Coby Adcock’s Scout AI raises $100 million to train its models for war. We visited its bootcamp.

At a US military base in central California, four-seater all-terrain vehicles roam hillside trails. This is a training exercise, but not for the people in the vehicles: This is an effort to train AI models to enter conflict zones. 

The autonomous military ATVs are operated by Scout AI, a startup founded in 2024 by Coby Adcock and Collin Otis, that calls itself a “frontier lab for defense.” The company said on Wednesday that it has raised a $100 million Series A round, led by Align Ventures and Draper Associates, following its $15 million seed round in January 2025.

Scout invited TechCrunch for an exclusive tour of its training operations at a military base it asked us not to name.

The company is building an AI model it calls “Fury” to operate and command military assets, first for logistical support but soon for autonomous weapons. CTO Collin Otis compares this work, which builds on existing LLMs, to training soldiers. 

“They start when they’re 18 years old, and sometimes they even start after college, so you want to start with that base level of intelligence,” Otis told TechCrunch. “It’s useful to start with someone who’s already made an investment and then say, hey, what do I have to do to teach this thing to be an incredible military AGI, versus just being a broadly intelligent AGI?”

Scout has secured military technology development contracts totaling $11 million from organizations like DARPA, the Army Applications Laboratory, and other Department of Defense customers. It is one of 20 autonomy companies whose technology is being used by US Army’s 1st Cavalry Division during its regular training cycle at Ft. Hood in Texas, with the expectation that the unit will bring along products that prove themselves when it next deploys in 2027. 

For Scout’s internal testing, the rubber meets the dirt at in the base’s hilly terrain. There, the company’s operations team, led by former soldiers, is putting the vehicles through their paces on simulated missions.

While autonomous cars are starting to be seen in more cities around the world, they are operating there in more structured environments with rules. Operating autonomously on unmarked trails or off-road is another challenge entirely. Otis, a former executive at autonomous trucking company Kodiak, said he was motivated to start Scout when he realized the system he helped build there wasn’t intelligent enough to operate in an unpredictable war zone. 

A new approach to autonomy

Scout is turning to a newer autonomy technology: Vision Language Action models, or VLAs, that are based on LLMs and used to control robots. First released by Google DeepMind in 2023, the technology seeded robotics start-ups like Physical Intelligence and Figure.AI, the humanoid robot company led by Adock’s brother, Brett. 

Adcock is on Figure’s board. He says that experience convinced him of the opportunity to bring broader intelligence to the military’s growing fleet of autonomous vehicles. His brother introduced him to Otis, who was advising Figure, and they set about applying the latest in AI to military solutions.

“If I handed you the controller of a drone right now and I strapped a headset on you, you could learn to fly that thing in minutes,” Otis said. “You’re actually just learning how to connect your prior knowledge to these couple little joysticks. It’s not a big leap. That’s the way to think about VLAs and why they’re such an unlock.”

Indeed, I got a chance to drive one of Scout’s ATV around the rutty trails, and the terrain was challenging: steep hills, loose sand on turns, disappearing tracks, confusing intersections. I’m not an experienced ATV driver but made a fair go on my first attempt (if I do say so myself). That’s the kind of general intelligence the company wants in its models, which it has been training via these ATVs for just six weeks after using civilian ATVs to start the process. 

I also rode in the ATV under autonomous control, and could feel the difference — it accelerates faster than a human who might be thinking about a passenger’s comfort. The operations team points out how the vehicles hug the right on wider trails but stay in the middle of narrow ones, like their training drivers. They also, when confused, suddenly slow down to think over their next move, something that happens a few times as it carries us on a 6.5 km loop before returning to base. 

Though the VLAs are new enough that they have yet to be deployed by any company in an operational setting, “the technology is good enough to be doing that experimentation in the field with soldiers to figure out how to most be effective to US forces,” Stuart Young, a former DARPA program manager who worked on ground vehicle autonomy said. And like other autonomy companies, Scout’s full autonomy stack also includes deterministic systems and other flavors of AI to round out its agents’ capabilities.

Young left DARPA this month to join Field after managing a program called RACER. It asked companies to create high-speed, autonomous off-road vehicles, helping seed this space the same way that the organization’s Grand Challenge boosted self-driving cars. Two competitors in this space, Field AI and Overland AI, were spun out of that program, and Scout also participated in as a later addition.

The first applications of ground autonomy, according to Scout executives and military technologists, will be automated resupply: Carrying water or ammunition to distant observation posts, or in a convoy where a crewed truck might be followed by six to ten autonomous vehicles, saving precious human labor for more important tasks. Brian Mathwich, an active duty infantry officer doing a stint as a military fellow at Scout, recalled a recent exercise in Alaska where he led a resupply convoy in total darkness and wished for autonomous vehicles to help him out.

Adding intelligence to the Army’s motorpool

Scout sees itself primarily as a software company, building an intelligence layer for military machines. It doesn’t intend to make the autonomous vehicles themselves but to build atop them.

Adcock expects the startup’s first product to be widely adopted will be one called “Ox,” the company’s command and control software, bundled on hardened computer hardware (GPUs, communications, cameras). It’s intended to allow individual soldiers to orchestrate multiple drones and autonomous ground vehicles with prompt-like commands: “Go to this waypoint and watch for enemy forces.”

However, making that software work requires training on real vehicles. Hence Foundry, which is what the company calls its training range at the military base. There, drivers spend eight hour shifts putting the ATVs through their paces, then work through a reinforcement learning system to log where they had to take over, which is then used to improve the model. The base commander has asked the company’s ATV to take a turn with security patrols.

One hypothesis Scout is testing is that VLAs will enable this relatively limited data set, alongside training data in simulations, to deliver a fully capable driving agent. While the the vehicle seems comfortable on trails, for example, it isn’t ready to operate fully off-road.

Scout is also practicing with drones for reconnaissance and as weapons, giving them intelligence with vision language models, a multi-modal LLM variant.

Scout is working on a system that would see groups of munition drones fly with a larger “quarterback” platform that provides more compute resources to command them. In one mission, the drones would search a geographic area for hidden enemy tanks and attack them, possibly without human intervention. Otis contends that the alternative approach in this scenario might be indirect artillery fire, which is imprecise compared to drone strikes.

While autonomous weapons are a flash point in the politics of defense tech, experts note the concept is old: Heat-seeking missiles and mines have been in use for many decades. The question for technologists is how the weapons are controlled, Jay Adams, a retired U.S. Army Captain who leads Scout’s operations team, told TechCrunch.

He notes the company’s munitions drones can be programmed to only attack threats in a specific geographic area, or only with human confirmation. He also says autonomous weapons platforms are unlikely to fire because they are scared, the way an eighteen year-old soldier might. 

VLAs, too, offer promise for better targeting. Scout says its models are pretrained on a specific set of military data to prepare them for, say, running into an enemy tank while on a resupply mission. Lt. Col Nick Rinaldi, who supervises Scout’s work for the Army Applications Laboratory, says that while automated targeting is hard and unlikely to be used outside of constrained environments in the near term, the potential of VLAs to reason about threats make them a promising technology to investigate.

Adams says the promise of drones that can identify their own targets is key to future warfare: While Russia’s invasion of Ukraine has generated intense interest in drone warfare, he believes having humans operating individual UAVs doesn’t scale enough for the US to face a large number of low-cost unmanned systems should they threaten US forces. 

A mission to counter anti-military vibes

Like many defense startups, Scout wears its mission on its sleeve, and executives will freely criticize companies that are reluctant to hand their technology over to the government. Google, for example, reportedly pulled out of a Pentagon contest to develop control systems for autonomous drone swarms, a capability Scout is also working on.

“The AI people don’t want to work with the military,” Otis told TechCrunch, referencing Anthropic’s spat with the Pentagon over its terms of service. “None of them are open to running agents on one-way attack drones, or running agents on missile systems.”

Nevertheless, Scout is actually using existing LLMs as the base to build its agents, though declined to say which ones. Otis says it has agreements with “very well known hyperscalers” to provide the pretrained intelligence for Scout’s foundation model. Otis also declined to comment on if it uses open-weight models, such as those offered by Chinese companies. Many companies reliant on AI inference build on these models to operate with lower cost compared to models from frontier labs like Anthropic or OpenAI.

Scout expects to address this by building its own model from the ground up in the years ahead, and the founders say much of its capital will go into those training and compute costs. Indeed, Otis wonders if Scout will beat the existing leaders to AGI because its model will be constantly interacting with the real world. 

“There’s an argument in the AGI community along the lines that you can only get so intelligent by reading the internet, and most intelligence comes with interacting in the world,” Otis said.

Does that mean Adcock is competing with his brother’s army of humanoid robots at Figure? No, Otis says, but “we can get to scale much faster because our customer has assets,” he said, referring to the Pentagon.