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

At a US military base in central California, four-seater vehicles roam the hillsides. This is a training project, but not for the people in the cars: This is an attempt to train AI-type people to enter non-standard environments.

The autonomous military ATVs are powered by Scout AI, a startup founded in 2024 by Coby Adcock and Collin Otis, who call themselves a “defense lab”. The company said Wednesday it raised $100 million in Series A funding, led by Align Ventures and Draper Associates, following its $15 million seed round in January 2025.

Scout called TechCrunch on her special educational trip to a military base she asked us not to name.

The company is developing an AI model it calls “Wrath” to operate and command weapons, first for support weapons but soon for autonomous weapons. CTO Collin Otis compares the project, which builds on the existing LLM, to training the military.

“They start at 18, and sometimes they start in college, so you want to start with the intelligence to start with,” Otis told TechCrunch. “It’s important to start with someone who’s already invested and then say, hey, what do I need to do to train this thing to be an amazing military AGI, versus just a super smart AGI?”

Scout has received $11 million in military technology development contracts from agencies like DARPA, the Army Applications Laboratory, and other Department of Defense customers. It is one of 20 independent companies whose technology is being used by the US Army’s 1st Cavalry Division during regular training at Ft. Hood in Texas, I hope that the unit will bring something that will prove itself when it returns in 2027.

When testing inside the Scout, the rubber meets the dirt at the base of the mountains. There, a group of the company’s employees, led by ex-servicemen, are driving these cars in similar hearts.

Although autonomous vehicles are starting to appear in many cities around the world, they are operating there in a strict legal environment. Working independently on unmarked trails or roads is another challenge. Otis, a former executive at the independent Kodiak company, said he was inspired to start Scout after realizing the system he helped build lacked the intelligence to operate in an unpredictable war zone.

A new way of self-control

Scout is turning to a new autonomous technology: Vision Language Action models, or VLAs, which are derived from LLMs used to control robots. It was first released by Google DeepMind in 2023, a robotics technology as early as Physical Intelligence and Figure.AI, a humanoid robot company led by Adock’s brother, Brett.

Adcock is on the Figure board. He says the experience convinced him of the opportunity to bring more intelligence to the growing fleet of autonomous vehicles. His brother introduced him to Otis, who advises Image, and he began applying the latest in AI to military solutions.

“If I gave you a drone controller right now and tied your head, you could learn to fly in minutes,” Otis said. “You’re just learning how to connect what you already know to these little sticks. It’s not a big leap. That’s how you think about VLAs and why they open up.”

Indeed, I had the opportunity to drive one of the Scout ATVs around the dirt roads, and the terrain was difficult: steep hills, unsettled sand in the turns, rare tracks, confusing intersections. I’m not an experienced ATV driver but I did pretty well on my first try (if I do say so myself). That’s the kind of intelligence the company wants in its models, which has been training through these ATVs for just six weeks after using non-standard ATVs to begin the project.

I also rode in the ATV and controlled it myself, and I can feel the difference – it runs faster than a person who might be thinking about the comfort of the rider. The task force shows how cars hug the right on highways but stay in the middle of narrow lanes, as their drivers are being trained. They, too, when confused, gradually slow down thinking about their movement, something that happens several times as it takes us to a distance of 6.5 km before returning to the base.

Although VLAs are so new that they have not yet been used by any company in operational situations, “the technology is very good for us to test with the military to find out how it can be used by the US military,” said Stuart Young, a former DARPA program manager who worked independently. And like other autonomous companies, Scout’s autonomy also includes deterministic systems and other AI enhancements to maximize the potential of its agents.

Young left DARPA this month to join Field after overseeing a program called RACER. It asked companies to develop high-speed, autonomous, self-driving cars seed this space likewise the organization’s Grand Challenge promoted self-driving cars. Two competitors in this area, Field AI and Overland AI, came out of the program, and Scout participated in the addition later.

The first independent tasks, according to Scout managers and military technicians, will also return: Carrying water or equipment to a remote observation point, or in a group of mobile vehicles that can be followed by six to ten vehicles, saving precious people in the most important tasks. Brian Mathwich, a Scout staff member, recalled a recent incident in Alaska where he led a rescue team in total darkness and needed autonomous vehicles to help him.

Adding intelligence to the Army’s motorpool

Scout sees itself as a software company, building an intelligent system for military systems. It doesn’t want to build autonomous vehicles but build on top of them.

Adcock expects that the first product that will be adopted by the masses will be the so-called “Ox”, the company’s command and control software, built on hard computing devices (GPUs, communication, cameras). Its purpose is to allow each soldier to organize several drones and infantry vehicles with a quick command: “Go to this place to see the enemy troops.”

However, developing these programs requires training on real vehicles. Hence the Foundry, which is what the company calls its training in the military. There, the drivers spend eight hours putting the ATVs on their tracks, then use a reinforcement learning method to enter where they need to grab, which is used to improve the model. The Base Commander has requested the company’s ATV to exchange with security guards.

One idea of ​​Scout and testing is that the VLAs will allow this limited data, combined with the training data in simulations, to provide a more capable pilot. Although the car looks good on the roads, for example, it is not ready for off-road performance.

Scout is also working with drones for reconnaissance and as weapons, giving them intelligence with visual languages, a multi-level LLM model.

Scout is working on a system that will see the armored units in flight with a large “quarterback” platform that provides more control over them. In one mission, drones search for locations with hidden tanks and attack them, possibly without human intervention. Otis argues that another option in this case would be indirect weapons, which are less accurate than drone strikes.

While autonomous weapons are a hot topic in the politics of technological security, experts see the concept as outdated: Heat-seeking missiles and mines have been used for decades. The question for the engineers is how the weapons are controlled, Jay Adams, a retired US Army Captain who leads the Scout operations team, told TechCrunch.

He also says that the company’s drones can be programmed to target only a specific area, or with the confirmation of people. He said that the autonomous weapons would not fire because they were afraid, as an eighteen-year-old soldier would do.

VLAs, too, offer the promise of better targeting. Scout says its models are trained in a specific battle group to prepare them for, say, running into an enemy tank while performing a reloading operation. Lt. Col Nick Rinaldi, who oversees the Scout project at the Army Applications Laboratory, says that while self-targeting is difficult and impossible to use outside of difficult environments in the near future, the ability of VLAs to anticipate threats makes them a reliable reconnaissance capability.

Adams says that the promise of drones that can identify their targets is very important in the future war: Although the Russian invasion of Ukraine has caused a lot of interest in drone warfare, he believes that having people using UAVs is not enough for the US to face a large number of cheap systems that are not used if they threaten the US military.

Anti-military work

Like many defense startups, Scout wears its mission statement, and executives openly criticize companies that don’t want to give their technology to the government. Google, for example, he says After pulling out of the Pentagon’s competition to develop an autonomous drone control system, the capable Scout is back in the works.

“The AI ​​people don’t want to work with the military,” Otis told TechCrunch, meaning Anthropic saliva and the Pentagon for its operational results. “None of them have the ability to chase down agents on one-way attack drones, or agents on missiles.”

However, Scout is using existing LLMs as a basis for its sponsors, although it declined to say which ones. Otis says it has partnered with “well-known hyperscalers” to provide trained intelligence for the Scout base model. Otis also declined to comment on whether it uses open source models, such as those offered by Chinese companies. Many companies that rely on AI concepts build on these models to operate at a lower cost compared to models from frontier labs like Anthropic or OpenAI.

Scout hopes to overcome this by building its brand from the coming years, and the founders say that most of its money will go to the training and accounting. Of course, Otis wonders if Scout will defeat the current leaders in AGI because his model will always be connected to the real world.

“There is an argument in the AGI community that you can become intelligent by reading the Internet, and great intelligence comes from connecting the world,” Otis said.

Does that mean Adcock is competing with his brother’s army of humanoid robots in the Picture? No, Otis says, but “we can scale up very quickly because our customers have assets,” he said, referring to the Pentagon.