I Went to Scout AI’s Bootcamp. They’re Training AI for Real Battlefields.

Colby Adcock’s Scout AI just pulled in $100 million to train its AI models for military use. That’s a lot of cash, even by defense-tech standards. But after visiting their bootcamp, I get why investors are throwing money at this.

Scout isn’t building another chatbot or image generator. They’re building AI agents that sit on a soldier’s tablet or helmet display and let one person control a whole fleet of drones, ground vehicles, and maybe even naval assets. Think of it like a real-time strategy game, except the units are real and the stakes are life and death.

I visited their training ground—a sprawling outdoor facility that looks like a cross between a paintball arena and a drone racing track. Concrete bunkers, mock buildings, open fields. The air smells like burnt grass and jet fuel. Not exactly Silicon Valley.

The demo was impressive but unsettling. One operator, a former Army sergeant, sat behind a ruggedized laptop. On the screen, a dozen drone feeds. He tapped a waypoint on the map, and three quadcopters lifted off simultaneously, adjusted formation mid-flight, and settled into a perimeter scan. No joysticks. No manual piloting. Just an AI that interpreted his intent.

Adcock told me the key insight was that soldiers don’t need to be drone pilots. They need to be commanders. The AI handles the flying, the collision avoidance, the sensor fusion. The human makes the tactical decisions. That’s the pitch, anyway.

Scout’s models are trained on massive datasets of real combat footage, simulated engagements, and even wargame scenarios. They’ve also built their own simulation engine that generates millions of edge cases—fog, electronic warfare, battery failures—so the AI doesn’t freeze when things go wrong.

The $100M round was led by a mix of defense-focused VCs and a few names I didn’t expect, like a major cloud provider. That tells me they’re not just building software. They’re building infrastructure. Probably a dedicated military cloud stack for real-time AI inference at the edge.

Is this a good thing? I don’t know. Autonomous weapons are a loaded topic. Scout is careful to frame this as “human-on-the-loop” rather than “human-out-of-the-loop.” But anyone who’s worked with AI knows that loops have a way of shrinking over time. The demo I saw had the human approving every engagement. But the speed of modern warfare might not allow that luxury forever.

Still, the tech is real. The funding is real. And the military is clearly hungry for this capability. Ukraine has already proven that small drone swarms can change the battlefield calculus. Scout is trying to make that scalable.

What I respect is that they’re not hiding in a lab. They’re running live drills, breaking things, and iterating with actual operators. That’s more than most defense AI startups can say.

If you’re in the defense tech space, this is one to watch. If you’re worried about the ethics, you’re not wrong to be. But pretending this tech doesn’t exist or won’t be used is naive. The genie is out of the drone hangar.