fundamentally AI-driven, has the potential to, you know, 10x, 100x from there, the number of people who are really doing what I think of as the spirit of programming, which is just telling your computer what you want it to do. All of these are just increasingly less arcane ways of telling your computer what you want it to do. And so we're excited about a future where everyone's doing that.

• 37:58Speaker 01

  Got it. And we have some time left here, so I just need to ask, so I mean, I wrote a story today about OpenAI saying that they were going to stop working with Scale AI. Google said earlier this week, or last week maybe, that they would stop working with Scale AI. There were reports of that, and that one wasn't exactly confirmed from the company. But investor in Scale, former Scale employee, what does the future of Scale look like?

• 38:27Speaker 03

  You know, I think investors in Scale at every fundraising round ask, what does the future of this company look like? You know, like, you know, back when it was just a sort of self-driving car data labeling company, I think the most common thing we heard was, I mean, there's just not that many companies doing self-driving cars. Is this really going to last? And then Scale actually had a huge spurt of revenue growth doing e-commerce. People don't know this, but actually the data enrichment for e-commerce was one of the sort of biggest business lines of all of Scale. And then, and then, you know, it kind of petered out. And then this human label booming boom for generative AI came to be and then at each scale there's been another order of magnitude growth and at this point scale it's it's run in a very sort of bottoms

• 39:11Speaker 03

  up business unit driven way where a lot of the people who are at scale are former founders want to be founders and so you have these sort of decentralized pockets of product and engineering which sometimes makes for a chaotic unified experience but on the on the plus side you have all this kind of innovation the skirts and now scale has a massive business with governments around the world it has a extremely large enterprise business you know I think meta is also committed to doing lots of data labeling with scale and and there's a few others as well so I think it's a it's definitely a new chapter but I would definitely the one thing I've learned is you never write off scale

• 39:49Speaker 01

  yeah and I know we're over time but I'm just gonna ask one more question so I mean you've known Mark Zuckerberg from kind of a long time for a while now a lot of people right now feel like meta is has their back against the wall in the AI race you've seen Mark Zuckerberg with his back against the wall in many different situations what do you think kind of Mark is going to do here to kind of get himself out of this kind of sticky situation he is in in the AI race

• 40:21Speaker 02

  I don't know if I know him well enough to really be an authority on this you know but I would say in the same way that you said don't write off scale I would not write off Mark I think he's I mean I'd say what you have in common here whether it's Alex Wang Russell Kaplan Mark Zuckerberg these are each you know different generations of people who are incredibly competitive incredibly smart and you know not just gonna like back down because the going got tough you know so I think that he's you know I think he's going to compete how exactly that looks it's to you know I don't really know how to predict but you know I don't think he's just gonna say oh this got difficult

• 41:06Speaker 02

  so I'm gonna give up and go relax and work

• 41:08Speaker 01

  all right well thank you both for being here this was great

• 41:10Speaker 03

  thank you

• 41:14Rebecca Bellan?

  Equity is produced by Teresa Loconsolo with editing by Kel. We'd also like to thank TechCrunch's audience development team. Thank you so much for listening and we'll talk to you next time.