Behind the $61 Billion Startup Building AI Weapons for the Pentagon | Term Sheet

People get nervous around the idea of lethal autonomous weapons. The reality is, the U.S. has...

...incorrectly. Like if it's done in a completely insane way, it would be a very bad idea.

...them. The volume of weapons that these modern conflicts are consuming is so high.

Hello.

Pretty cool.

So I think of myself as like, you know, I got two little girls, so I'm a dad. I love that. I've always wanted to be a dad. I love that part of my life. And you know, I just think of myself as a builder. I just like to build things. Kind of doesn't matter what. Like I've been doing it since I've been, you know, a little kid, since I've been like five, and just always building something. And it's just curious and wanna just make something. Like that's just who I am. Sometimes it's a physical thing because I'm an engineer. Sometimes it's organizations, just whatever it is.

Yeah, so you know, well, the engineering was, you know, kind of always there, but the management side was interesting. Where like, you know, I came in as a forward-deployed engineer, one of the first ones on the team.

Yeah, yeah, no, sure. They were actually calling it that, forward-deployed engineer, maybe like six months earlier. And you know, I was working for this guy Mike Schaffer, who I knew in college, great friend of mine. And you know, I remember just like a handful of months in, he was like, I don't really like managing. You know, more or less, you want to swap? And I was like, all right, this sounds good. I ended up turning out I liked it. I liked, you know, leading teams. I liked being able to work on a lot of these organization problems that were just, you know, kind of tricky puzzles. And how do you get people to be motivated to do the right things? It's like one thing you learn with engineers is, you know, I see a lot of people who, you know, kind of come from like a business background or whatever, and they think that if they're in charge, finally the engineers will listen to them. It's like, no, they don't listen to anyone. You have to convince them of everything, and they've got to want to do it themselves. And so just really thinking about like organizations through that lens of like the people, what are their motivations? How do you just enable them to do epic stuff? Like that's been a super interesting theme throughout, you know, everything I've been working on.

Yeah.

It will be bleeped. It will be bleeped. Can we...

The thing I always value is that you see folks who just have this like creative spark, just have the right ideas, who are just like really driven to, you know, actually get stuff done, and that's pretty rare. So like just that creative ability. When you're building something new, I think people get confused a lot. Most people live in organizations where the company already exists, what they do already exists, and most managers, like most people, are just optimizing what was already created. This like process of creating a new thing, of, you know, defining, you know, a new business, everything has to be invented. So that invention skill is super rare. And so for what we're doing, that ability to just always be unsatisfied, really just pushing the envelope when they have an idea, it's just better than anything you can come up with. That's super rare. You can kind of tell. And for us, that's just a big part of what's made us successful. And then, how do you, you know, it's like nobody's like perfect at running a holistic organization. It's not how it works. Like, you know, there's so many dimensions to these things. The people who do really well are the ones who can be really self-honest about where they're good, where they're bad, and it's okay, right? Like, if you suck at managing people, but you're really creative, that's fine. Just defer that part to someone you can work with on it, or you're not great at operations or whatever it is. And so that self-honesty is probably the most important skill for any person to have, and then it bleeds over into an organization that is either actually reflective and self-honest and can critique itself and get better, or just has to lie to itself that it is great and everything's working. And organizations and people that lie to themselves, they don't survive very long. It just doesn't work out. And that's a big part. We've always tried to like drill in, I've always tried to drill in, where people just have to get adjusted to the fact that we don't do a lot of compliments, we don't tell you did a good job. It's mostly like what didn't work, you know, what could we do better, and then like how we're getting better next time. Like that's really all we talk about, which is really uncomfortable for a lot of people where they're used to like a reasonable degree of praise. Just not my style, not how we work, and it's once you get used to it, it's just really productive, though.

It's pretty hard. I mean, like the washout rate in kind of senior roles is probably like 50, 60%. That sounds high. I don't know. I mean, maybe it's like actually probably accurate to what these things should be. There's always a question of like, does it fit? Not just like, are you competent, but like, do you fit with how we want to operate, your peers, your boss? You know, it's just like the whole system, and that's just impossible to screen for. The like introspection and like how they think about problems, you can tell that pretty fast. Like if you met a lot of folks, you can kind of get a sense of who is just deeply introspective and thinking about what worked, what didn't, what was their contribution, what would they have done differently. That's normally pretty easy to suss out. But then, like, are they going to be effective in our setup? Are they going to be effective for what we need? That's always just, it's so circumstantial and like actually, you know, not 100% correlated with how good they are. It's just, are they good in this context?

No, no. Yeah, you're not sports.

Oh man, not gonna name names, because they're mostly still around. You know, a lot of folks were just like really creative, and maybe from a prior life, we had, you know, one person who is just like just absolutely brilliant engineer. And the challenge is always just, you know, can they get to that self-honesty and build a support structure around themselves? Can they take the help? Can they be honest about, you know, where they need somebody who's got complementary skills on management or whatever it is? And as they're going through that, it's often quite hard for the people that, you know, work with them or their peers. There's a lot of friction. There's a lot of difficulty as they like get, you know, like a lot of these people are like brilliant artists who just get offended that something's not perfect, and that causes a lot of friction and chafing with like people they work with. And that's, you know, that's what makes them great, but it also makes it hard for them to exist in these organizations. So, you know, how do you protect them, get them, you know, peace, plus get them to understand how they need to be most effective to operate, get the rest of the organization to understand like this person's brilliant in this way, you know, you're going to tolerate like these issues. Just suck it up, deal with it, because it's worth it.

Really good. It's so good. And I think most organizations reject these people because it's just like, it's too hard, it's too annoying. Is it really worth it? You know, if you're changing like 5% per year as a company, or as an organization, is it worth someone who's high friction? No, right? You're like, it doesn't really matter. For you, it actually is worth it. It sounds totally worth it. Why? It's just like the rate of growth, the number of new things we're going after, the new products we're building. Like all of these require someone who's just like a brilliant artist. Like you just can't make a new product unless it's like an actual, like someone really brilliant leading it. You get very mediocre results, right? Like you don't build an airplane by consensus or committee. It's not how it works.

There's a weird balance where you're trying to create an organization where people feel genuinely and are genuinely engaged in like their opinion matters and what they think matters, right? Like it doesn't work to just tell everyone like, you know, you just can't have an opinion on this thing. This doesn't work, right? It's like especially if you have a lot of, you know, brilliant people. So, you know, that balance of how do you create that real buy-in, get people really valuing their opinions, getting to a better outcome while still you've just got to be able to say somebody's in charge. We're moving out. They're making decisions. And I've screwed that up at different times in different ways, trying to, you know, manage the more consensus approach versus more like, you know, empowered, decisive approach. It's just genuinely hard, right? Like it's a genuinely hard balance. And, you know, these projects are always, anything we're doing is just like, you know, take the literal case of building an airplane, right? It's like, okay, well, there's like the design of the plane, but then there's like the testing program or the software approach or, you know, how we talk about it on business development or, you know, comm strategy. Each of those require a degree of like creative insight and how you're going to approach it to generate the right, you know, the right strategy for each of those dimensions. And it's just like, you know, it's not true that one person's going to be great at all of those things. So how do you create the right environment that the right experts, the right brilliant people, can just be empowered to run, and it works.

I think a lot of companies end up pretty thin where the founder ends up being the really like kind of brilliant artist and struggles to recognize that their job is to actually mostly just get brilliant people and, you know, empower them, right? Like that's most of what your job is. Not you being right; it's about getting the best ideas and getting the best out of people. And so the bench depth that a lot of companies actually like often ends up pretty bad, right? Where it's like the founder, the CEO, or CTO wants to be the person with all the ideas. I just think it's, if that's what you're optimizing on, if that's what you're subtly optimizing on, it goes really badly. You just can't do that much. So that ability to like really bring in and empower these brilliant artists, that's probably the number one skill. Like if you get that right, it tends to go pretty well.

At least with AI today, and it's sort of hard to tell where it's going to be in a year or two, AI has terrible taste. It's like really good at thinking through like the technical details of a problem or the mechanical aspects of it, but the intuition around, you know, what's good, what's bad on any dimension. You know, it's like if this is like you're building a software product and thinking about like the right architecture or the right way this is going to operate, that taste around what is this thing and what should it be and how should it fit is often quite bad and it just doesn't have the context and it just, that's not what it's good for right now. But then you're like literally implementing the code. It's great, right? And I've kind of seen this across lots of different things. We're even seeing with internal apps, you know, product managers are writing 80% of the code via Claude Code at this point. The software engineer is thinking about how's this thing going...

A year ago, no. Like it's changed dramatically. Like a year ago, the people who were getting the most out of the AI stuff, you know, talking with a bunch of different folks about this, they were a lot more of the like junior engineers, where, you know, they were already given fairly, you know, task-specific work, and it was helping them just write the code on the task work, right? Like fairly bounded stuff. Now it's like senior engineers. It's like the people who are like architects or product managers or any of these folks who are like, I know what I want this to do. I can think about what I want the outcome to be in a really high-end way and what good or bad looks like. Those people are doing incredibly well with AI right now. So it really shifted from a year ago. A year ago it was like good, but just not that impactful. Now, like, there's a reason Anthropic is taking off so much. Like how this is really manifesting in a very different way. So I think that's a big piece of it. The other piece is like, I don't think there's enough, you know, kind of honesty around, or kind of curiosity around how this is going to play out in the job market. I think it's sort of all doom and gloom. But there's going to be grades of this, right? It's like, you know, if you're a small business, it's probably in a lot of ways great for you. Where now all of a sudden you have access to world-class marketing talent, you have a finance team that's actually very good, right? Like you could start to do things that were never accessible to you before, and really amplify your reach. You know, is it going to lead to a real reflection on do we really need the, you know, marginal software engineer? Right? Can we get more efficiency? Yeah, that's going to be true. But, you know, what we're seeing is just we were always limited by the people, the rate we could hire. We had more ideas than we had bandwidth. Now that's like more unconstrained. Like that's a huge, huge difference. And so just the quality and volume of what we're going to be able to produce is just massively higher. And then the value that each person brings is going to be that much higher, right? Like we live in a world where doing more will yield more growth and a better outcome, right? Like better products, better revenue growth. For a lot of business, that's not true, right? Like if your back office is better or whatever, you know, it doesn't matter, right? Like you're still saturated in your...

Create jobs, on the whole. I think the efficiency you're going to get out of people in terms of the rate you can bring new products to market, you know, the ways you can get like kind of new experimental capabilities move through to production very quickly, that's going to change dramatically. And the part I think everyone agrees is over the next five to 10 years, the world of atoms is going to do quite well. So if you're bending metal, you're making hardware, there's just going to be a resurgence in, you know, kind of manufacturing. And I think the AI is also going to have a massive benefit on the manufacturing efficiency. So that's going to yield, you know, better jobs, more jobs where it actually makes sense to manufacture in the U.S., right? And we're in a lot of these more expensive locations. So I think it's going to yield significantly more jobs into the manufacturing space. I think you're going to see the software engineers and a lot of the kind of product managers, program managers, just be wildly more efficient. They're going to be doing more valuable stuff. And I think the, you know, costs, you're going to make it easier to adopt new things, which is I think net going to drive a lot of new jobs.

Yeah, so we don't make our own foundation models. We work with all the foundation labs, and they're great, and, you know, we have a good relationship with nearly all of them. The way we think about where AI is going to be most beneficial and sort of most ethically applied in defense, we kind of think of it in two ways. You know, one, you're going to be able to be more precise, right? So you think about, you know, people get nervous around the idea of lethal autonomous weapons. Well, the reality is the U.S. has had... the killer robots have people very scared.

...incorrectly. Like if it's done in a completely insane way, it would be a very bad idea. But nobody is proposing that. And...

I don't. And I think the reality is when we think about it, it's like the U.S. has had this for a long time. In the sense of, you have a torpedo that you say, hey, in this region, in this, you know, geographic area, something matches this acoustic signature, it's a military target. Or anti-radiation weapons: if it's emitting like a radar, like a military radar, and it's in this box, that's a military target. And so we've already had this notion of like you can bound it in a way that's, you know, it's not making arbitrary decisions; it's making very narrow, very precise decisions. And I think that framework has existed for a long time. The way you test that has existed for a long time, the way you think about employing that, the policy around it. And so extending this to something that's using, you know, say computer vision instead, or taking a bunch of different sensor data, it only gets more precise and more selective and more accurate. So I think that net is going to be, you know, it's hard to make the counter argument that we should have less precise weapons. Like it's just kind of a crazy position to take. And so, you know, I think the precision side of this makes a ton of sense to me. The other dimension is on dealing with, you know, kind of fog of war, the volume of information. How do you make better decisions? And again, there's sort of an irresponsible version of this, and then a responsible version of this.

Yeah, like so, the irresponsible version of this is, you know, humans aren't in the loop. The AI is just making arbitrary decisions. Like that's a crazy position. Nobody is arguing for that. And the fundamental to the way the U.S. policy apparatus works, the way we think about this, is that you have humans be accountable for the decisions they are making. So if a commander on the battlefield employs, you know, makes a use-of-force decision, right? Like they're engaging a target or they're, you know, firing a weapon, they are accountable for what happens there. And so, you know, AI can do a lot to cross-check their assumptions, right? Like, hey, let me run through and make sure that, you know, this is, you know, definitely a military target, or cross-check every database. You know, we can kind of review the decisions that are being made. It can synthesize a huge amount of additional information so they can make those decisions faster, but, you know, kind of fully formed. Because again, today the alternative you're comparing it to is, you know, fleets of, you know, 18- to 20-year-old enlisted folks who are going through and running a fairly mechanical process. When you think about a lot of these targeting or lethal decisions that are being made, and, you know, if you're a commander, how much do you trust the repeatability of that process versus, you know, I want all the information, I want to be able to ask the questions, and I want something adversarially reviewing my decisions on this? That seems like a very responsible world to live in. And so I think that is going to have a lot of, you know, kind of benefit where the decisions that are being made, you can have a lot more confidence that they've been reviewed from multiple aspects, they've been looked at in different ways, and do it really efficiently in a much more...