# Anduril started with a deployable product, not its ultimate ambitions

Markdown copy of https://realalpha.app/s/anduril-started-with-a-deployable-product-not-its-ultimate/ on Real Alpha, the best minutes from long interviews with the people building AI. Cite and link the page, not this copy.

- Speaker: [Trae Stephens](https://realalpha.app/builders/trae-stephens/) · Trae Stephens is cofounder of Anduril Industries and Partner at Founders Fund. · Co-founded Anduril, building autonomous defense systems on a shared software platform, and invests in government-focused startups.
- Host: [Aditya Agarwal](https://realalpha.app/people/aditya-agarwal/) · Partner, South Park Commons · One of Facebook's first engineers, he helped build its first search engine; later CTO of Dropbox.
- Show: Minus One
- Episode: [What Nobody Tells You About Building Defense Tech Companies | Trae Stephens](https://realalpha.app/v/wlH0lfkLRRU/), Nov 5, 2026
- Moment: 34:20 to 39:15 of the episode (5 min), playing on https://realalpha.app/s/anduril-started-with-a-deployable-product-not-its-ultimate/; the episode from there: https://realalpha.app/v/wlH0lfkLRRU/?t=2060

## Quote

> “The company was started in June. We had a **test unit in September**. We had first towers deployed in December.”
> — [Trae Stephens](https://realalpha.app/builders/trae-stephens/) at [38:19](https://realalpha.app/v/wlH0lfkLRRU/?t=2299) · Minus One

## Takeaways

1. The team chose a product it could deploy in six months.
2. Computer-controlled sensors required a more accurate pan-tilt unit.
3. A test unit came in September; towers were deployed in December.

## About this segment

Anduril began with a counter-intrusion tower it could field quickly, then built components when existing hardware fell short. The first government program followed within 22 months.

From “What Nobody Tells You About Building Defense Tech Companies | Trae Stephens”, Nov 5, 2026. [Watch the full episode](https://realalpha.app/v/wlH0lfkLRRU/)

## Transcript

Transcribed by machine (Whisper) from the episode’s audio, so a word may be off here and there.

[34:24](https://realalpha.app/v/wlH0lfkLRRU/?t=2064) kind of hardware platform that kind of works across, you know, autonomously across a bunch of different form factors, but what did that mean actually in terms of building the first product or the first set of products? Was it that you needed to make sure you're working with the same software foundation across all of the different things? And how is that, you know, I haven't worked in defense, I haven't really worked at a hardware company either.

[34:47](https://realalpha.app/v/wlH0lfkLRRU/?t=2087) So what does that mean in terms of what you had to do for the first couple of years? Yeah. So we had this like Google Slides pitch deck that we used for our seed round. This was before there were three rounds before your seed round. It was just our first money in was our seed round. Old school. Yeah. So weird. And there was a slide in this deck where we had like a list of all the things we wanted to build.

[35:08](https://realalpha.app/v/wlH0lfkLRRU/?t=2108) And it was like crazy stuff. I mean, it was like fully autonomous fighter planes and like submarines, autonomous submarines and like low cost cruise missiles to compete with Patriots, like all this crazy stuff. We had this recognition that SpaceX did not build the Starship on day one. And this is actually an important lesson for entrepreneurs. SpaceX did not build Starship from day one. They built Falcon 1.

[35:38](https://realalpha.app/v/wlH0lfkLRRU/?t=2138) They built a rocket that had a single engine, and then they built Falcon 9, and then they did reusability, and then they did Starlink, and now they're doing Starship. It's an iterative process to get to Mars, but it's not a one-step jump to Mars. And so that was the approach that we took, as we said, we know that all of these things are going to require Jarvis from Iron Man. We need that AI brain, but we also need a quick way to deploy that, that we're going to be able to test on real world data, that we're going to be able to deploy quickly, that doesn't have this like massive basic science problem associated with it.

[36:14](https://realalpha.app/v/wlH0lfkLRRU/?t=2174) And so we wanted to build something that we could get deployed into the field in six months or less. And so that was really the origin as we build this counterintrusion product called Sentry. That's just a, it's a camera on a stick. It's a little bit more complicated than that. But basically, solar power, a big steel pole, solid-state metamaterial radar, an infrared camera, like a standard EO optical camera, a pan-tilt unit, and compute at the edge with comms.

[36:43](https://realalpha.app/v/wlH0lfkLRRU/?t=2203) And this sounds like not that complicated. It turns out it's really complicated. No one had really deployed, like, NVIDIA GPUs at the edge with bad comms before, doing, like, object recognition. So doing edge inference, basically. Yeah, doing edge inference, doing like object detection, tracking at range in like weird desert environments with, you know, environmental distortion from temperature. Like it was really hard, but it was doable in six months.

[37:10](https://realalpha.app/v/wlH0lfkLRRU/?t=2230) There was no basic science. In fact, at the time, we didn't even think we needed to build any of the components ourselves. It was just a hardware integration problem. We ended up having to do that, which is a fun story that I'll tell. So pan tilt units, in theory, it should be the simplest part of the entire stack. It literally is just like a thing that moves the sensor payload. But it turns out that if the normal way that a pan tilt unit is operated is with a joystick, where it's just like you move until you see something and then you stop and you look at the thing.

[37:39](https://realalpha.app/v/wlH0lfkLRRU/?t=2259) And if you miss, you just slew back a little bit, a few degrees in the other direction until you have target on. We had a computer doing all this. And so the pan tilt unit, we'd say go 117 degrees. it would turn 116 degrees. It would just be like a few clicks off for some reason. Now you multiply that against 100 movements within an hour. You're no longer even looking remotely close at the thing that the radar is telling you to slew to.

[38:05](https://realalpha.app/v/wlH0lfkLRRU/?t=2285) So no one had built a pan-tilt unit that needed to have the level of accuracy that we needed it to have because we were using a computer to control it. So we had to build our own pan-tilt unit. Super annoying, wish we didn't have to do that. This is still the philosophy of the company today. We don't want to build hardware. unless we need to build hardware. And we were able to get that deployed. The company was started in June.

[38:25](https://realalpha.app/v/wlH0lfkLRRU/?t=2305) We had a test unit in September. We had first towers deployed in December. And then 22 months from founding, we had a multi-hundred million dollar program of record with the federal government to operate Century Towers. And that program is still running today. It's worth billions of dollars. We have hundreds and hundreds and hundreds of towers deployed at military bases, national borders, maritime environments, both in the U.S. and abroad.

[38:54](https://realalpha.app/v/wlH0lfkLRRU/?t=2334) And it's the best margin profile of any product in our entire portfolio, which is with maturity and, you know, you can tighten things up. That's the way that all of our products should work over time. But the key lesson from that is that we got to build the software that ended up being the brain that goes into everything else that we build.