How AI Is Changing Warfare with Brian Schimpf, CEO of Anduril

0:00The American defense industry is the largest in the world at nearly $1 trillion, accounting for about 40 % of military spending around the world, and also arguably impacting every person on Earth. Now this sector has also phase shifted throughout the decades, including the consolidation of crimes, shrinking from over 50 to less than 10 large crimes receiving a majority of defense dollars. Those are companies like Lockheed, Raytheon, or Billing. But there are some new companies in town, trying to disrupt how defense is done through new hardware and software. One of those is Andro, a company that just announced Arsenal 1, a billion -dollar factory in Columbus, Ohio, expected to create 4 ,000 jobs in the region.

0:48Now in today's episode, Andro Co -founder and CEO, Brian Chim, sits down with A16C Growth General Partner, David George. Together they discuss how Andrew got its first product off the ground, competing with some of the largest companies in the world, in navigating the US government's complex procurement processes. They also discuss how AI changes the modern battlefield. Being able to pull out that signal from this overwhelming amount of information that it lists. Plus, what most people get wrong about these technologies. It is unethical to not apply these technologies to these problems. And how we shape up to the competition.

1:27They are running hundreds of tests a year of hypersonic weapons. Right. US is running like four. Now, if you do like this episode, it comes straight from our AI Revolution series. So, if you miss previous episodes of that series, with guests like AMD, CEO Lisa Sue, and Throbbit co -founder Dario Amade, or the founders of companies like Databricks, Waymo, Vigma, and more. Head on over to a16z .com slash AI Revolution. Alright, let's get started. As a reminder, the content here is for informational purposes only. Should not be taken as legal, business, tax, or investment advice, or be used to evaluate any investment or security, and is not directed at any investors or potential investors in any A16z fund.

2:11Please note that A16z and its affiliates may also maintain investments in the company's disgust in this podcast. For more details, including a link to our investments, please see a16c .com slash Disclosures.

2:36Let's jump right in. What is the end roll? Tell us what you do. Alright, so we were founded 2017. We're about seven years in. The basic idea was we thought there was a better way to make defense technology. So number one, the tech for the next 20 or 30 years was going to be primarily. How do you just have more cheap autonomous systems on the battlefield, just more sensors, just more information flowing in? That seemed like it had to be true. So we invested in the core software platform we call lattice that enables to make sense of all these things. We have built a variety of autonomous products that we've fielded over the last seven years, just an outrageous pace.

3:12And we're really working on all aspects of national security and defense. And how did you guys get onto national defense as the place to go spend your time? Obviously, I know your background, but maybe you can share that. Yeah. So I was at Palantir for about 10 years. I had been working on a variety of government programs. And then several of the co -founder, so Trace Stevens, was also a Palantir Mac Grimm, our COO, as a Palantir. We're all really good friends. And we've been talking about doing this idea of there needs to be a next generation defense company. And then Tray and Palmer Matt through VcWorld and Palmer was just getting out of Oculus and did the same thing I wanted to do.

3:46And so we decided to kick this off together. But for me, working in defense, it was just obvious the degree to which there was a problem. You work in this space, the tech is old, it is not moving fast, it is very lethargic. There are relatively few competitors at this point. It just felt very right to do something different. And it's the sort of thing that once you get into it, the people who are actually serving just have this patriotic motivation to solve the problem. It's just very, very motivating problem to work on. How did you land on the first product? So first product we worked on was what we call centuries for border security.

4:21And this was a Palmer idea. He believed that tech could actually solve this. So we have these automated cameras with radars. We can monitor the border for miles away from these cameras. And he was like, this is something we could solve super fast with technology. And it really kind of hit what has ended up being a very good pattern for us is find an urgent problem that actually has a real tech solution that we can apply the cutting edge technology too. So early on in 2017, computer vision was just starting to work. It wasn't even really embedded GPUs yet. We were literally taking desktop GPUs and liquid cooling them to get these things to work in a hot desert, under solar power.

5:01But we were able to go and get a prototype up in about three months and then move into like a pilot in about six months and then full scale in about two and a half years. So really, really quick timeline. But I got to fit this problem set of, we had a technical insight of how you could do this better. And there was urgency to solve the problem. They actually wanted to make a dent in this. All right, I'm going to ask you a lot more about that stuff. But one of the things that people say to me all the time, and you hear it in speeches and all this stuff, like AI is going to change the nature of warfare.

5:28Yeah. On the one hand, the major breakthrough that we just had, the way everyone interacts with it, is like a chatbot and an LLM. It's pretty cool. It's amazing. It's awesome. I use it for everything. But what are the implications of this new wave of AI, gender of AI on modern warfare, physical sense on the software side? Just talk about that. So when I think about where AI is going to drive the most value for warfare, it is dealing with the scale problem, which is really the amount of information that is the number of sensors, the sheer volume of a system that are going to be fielded, it's going to go through the roof.

6:03So this is like lattice maybe start even there everything has a sensor That's right. So like what do people do in the DOD? There's a lot of things they do but what's the primary war fighting function? They are trying to find where the adversaries are yes, they need to then deploy effects against them that can be Strike that can be deterring them by a show of force that can be jamming and non -connected things and and they've got to then assess did that actually work. Right? Find them, you got to engage, and you got to assess. It's like pretty straightforward. That is the primary thing that the military does.

6:39What do you need to do that? You need a ton of sensors. You need a ton of information on what is going on, with an average who's constantly trying to hide from you and deceive you. Yeah. So just huge amounts of information to make that problem as intractable as possible for them to be able to hide or when they are deceiving, you can figure it out. Yeah. Yeah, yeah, yeah, yeah. I just even every single phenomenology of sensing. This technology exists, right? The sensors exist. The sensors all exist. The sensors are deployed, right? They're gonna get better and you're gonna be cheaper and you're gonna be able to do more of them, but a lot of the limit of why can't we do more is what color you're gonna do with the data?

7:10Processing capabilities. Yeah, processing, but also just operationally. So okay, now I say I had a perfect AI system that could tell me where every ship, aircraft, and soldier was in the world. What are you gonna do with that? Now I know everything that is overwhelming. right? And so then being able to sit through that information to, well, okay, they're maneuvering here. What does that imply? Is this an aggressive action? Is it outside their norms? Is this different than we've seen in the past? Being able to pull out that signal from just this overwhelming amount of information that exists. And then the other side, you got to act, right?

7:43So now I've got to actually be able to carry out these missions. So this is where on the autonomy side, it really comes in, which is, okay, I want to send fighter pilots out. So the way they do like a predator drone today is like a guy with a the joystick. Yeah. No, we've all seen the, you know, Ukraine Russia. Yeah, exactly. It's all like manually piloted. But that doesn't really scale. And there presents a lot of limitations on communication shaming all these things. So I want to be able to task a team of drones to go out and say, Hey, go in this area and find any ships and tell me where they are.

8:11I just wanted to be that simple. They just need to figure out their own route. And if I lose some of them, they rebalance, they just go out and handle it. Yeah. And target recognition, they can pop back whatever's relevant. That is where I think the autonomy side really comes in, which I can just drive scale into the number of systems I can operate in the environment. The promise of AI in a lot of ways and the long run with this is just the ability to scale the types of operations I'm doing, the amount of information I have. And if done very well, it will put humans into a place of sort of better decision making, right?

8:40Instead of being like inundated by a volume of data and then all of our capacity goes to these mechanical tasks. We can have humans with much better context, much better understanding, historical understanding of what this means, what the implications of different choices are, those are all things that AI can enable over time. Ideally, better decision making. I think it goes through your full -time decision making. We're working with both limited information and imperfect judgment. That's right, I guess, right? Yeah. And so the more you can have AI augmentation for these things and synthesis and clarity, that is where the promise of this is.

9:13And so the US posture on this is very much we want to have humans accountable for what happens in war. Yes. That is how it should be. right? The military commander that employs a weapon is accountable for the impact of those weapons. Yeah. That is correct. I think that is the system we should have. And so then nobody is talking about having full, blown AI is going to decide who lives and dies. That is a crazy version that nobody wants to have. Well, I think it's also far -fetched in the sense that it presumed some sort of objective function that is in terms of my... This is my conversation with everybody when they come and they're like, oh my god, what about when the AI goes terminator on us?

9:49I'm like, it's a tool for humans. It doesn't have an objective function. That's a leap that is not on the scientific roadmap today. That's right. Like why would that be the case in warfare? That's right. And so I think the reality for these things is it's going to be human augmentation and it's going to be enabling human software to much larger scale with much higher precision on these things. And that is the opportunity with it. Yeah. And so to me, it is unethical to not apply these technologies to these problems. And I think our view has always been, We're the best technologists on these problems where we can get the best technologists to it, giving the best tools on these absolutely critical decisions that are extremely material.

10:26That seems like probably a good thing. And engaging in the question of how can you use this technology responsibly and ethically is incredibly important. Yeah. Is it more humane to have a fighter pilot in the way of danger or having an autonomous system? That's right. Piloting in a conflict. That's right. And by the way, I have friends who are fighter pilots, I love fighter pilots. Yeah, but you know the technology has advanced significantly and you can make the argument that it is more humane Not to put them in the line of fire. That's right. Yeah, we're not gonna want to put US troops at risk Yes, and I think those are the turn specter of the US saying I have this capability and I've reduced my political Cost of engaging on these things.

11:04It's actually a pretty good deterrent as well. Yeah, I'm not putting US troops Or I can give this to Alice. Yes, and they can defend themselves. Yep, and so I keep it out of the fight Yeah, exactly. You're keeping our troops out of the fight. Keep the troops out of the fight. And it changes the calculus quite a bit. And so I think that actually in a lot of ways, if done well, has a significant kind of stabilizing impact and a turn impact. It's just as harder to use force to get your political ends. Yes. And I think that can be a very positive thing. Yeah, exactly. Yeah. I keep coming back to deterrence.

11:34And we need to find a way to create a sense of urgency for the sake of deterrence. Yeah. And that for the sake of going to war. That's right. And so it feels like that's universally like a people we talk to I feel like that's universally known and hopefully we can Make some progress. Yeah, I think people largely agree what Vladimir Putin was very convincing of this Yes, like turned out invading Ukraine was probably the single biggest Shift I've seen in terms of people recognizing that look there are still bad actors in the world Mm -hmm They will use force to get to their political well if they think it will work Yeah, if the cost is worth it.

12:05They're gonna do it and I don't think there's any reason to believe that's gonna stop I have sure for tens of thousands of years. Do you think the future of warfare so you set AI as an augmentation for humans? Yep. How fully automated do you think a conflict can become say in the next 10 years? Look, I think the mechanics of okay, there's this airfield and you want to go surveil it and take it. You can do some strike, you can do some surveillance, you can do all these things. there will be a large degree of automation in that. Right? Like I can just say, hey, send this team of drones out in these waves to go conduct this operation, find things that pop up that are a threat.

12:45Tell me if you pop up to the human to say engage or not, it goes. Yeah, right? Like you can move in a much faster pace. I think a lot of the things that were starting to happen in Ukraine, a lot of the great work Peltar did was on things like this, where it was like the targeting process of going from satellite imagery through to, hey, this looks like a tank through to an approval of is this a legitimate military target or not? I was streamlining compressed. Much faster, much faster. Much faster. So I think those things will happen very, very quickly, like very, very quickly. Then, okay, now it turns into a matter of policy and degree and scope.

13:18That is a thing that I think we're just going to have to figure out as we work through it with the military. So we think about it from the technology side is, okay, I don't want to design anything that precludes more advanced forms of this over time. I architected it to correctly, but the crawl phase is just get a lot of the basics, just automated, very mechanical things, make it very predictable, don't have any surprises. And then you can add more sophistication as you build trust, the AI advances, these things get more sophisticated over time. And one of the best examples is on the defensive side where it's right for AI's.

13:48We do a lot of work on kind of drone systems. This is one of the areas we're partnering with OpenAI on. And it's looking at this question of if you have multiple drones flying at you and you have minutes to respond before the strike happens on you. How do you make an optimal decision? How do you make an optimal decision when you are panicked, you are nervous, and your life is at risk. It's very hard. Is that a person manually sitting there in those decisions today? Yeah, yeah. It's often three, because they have a separate radar from a camera or separate from the guy pulling the trigger around the weapon systems.

14:19And so then the coordination cost can be significant. So you can automate a lot of this. And then the other problem with this is then, as we see in Ukraine, every single unit, every single soldier is now at risk of drugs. Right? So this has to proliferate out from being a specialty that you do in an operation center now to every vehicle in the field. In the field, everyone has to have this capability. You need the ability to have these systems just process all that centered data automatically, fuse it together, tell you viable options for countering this, tell you what's a threat, what's not a threat.

14:51Like these are the types of things you need to be able to do, responds with intelligence suggestions and then have the system just automatically carry it out from there. Yeah. These are the types of problems we're working on. In the defensive side is just you need it, right? There's no choice because the timelines are too short and the urgency is too high. Yeah. And it's a very straightforward area to understand where technology can really improve the problem. Yeah, it's like the highest stakes version of decisioning that autonomous driving cars are doing today, but with way more sensor information.

15:25Yeah, it's not a road. Yes. Yes, with an adversary who's constantly trying to fool you, to see you. And yes, it's very, very hard. So that's one of the big parts of the partnership with OpenAI. Yeah, yeah. So they've been great. And I think Sam, especially, has been very clear that he supports our war fighters. And he cares about having the best minds in AI, working in national security. And who better exists to work through these hard problems? Yeah, and so I was just incredibly proud of them for coming out in favor of this and saying they're gonna work on this They're gonna do it responsibly.

15:57They're gonna do it ethically But this is an important problem that the best people should be working on the defense industry is notoriously difficult for startups to navigate so How did you guys actually get traction in the first place and? Do you think that's gonna change in the future? Do you think it will continue to be hard? Do you think the primates will continue to have a stranglehold? I'd love your take on that. It is very hard. And I think we built a lot of the right technology, and the right business model of investing in things that we believe need to exist. I think we're picking a lot of the right problems to go after.

16:33But probably more than anything, I think we understood the nature of what it took to sell. And the congressional relationships, the Pentagon relationships, the military relationships, like all of this that you need to be able to say, hey, we have the right tech. You can trust us. we can scale, we can actually solve these problems for you, proving that it works and then like, catalyzing all of these really complex processes around it. I think the other part that we've done quite well is we're just finding ways to find those early adopters and we understand those playbooks. Who's going to move quick?

17:05How do you just build that momentum and advocacy in the government to make this go? Look, it's like more bureaucratic in certain ways. Is it much worse than selling to a bank or an oil and gas company? I don't know, maybe 30 % worse, but like probably not 5x worse. Yeah. And I think the reality is it's like enterprise sales are actually very hard. Yeah. Especially the ones with long sales cycles and massive commitments. That's right. These are the large capital investments customers making. That is a slow sales cycle. Yeah. That is how it works. And so I think there's like a lot of complaining and frustration.

17:35It's okay. Well, also being bad at business means you're bad at business. If you don't understand your customer, you're going to lose. Yeah. That's how it works. Yeah, yeah, yeah. So do I think the government needs to be a better buyer of these things? So I think they need to like take better strategies. It'll get them more what they want. Absolutely. They're taking observably bad strategies to get to the outcome they actually want. Do I think it's necessary to change for us to be successful? Not really. We're just gonna play the game that they present. Okay, I wanna talk about the observably bad strategies.

18:04Yeah. What are the observably bad strategies? Right. And then what are the good ones? And maybe also wrap it into this idea that how do you actually convince the government that your ideas are the right ideas? So, take, should you go spend money building a whole new generation of F -35s with man pilots or a whole new generation of aircraft carriers, or should you do something different and how do you actually get your points across to them? Okay, so they're sort of like, how do they contract and buy and what's been going wrong there? And then it's, what's the right composition of even if you could buy perfectly well, what should you be buying?

18:38What should you buy? Yeah. Two different questions. Yeah. And so how are they buying poorly? So the typical government contracts are done in what's called cost plus fixed fee. And this actually came out of World War II when we were retooling industry to work on national security problems. It was like, we're going to cover all your costs and we're going to be a fixed profit percentage on top. And so the incentives here are sort of obvious. If it's more expensive, you get more profit. If it is less reliable, you get more profit, right? The longer it takes, the longer it takes. The longer it takes.

19:05The less reliable it is, the more complicated it is. there's no point of incentive in there to actually drive down costs. And you see this play out, right? It's like the companies have gotten so used to this where you look at even something like Starliner, where, you know, SpaceX had a third the amount of money that Boeing was given to make Starliner work. And SpaceX did it on time probably faster than they even predicted. Did it probably incredibly profitably, and it worked. And so I think these incentives that don't hold you accountable are actually bad for your company. It just makes you a worse company.

19:36But do people in the government realize that it's bad for the country? I think they are frustrated. I think they understand that this is not really working. So you look at like F -35 as an example of one of these programs that took 25 years to get it from initial concept to fielding. There's this, that's awesome chart, which is like shows how long it takes to get commercial aircraft or autos from kickoff to fielding. And it's been like flat to slightly better for all of those things, like on the order of two to three years. The military aircraft side just went linearly straight up. Like these things are taking longer and longer and longer.

20:11There's an amazing quote that if you extrapolate this out, this isn't the 90s, this guy made this quote. If you extrapolate this out by like 2046, the US government will be able to afford one airplane that the Air Force and Navy share and the Marine Corps every other day of the way gets a better. It would be a good airplane. These things are just crazy. And so I think they recognize that this is not working, right? This is broken. Now, the other part of this, they haven't had a lot of alternatives. So you have a relatively small cartel of these companies who sort of all say we won't do fixed price programs anymore.

20:41They won't do things on a fixed cost basis. So okay, if you're the government nearby, what are you going to do? You don't have a lot of choice here. And there's been a lot of problems with trying to get this model right. Now, in terms of things that can work a lot better, I think SpaceX really proved this where they literally built a reusable rocket that you catch with chopsticks commercially. Like that is, I think we could solve these things, guys. What thing can't we build? I think we can build an airplane. I think we can build it. So there's not really a question that this is the only part of some magical thing that can only, like this is the only people can do it anymore.

21:18You guys with autonomous, it's proven now that the future is here. Yeah, yeah, yeah. Exactly. And so I think the alternatives are there now. And then models that can work a lot better. It's like one of my crazy ideas is new missile It takes about 12 years to go from concept through to fielding, like 12 years. That's insane. It's insane. And so, okay, if you're in that world, how fast is the technology evolving? Oh, like, this is the like 12 years right now. Like, what we'll be able to do, right? Exactly. And then we'll still be on the previous system. No, there's even crazier examples like the Columbia class nuclear submarine is going into service in 2035 and it's expected lifetime, this year, 2085.

21:54So how good were we in 1960 at Guest Academy? Where were we today? It's like, it's unclear. It's unclear. Yeah, we had technology to go to the moon. Yeah, it's very good. It was like the quality of the phone. Those are the computing power of the phone. Yeah, and so these timelines get longer and longer and this is death spiral these things. Contrast that cycle of development with China. Do they take 12 years? How is their tech stack up to ours? The single best stat for this is they are running hundreds of tests a year of hypersonic weapons. The US is running like four. Right. Anyone who's worked in technology understands the compounding value of iterating on these things.