Cognition AI closes $400M at $10.2B valuation — Scott Wu on Devin's 8-15x engineering speedups and the Windsurf integration

and uh properly send Antarctica apparently and just see like how how thrilling would it be to have nothing for days and weeks and months on end and then and then somebody just comes up on the video and just takes it out. What are they doing down there?

Uh anyway, we have our next guest, Scott W from Cognition joining the stream. Welcome to the stream. How are you doing, Scott? How's it going? It's good to see you again. It's been a little bit. Great to see you. I think I recognize that sweatshirt. Is that a founders? It is the founders. It is the founders.

It felt like the right thing to do today. We interviewed Alex Karp last week and he he referred to it as the founders fund because back in the day it was actually called the the founders fund and they dropped the which I think is funny. Yeah.

Uh anyway, uh that we're not talking about Founders Fund, we're talking about your company. Give us the news. What's latest? Yeah, so we we just closed a big fund raise led by Founders Fund actually um at uh so so we kind of are talking about Founders Fund. We we we closed a raise um at a $10. 2 billion valuation.

You know, super grateful for for all the support that we've had. Congratulations. And yeah, busy busy month and a half for us since the Windsurf news. I mean it's uh it's it's been a crazy couple months for code and and for us especially getting to just see the two products mesh together has been has been a lot of fun.

Yeah. Um where is the growth coming from?

I remember uh seeing a presentation by uh at Microsoft Build and they were talking about the value that Devon can bring to replatforming and that was super concrete for me because you know I can just imagine having an enterprise application written innet and wanting to move to Python and that being a lot of work to write the test suite and move everything over and just being able to unleash a ton of AI agents and coding agents seemed like incredible value and an incredibly fully high leverage use of technology.

Is that a big piece of the business? What what what else is changing? What else is growing in the last year? Yeah, I think that's right.

A big part of what we do is is I kind of describe it as going and taking on engineering toil, you know, and so that includes like all these replatforms and migrations that folks have to do. That includes version upgrades, testing, documentation, you know, issue triaging, all these various things.

Um, and it's, you know, the reality is that that that's a lot of of what takes up time for software engineers today. I don't think it's anyone's favorite thing to do, but it is it is a lot of the time for people, right? And so so that's often what we see.

And um, you know, we we've been working with a pretty big range of companies all the way from from the smallest startups all the way to to some of the biggest enterprises in the world. But yeah, because uh, Devon is in GA, so any startup can sign up for it. Correct.

Um but then obviously you have a you have a very talented team that goes and sells into very large enterprises as well. Yeah. What what's it been uh yeah Windsurf was known for their GT you know GTM team. What's it been like both just like bolting on or or really integrating a team like that you guys?

Um yeah I'm sure it's really added a lot of firepower. It's been incredible honestly. I mean, even I think just the two products themselves because um you know, as we were saying last time, like a lot of the the the there's kind of these two categories of product experiences that devs are using today, right?

There's the idees and the agents and even just being able to have both and kind of serve that all-in-one solution with Windinsurf and Devon um has been amazing for us.

And so we've seen a bunch of I mean Windinsurf and Devon were both already each individually growing, but but in the last month and a half it's it's grown even faster. I want to go into a little bit of a debunkathon, hit you with some uh some data points and you can give me a lot more context.

So, uh the first one that we saw, this is all from uh extremely reliable sources, usually uh random screenshots that I see on X. Uh but one was that uh uh when when a when a developer uses a an AI coding tool, they generate twice as much code but 10 times as many security vulnerabilities. Do you have anything to add?

Does that sound right? Does that sound wildly off? What can we do to stop creating security vulnerabilities? Yeah, there's there's a lot of studies out there. You know, studies have different sample sizes and they have different ways that they measure these things.

Um, for us, you know, with a lot of the projects that that we deliver because, you know, if somebody's doing a whole replatform or something, you know, that is a concrete project that you're, you know, uh, you're you're setting up a team to do and kind of accounting for how many hours you need to do.

um with with the enterprise customers that we work with, we typically see speed ups of around 8 to 15x in terms of how long it takes. Yeah. So so they're basically they're doing in one hour of an engineer's time using Devon what would typically take eight hours without Devon. Wow.

Um and and I think the main thing that you really have to call out here is like it really depends on the use case, right? And so I think for a lot of these kind of like repetitive tedious things that frankly the engineers don't want to do anyway like that that's where tools like Devon work really really really well.

You know Devon is not necessarily you know you could use Devon to try and kind of like be the ideator with you or something like that but it's obvious it's it's not today what we've optimized Devon for. Um and so that's that's perhaps uh um what what folks are seeing in some of these different coding products. Yeah.

And then on the security vulnerability side, do you need a, you know, okay, you have Devon, but do you need like Bill, like a different agent that just is watching for security vulnerabilities?

In in traditional software development, you might have like, yeah, there's this great developer that can just churn out code really fast, but then we got the guy over here or the gal over here that really knows what what is secure and what's not and and where the flags are. And you have pen testers.

Like, do is that a different product? Is that just a different layer of what you're training for or have you have you kind of built Devon in a way that you don't inject security vulnerabilities? Yeah, it's a good question.

So, so I I think with security I I think the main thing I would just call out here is like this is a problem that all engineering orgs have already had to think about and deal with. You know, it's not necessarily new in the age of AI that you have to think about developers introducing security vulnerabilities, right?

And and yeah, you know, to to your point, we have a lot of these processes in place for that, right?

which is you know the first one probably is just code review itself right I mean people you know typically you you don't allow people to just merge their code straight to master or or or something like that and and similarly you know you want to have similar processes in place for your AI agents I think there will be specific kind of like security based agents but I I kind of imagine it draws on a lot of the same um codebased intelligence and just like understanding you know what what these functions are meant for or how these are meant to be used which is kind of a shared intelligence and so you know maybe we release our own kind of like security theme to Devon that that specifically does you know uh does this like security review or or something like that.

Cool. Are you seeing uh companies change their hiring practices at all after implementing Devon and getting getting starting to get value out of it? I've just been curious to learn how companies are actually adapting as they sort of ramp up AI AI coding tools. Yeah, it's a great question. It depends a lot on the space.

uh of company because obviously you know all companies are competing with the others in their space and that's what ultimately comes down to this but honestly I think what we've seen actually it looks like folks taking more and more projects in-house um and so a lot of things that folks typically would have outsourced to a team of contractors or um you know handed off to a system integrator or something like that they're actually just doing in-house with Devon um and they're able to do that faster and you know one one of our biggest customers actually was um you know is a bank in Latin America that that started using us around the start of this year for um you know some of these refactors and migrations and so on.

Um and then just got to the point where their their team was seeing like a lot more effectiveness using the tool and then they they 10x their contract within the last eight or nine months and if anything they're growing headcount they're not shrinking every single one of their you know their engineer if if every engineer is worth way more then obviously you you want more engineers want to add more engineers.

Uh okay let's continue the debunkathon. Uh we saw some data from the census that showed that that AI is cooked. It's done. It's it's over. Uh it dropped uh the actual data point was for firms pulled by the census of that they have over 250 employees. The level of AI adoption quote unquote had dropped from 14 to 12%.

A little dip in the chart. Of course, people are joking about this, memeing about this, uh you know, uh saying it's all over, but also joking that of course this is just one data point.

Uh to me, it seemed extremely low to think that only 10% of businesses are even using AI since AI is kind of everywhere now and everyone uses AI and all sorts of things.

Um anyway, uh what what is your take on like what's going on in that data point or in that stat specifically and and what what other kind of context can you add for us? Yeah, I think I saw this.

This is the one who's going around Twitter where it says like in July and August there was like a tiny, you know, and then everyone was freaking about Yeah. Yeah. So, same, you know, I I don't really I haven't looked into the methodology of how they figured out that number.

It's kind of interesting because I mean for us obviously like a lot of our work is with big enterprises, you know, specifically working on these massive code bases and July and August have been our biggest months ever, you know, by far.

Um, so that's that's that's that's in terms of the the data, but but I feel like the high level thing which I would just call out is, you know, I I think there's still a kind of like like collective like I I don't know what there there's like a there's like a collective like refusal to to see, you know, reality, which is, if I were to say it, is is basically we have AI that passes the Turing test.

We have AI that gets an IMO gold medal.

have AI that you know does all these crazy things which 10 years ago we we would would have put it clearly called AGI you know and it doesn't mean that you do all of this work for free as a result there's obviously still a ton that you have to do to teach the specific capabilities and and to make the right product experiences and to get you know actual businesses out there in the world going and using these things but but I I just think the story you know the things that we're asking the AI to do are not harder than getting an IMO gold medal put it that way and so so there's There's a lot of work to do.

You know, I don't want to diminish like it's going to be years and years, you know, of taking all of the techniques that we now have and figuring out how to build the right experiences and get them out to everyone. But but I really don't think also I love that every time I come.

I'm just the I'm just the the crazy AI bull, you know, my new uh but but like, you know, I think it seems very clear that that AI will um do more and more of this work over time, you know, and software engineering.

I think you see this especially which is if anything I think the best coders are the ones whose whose flows have changed the most dramatically right and and you just think about how many engineers there are out there that still you know have to learn these tools and have to get on these things like I I don't think we are due for a slowdown anytime too soon.

Yeah, the the the methodology of the census data was a little bit odd to me. Basically, they called up uh individual proprietors, business owners, and they said, "Are you using AI?

" And they defined AI as uh predictive analytics, data analytics, image processing, voice recognition, machine learning, uh natural language. So, a bunch of people said no, we're not doing we're not doing data. We're not doing data analysis, which was very odd. Only 10% of people said yes to this question.

I feel like it should be like 100% because like if you if you're using Stripe like there's machine learning under the hood for fraud detection. If you're using Gmail like Gemini is right there. There's going to be autocomplete on your phone is using data analysis.

Like anyway uh my real question is one of the one of the lowest hanging fruits that I see in the software that I interact with is in government, you know, government websites basically. Uh, I imagine that the census could benefit from having Devon run around write better polling software.

Is there anything on the horizon where you might be working with the government? Is that something where you need to get ITAR compliance or something before you go in there? Is it on the road map? What do you think in terms of improving the software that our government uh serves to us? Yeah.

No, I I I think it's a great point and as you say, I think there's a lot of work to do and it's a great example of the kind of like, you know, the the messy problems which I I think it is very clear will be solved over the next couple years.

You know, I we there's no reason that the government shouldn't be using tools like these to go and build their websites better and faster and everyone knows, you know, how many bugs there are, how many simple things that could just be fixed, you know, if if there were an AI software engineer available to go and do this.

um a lot of you know it's it's it's it's funny because I think Windsurf actually funnily enough was ahead of us on on you know getting things moving with Fed Ramp and getting certified and all these things but but it's something that that we're working on as well. So that's very exciting.

Are you uh are you thinking at all about other acquisitions over the next couple years? You got a big balance sheet at this point. Yeah. Yeah. You know, I mean the last one went pretty well.

So now I um it's I I think um it's I I think for us it's it's obviously really valuable to have this capital for a few different reasons you know with model training I think with hiring great talent um with expanding go to market and so on um you know is it possible that that part part of that will also you know go towards you know buying other companies of course it's possible there are not any plans at the moment but yeah yeah um in terms of your position in the stack I see you as uh sort of like initially like a enterprise level application layer company that then is now going down with Windsurf into more of a B2B context, not full consumer but more broad usage um but famously model agnostic, not training foundation models, not working down the stack.

And we're seeing the foundation model companies um do deals with Broadcom to co-develop chips and and uh what Elon's doing with Samsung. And there's a lot of these data points. Are there any other places where you want to at least have better relationships with other with other layers of the stack?

Is any of that important or is it really just like you have your slice and you're just going to go and run at that for the foreseeable future? Yeah, it's it's a really good question. I mean, I think in terms of relationships, for sure. I mean, we work very closely with the foundation labs, for example.

you know, we have lots of things going on in terms of um how we do compute or even kind of like partners that we work with on distribution, right, which I think are great.

I I would say, you know, I think the um for for us it's we we obviously I think it's very important for us to just have a lot of humility about where we're coming from and what we're doing. You know, we're obviously, you know, a very new entrant to the space and and you know, are only like 0. 1% of the way there.

And I think you know for us to be able to meaningfully succeed is going to require a lot of focus.

Um and so I think from that perspective you know I I think the like you know we we know what what is the area that we really kind of want to own and that's that's basically you know everything between the the you know that goes from taking the model the base model itself and then doing the right kind of capabilities work and building that into a product experience that you know real engineers can use every day.

Um, and I think I think we will stay quite narrow kind of within that focus ourselves for for the future. So, you're saying job's not finished. Job's not I mean it's it's it's it's barely even started honestly.

I think and I think like the um I think obviously there there's a you know there there's there's I think a vision of the future where um where as we said you know where where software engineering is just as easy as telling your computer what to do.

Um and and I think the world of software engineering has already changed in some meaningful ways over the last two three years because of you know all these AI coding tools but I think we've got another like 10 20 generations of product experiences you know as a community to to build and unlock until we get to that full future.

So uh important question was this latest round of financing negotiated with Founders Fund over poker or chess? Unfortunately, it was needed. You know, it was funny actually. We were um on the weekend that we did the Windsurf deal.

We um obviously I mean it was it was a pretty substantial commitment from us in terms of Windsurf. And so, you know, we want to we we explicitly needed board approval, but also, you know, wanted to give our investors an update and kind of hear what they thought and so on.

And so, you know, we were on with the Founders Fund team with Napoleon and Peter and and getting their thoughts on on what they thought makes sense.

And I I mean, first of all, they they were they were really supportive of us doing the deal and they thought, you know, the partnership made a lot of sense for for a lot of the same reasons that that we did.

Um but but they also, you know, were very clear of like, hey, like I know this is going to be, you know, a big thing for you guys on the balance sheet and and for for um for for you guys as a company and, you know, we want to be clear that we we we stand ready to to support you in the next thing if this is the deal that you want to do.

Um, and honestly, that's a lot of what gave us the confidence to be able to go ahead with it and close the whole deal, you know, in the 48 hours or 72 hours or whatever it ended up being. Um, and so we're good for our 400 million. Yeah, it's it's it's it's really been amazing. I I can't I can't shout them out enough.

Um, you know, Napoleon and Peter and the whole team that we've gotten to work with have have been great to us. That's fantastic. Um, tell us the story of the deal that was negotiated over was it poker? How did that come together? Yeah. So, so it's a funny one in retrospect. So, so it didn't actually happen.

Uh but but basically when we were doing the the series A, you know, one of the very early rounds of the company, um it was me and Napoleon and you know, I was here and he was there or something like that and and there was like a I think at some point it was clear that the deal was going to get done, you know, but it was just kind of like going through some of the more minor terms and and making sure we were on the same page of everything.

Uh, and we kind of proposed, you know, what if we uh what if what if we just had a nice heads up match to go and settle this and decide whose terms we're gonna take. We we did not do that, but it's probably probably for the better that we didn't do that, but but but yeah. Yeah. Oh, yeah.

But the story lives on as as something on. It almost happened or was at least like joked about, which is fantastic. Uh, that is great. Well, thank you so much for taking the time. Jordy, do you have anything else? No, congrats. I know you got a busy day, so we'll let you get back to it.

But we always try to appreciate try to try to have another reason to come on before the end of the year. At least a couple, you know. You know, it's a great lock. Well, there's going to be plenty more stuff to debunk uh through the rest of the year.

There's going to be tons of data points that people read into way too much and we will be giving you a call.

I was going to say I look forward to coming on again and being the idiot optimist that just, you know, everyone else is giving their rational arguments about this is going to work, this is not going to work, and I'm just here saying everything's bullish. I mean, if you want one more, we can talk about inference costs.

I mean, people are saying that, oh, inference costs aren't falling. You got to use the latest and greatest model and your gross margins are going to be terrible forever. Uh, debunk that one for me. Sure. No, I mean, I I I think there's so so I I think there's two separate questions there, you know.

One is the question of is the business going to be defensible? you know, are you going to be able to offer something that that you know the you're competing with your competitors on something that is not just purely a race to the bottom, right?

And then the second thing is like is the cost value trade-off for the customers going to actually make sense, right?

I think the second one is just very obviously true because you know a as these tools if the trade-off is you know doing more and more and speeding up labor right is you know making every accountant faster and making every lawyer faster and so on.

I mean, at some point it's pretty obvious that the machines are cheaper, you know, that that that cheap enough that making, you know, every lawyer three times faster is just obviously a no-brainer, right?

I think on the the point of the first one, you know, obviously it's a great point and and I think that's it's different for every different company. You know, I think a lot of the businesses in the application layer need to think a lot about, you know, what is their special sauce and what is their differentiation.

Um but I think the answer for for most probably of what that comes to is a combination of one is just really really focusing on specific use cases and delivering solutions that are that are you know really kind of custom solutions for the problem that they're specifically solving and then two is you know I think there is like a real kind of um I'll call it like a personalization effect that that I think we're seeing more and more you know there's chat GBT memory for example obviously Devon has its own whole knowledge system where you know a a tool just understands the customer and their own business and their trade-offs and everything much much better.

That itself is obviously something that just makes the product, you know, more and more powerful just for them and and is the kind of thing that would command that value.

So I I think that's it's it's, you know, from a perspective of is AI, you know, are we going to have monster intelligent AI that's just so smart, but we just, you know, we just don't we just turn the machines off because they're too expensive, you know, I don't think we'll have that future.

I think there's a there's a a very reasonable question about, you know, where that value capture comes in and and I think that's why all these businesses are thinking about, you know, what is this one thing that I'm going to do really really well that's going to make my business durable. It's fantastic.

Well, thank you so much for taking the time to hop on on a busy day. Uh, good to hear that the job's not finished. I love unfinished jobs. It's the best. I hate when I hate I hate when the job's done. I hate when the job's done. Congrats. Give our best to the team. We will talk to you soon. We'll do.

Thanks for having me. See you. Hi. Uh, how'd you sleep last night, Jordy? I had You had a terrible night, right? I had a brutal night, but not because of my aid sleep. I had No, I I got an 82. I'm sleeping pretty good. I think I get the sound effect. I managed to pull a 75 despite eight sleep is working overtime.

It was like, we got to get this guy some