Factory raises $150M to help enterprises manage fleets of AI coding agents more efficiently

Speaker 1: Matan You dog.

Speaker 2: How you doing?

Speaker 1: You dog. You absolute dog.

Speaker 2: What's happening?

Speaker 7: Well. Happened? I am doing well.

Speaker 2: Tell us what guys are doing.

Speaker 1: How you doing? How big is this market? Because

Speaker 7: This is the is the most important market there is.

Speaker 2: It seems like it.

Speaker 1: It really seems like it.

Speaker 2: Break down the news for us first.

Speaker 1: What happened? It is with the news, and then we'll get into

Speaker 7: the details. Well, first of all, so I'm reporting live

Speaker 2: Oh, yeah.

Speaker 7: From and let's play a quick word association game. I'm gonna tell you where I'm calling in live from.

Speaker 4: Okay.

Speaker 7: You gotta give me a first first word that comes to mind. Coming in live from the Rosewood in Menlo Park.

Speaker 2: Oh, Drew Horowitz.

Speaker 7: Enterprise software. Come on.

Speaker 2: Enterprise software.

Speaker 4: The rose

Speaker 7: The temple the temple of selling enterprise software. I'm in here Yeah. Closing deals.

Speaker 2: But more

Speaker 7: importantly here to share with you guys about our latest fundraise. We raised a $150,000,000 from there

Speaker 2: we go.

Speaker 7: From the great folks at Coastal Ventures, Sequoia Capital, Blackstone, Insight, NEA, and some other some other great partners, and we're very, very excited to have their support.

Speaker 1: That's amazing. Talk about the last, I I feel like it's at least three months, maybe a bit more since the last time you were on the show. Talk about how the how the space is evolving. I mean, there's just so much noise. Like, every single day, you know, somebody saying Battling this entire paradigm is dead. You guys clearly have just kept your head down and Yeah. And are are making a lot of progress. But yeah, walk us through kind of how the how the space is evolving, how how your business is Yeah.

Speaker 7: Yeah. Absolutely. So I guess, first of all, I think, you know, there's obviously a lot of excitement in this space because there's just so much to be done in terms of software development. There's so many things that software developers don't enjoy doing, but they unfortunately have to be doing. Mhmm. And I think there was a there's kind of an early phase of excitement where everyone was just, you know, thinking about all the possibilities and kind of investing kind of the resources both from the the financing side, but then also in the enterprise side in terms of, you know, making sure they stay up to date. But I think there was a bit of a lag where first, you know, they were like, hey. Here are all the things we want to do. Let's go and tell our developers to actually do it. And there's kind of a lagging period where they didn't do it. And then in the last six months really, everyone started adopting Agentic AI. You see that in the revenue numbers of every Yeah. You know, model provider. And I think now there's a bit of a a bit of a hangover where, you know, some people are realizing that they wanted their engineers to go and adopt things and adopt what they did. However, they might not be doing so in the most efficient manner. You know, there are large enterprises that we work with that found that on a monthly basis, they spend on the order of hundreds of thousands of dollars on developers saying things like hello to Opus 4.6 fast or GPT 5.4, like ultra high

Speaker 2: Yeah.

Speaker 7: Which, you know, is probably not the best use of those tokens. And I think, you know, a tool like Factory where we can be model agnostic and dynamically route them to the appropriate model for the appropriate task

Speaker 2: Sure.

Speaker 7: That has been something that's been really getting a lot of enterprise excitement.

Speaker 1: Yeah. The on another token this token kinda token maxing trend, Meta Meta, of course, was they love to spend money on on AI. They've they had their internal leaderboard to see who could basically produce the most tokens. There have been some chatter that that people were had effectively just created loops to just, like, go to get to the top of the the chart even though I'm not sure it's an award worth winning. Do you think that, like, large corporates are already, like, hey, this or, like, do you think this is a period where it'll be, like, for the first half of this year, people are, yeah, just try a bunch of stuff and then we'll see where we land? Or do you think there's already gonna be more of a pullback? I know the Uber CTO had had, went on the record and was talking about, like, hey, we basically spent our whole inference budget, in in in, you know, the first quarter and we gotta figure out what our plan is on a go forward basis.

Speaker 7: Yeah. I mean, I think what we're seeing is that every company kind of has to go through these phases. We're like, phase zero is reluctance to adopt because, you know, developers have their workflows that they've that they've had for the last twenty years. They might not wanna change it. And phase one is kinda throw the kitchen sink. Just use as much AI as possible. Whatever you do, just change your workflows. And then I think a lot of people and maybe more of the frontier companies like Uber, who's always kind of very ahead of the curve as it relates to these things, now getting into the phase two, which is Meta, obviously, is an example as well. Okay. Great. People are now adopting. We're proving that they can change their behavior and use these tools. Now we need to make sure we're actually doing it efficiently and effectively. And I think that's it's fine as like a natural process. Yes, you'll overspend in that phase one, but the point of that is to get to the phase two where then you're actually really efficient on a on a per token basis, moving the needle for for the business, delivering software faster.

Speaker 1: What's going on international? We saw you we saw NSF, I believe the day of the Super Bowl, and you talked about going international. How how are companies and and developers abroad thinking about CodeGen and and the category broadly?

Speaker 7: Yeah. Absolutely. I mean, think one thing that's really important is that, you know, people build software around the world. Even though the best software is probably based in The US, there's still some fantastic software out there. And so I think, you know, the way we've built Factory in particular is amenable to, you know, places like Europe or Asia or Australia because, you know, they have different rules about where the data needs to live, where the models need to live. And factory is one of the only solutions that is fully on prem, fully modular. So, you know, my cofounder, Eno, likes to say you could deploy factory on a nuclear submarine as long as you have GPUs down there. And, you know, in places that tend to have a lot of regulation like Europe, that actually works quite well. And so with some of this funding, we're we're opening up an office in London and

Speaker 1: Air Horns for European regulation. There

Speaker 4: we go.

Speaker 2: Can you talk about some of the projects that you're seeing speed up on the back of AI agents? Because there's this weird dichotomy where we see huge token budgets, huge AI spend, a lot of a lot and then you you dig in and you see people tell stories about building a lot of internal tools, a lot of new dashboards, automating workflows, being more efficient. But people it feels like they haven't really felt the external facing I don't know. It just feels like if you like, Meta is using a bunch of AI. If you open Instagram, it's sort of the same app. It's not like, oh, wow. They have, like, an entirely different paradigm, and it feels like an entirely new app. And maybe that's just because that particular platform is mature. But where where is like, in terms of internal tooling, customer facing software, entirely new ideas, automations, like, where are you seeing the most adoption, most impact?

Speaker 7: So I I I love the framing of that because I think there's kind of two separate types of usage that we see. There's one that's, like, the fun and cool stuff of, like, let me build all these new apps from scratch Yep. Which probably doesn't move the needle for the business.

Speaker 2: Yep.

Speaker 7: And then there are the less sexy things, but actually save developers a ton of time, and that's where we're seeing kind of under the hood a lot of the ROI coming from. And I think generally the name of the game there is, you know, developers are really smart. They spend years of their lives becoming experts, and they get paid a lot of money. Mhmm. They should not be spending their time on low leverage things like documentation or testing or, you know, to spend two years doing legacy code migrations. So generally, the the orgs that we see get the most ROI are the ones who basically play whack a mole with what is the lowest leverage thing that our developers are doing right now. Great. Let's use droids to automate it.

Speaker 2: Mhmm.

Speaker 7: Or, you know, some language that's really emerging is let's build a software factory

Speaker 2: Yeah.

Speaker 7: That goes in and automates these low leverage tasks. And I really think we take a lot of inspiration, and it is even where our name comes from is what Elon and Tesla did to the physical factories. Like, if you go to Tesla factories, they're mostly like, you know, machine arms going and doing things, and there's no reason that software can't be very similar. Like obviously, you still need humans involved, but those humans tend to play a role of more designing that software factory, figuring out what is the most efficient way to to configure that factory so that you can move the needle on your business, you know, produce more more output.

Speaker 2: How important are are is wide diffusion of models? We we've been following the the latest model from Anthropic Mythos. That's only available to a few companies. Is the strategy to focus on being cross platform or get access to that model earlier and then act as a diffusion layer for that? How are you thinking about the future where more advanced models are sort of gated?

Speaker 7: Yeah. I think for us, the biggest thing is enterprises need to be model agnostic. It's just it is a non negotiable that they cannot standardize on just one provider for a number of reasons. One might be, you know, if that API endpoint goes down, which, you know, some models these days haven't been the most reliable. And if you're a highly regulated industry and you get your developers to become agent native and, you know, sort of delegating tasks to these agents and that endpoint is down, what are you going to do? Like, it's just it's nonnegotiable. And so being model agnostic for every serious enterprise is kind of table stakes. And for us, what we can do is make sure that we get those models, you know, day zero that they're released just so that they can stay at the frontier. Same with the open models because also different models end up being good at different tasks. They're better at different languages perhaps.

Speaker 2: Mhmm.

Speaker 7: And making sure that we kind of give the optionality to the enterprises to to adjust accordingly Yeah. Is pretty important.

Speaker 2: How are you thinking about the forward deployed model? Is that more important now? I I know that you have a very small team. It's, what, 70 employees and But probably growing very how much about, you know, enterprise adoption is actually spending time on-site with the customer, answering questions, integrating deeply into these, like, large enterprises?

Speaker 7: Yeah. The way we think about the the deployed engineering is they should never be doing the same thing more than once. Because with the tool that we've built, if they've learned, hey. Here's something that enterprises care about, we should be able to build that into the product very quickly. Like, our core competency is not providing services, but it's building product. And we treat our deployed team kind of as like the tip of the spear where they're figuring out live, you know, shoulder to shoulder with the enterprise engineers. What are the things that if we put into our product would allow them to become agent native faster?

Speaker 2: There's some news from Cursor today that they're teaming up with XAI to potentially train the next version of Composer. Are you thinking about training your own model at some point? Is that interesting to you?

Speaker 7: I think it's interesting at some point. I think it it doesn't make sense right now because I don't think there is a I don't think right now enterprises need another, you know, open model that you fine tune.

Speaker 2: Sure.

Speaker 7: I think there's there there's sufficiently good ones out there, and I think most of the alphas actually on the research side as it relates to the agent itself.

Speaker 2: Mhmm.

Speaker 7: So for example, Droid, which is our agent, ends up so it's model agnostic, but it also outperforms all of the agents that are coming out of the model labs.

Speaker 2: And

Speaker 7: so as an agent lab, most of the alpha on every incremental hour of our time is from the agent itself. And then we'll probably get to a point, you know, where eventually maybe there's just there's sufficient alpha in in, you know, RL ing a model for ourselves. But right now, every incremental hour on the agent ends up being very, very high ROI.