The vehicle knows enough to hesitate. On a 6.5-kilometer test loop through steep, loose California terrain, one of Scout AI's autonomous ATVs sometimes slows as the trail turns uncertain. It holds right on broad tracks, centers itself when the path narrows, and pauses when the next move needs more thought. The behavior is small, almost familiar. It is also the practical edge of Colby Adcock's large wager: that machines can learn to carry context into places where lane markers, constant connectivity, and tidy instructions disappear.
Adcock is not the person engineering every turn. His co-founder, Collin Otis, spent years in autonomy, including as a founding engineer and director of autonomy and AI at Kodiak Robotics. Adcock arrived from another technical system, one made of capital, boards, incentives, and companies. He spent fourteen years in investment banking and private equity before becoming a first-time operating founder. Scout AI is what happened when that financial apprenticeship met a close view of physical intelligence.
The company, founded in August 2024, is based in Sunnyvale. Its central product is Fury, a family of models designed to interpret natural-language orders, understand what cameras see, and guide unmanned systems. The ambition is not to manufacture every ground vehicle, aircraft, or sensor. Scout wants to supply a reasoning and coordination layer that can sit across hardware already in service. Adcock describes the opening as an intelligence gap, a missing brain between a commander's objective and a mixed fleet of machines.
“There's a very big white space for somebody to be the AGI brain for defense robots.”Colby Adcock, 2025
An operator assembled in finance
The route begins before Wall Street. Adcock attended Culver Military Academy in Indiana, where he held the student rank of Sergeant in Battery B Battalion. He graduated in 2006, then completed an undergraduate degree at the University of Florida in 2010. At his Culver 20-year reunion in 2026, he posted a photograph from junior year and wondered, with a wink, whether a defense technology career had been foreshadowed. Careers rarely look so linear while they are happening.
After Florida, he joined Bank of America Merrill Lynch and worked on mergers and acquisitions in technology. In 2012 he moved to GCP Capital Partners, where he spent eight years and rose to vice president. Then came Thomas H. Lee Partners in Boston. As a principal in its Financial Technology & Services group, he worked with founders on strategy, operations, and transactions. His public biography from that period uses a revealing phrase: “in-the-trenches support.” The trenches were still metaphorical, but the work trained him to look for leverage inside complicated organizations.
Technology mergers and acquisitions at Bank of America Merrill Lynch.
Joins GCP Capital Partners and begins an eight-year private equity run.
Moves to Thomas H. Lee Partners, later serving as a principal.
Joins the board of humanoid robotics company Figure AI.
Co-founds Scout AI with autonomy engineer Collin Otis.
A board seat at Figure AI changed the proximity of the problem. Figure, founded by Adcock's brother Brett, was developing general-purpose humanoid robots. Colby could watch a new class of models move from language and images into action. The brothers were already close. Brett has said they grew up together, attended the same college, spent about fifteen years in New York, and later lived roughly a block apart in California. Their companies also ended up about ten minutes from each other.
The family connection supplied one crucial introduction. Brett connected Colby with Otis, who was advising Figure. Otis understood where traditional autonomy became brittle. Adcock understood what it took to capitalize and organize a company around a narrow opening. Together, they focused on military environments, where the conditions are less forgiving than a mapped city street and the demand for coordination extends across different kinds of machines.
The command is the interface
Scout's product idea can be reduced to a scene. A commander states an objective by voice or text. A large model interprets the goal and drafts a plan. A person reviews it. Once approved, smaller agents translate the plan into instructions each vehicle can understand. Cameras, telemetry, and the agents' own reasoning update a shared view. If conditions change, the system can reassign tasks. Human intent stays at the center while software handles more of the coordination.
This architecture is distributed on purpose. Adcock's thesis is that large models may run on secure clusters farther from the action, mid-sized models may operate deeper in the field, and smaller inference systems may sit on individual platforms. If one connection is lost, the remaining agents should still understand the objective and continue collaborating. Fury is intended to run with low-power commercial hardware and as little as one RGB or thermal camera, without dependence on cloud access, lidar, or radar.
The word “intended” matters. A demonstration is not a deployed capability. The hard problems are reliability, cybersecurity, judgment, integration, and evidence that a system behaves under stress. Scout's own field work makes that gap visible. Human drivers spend full shifts gathering demonstrations on real vehicles. When a safety driver intervenes, the team logs the moment and uses it to improve the model. Failure is not hidden from the curriculum. It becomes the curriculum.
Models learn in the dirt
Scout began with civilian ATVs, then expanded training on military vehicles. Its operations team, including former soldiers, runs simulated missions over hills, loose turns, disappearing tracks, and confusing intersections. Adcock calls the Central Valley proving ground Forge. The name is good because it refuses the clean abstraction of a benchmark. Here, intelligence is shaped through dust, repetition, correction, and the occasional hard stop.
The first practical uses are less cinematic than robotic armies. Automated resupply could allow one crewed truck to lead several autonomous followers carrying water or ammunition. Reconnaissance drones could survey terrain while a ground platform provides more compute and communications. Ox, Scout's command-and-control package, is meant to coordinate such mixed fleets with hardened computers, cameras, and radios. Fury is the reasoning layer; Ox is one way it reaches the motor pool.
The stakes rise when the same tools are connected to weapons. In February 2026, Scout demonstrated seven agents planning and carrying out a mission involving an autonomous ground vehicle and two drones against an unoccupied target truck at a military test site. The demonstration made the company's direction concrete and the unresolved questions impossible to ignore. Scout says its systems can include geographic constraints and human confirmation. Outside experts have warned that convincing demonstrations still need to become reliable, secure, fielded systems. Adcock's bet is being tested in that space between possibility and proof.
“The most important frontier in AI is the physical world, and it should be pursued in service to the men and women who defend this country.”Colby Adcock, 2026
Capital, compressed
Scout emerged from stealth in April 2025 with seven employees, two Department of Defense contract selections, a $15 million seed round, and Fury. A year later it announced a $100 million Series A co-led by Align Ventures and Draper Associates. The company said it had grown to 34 people and booked $11 million in contracts during its first 18 months. The money is intended for model training, compute, multi-agent coordination, and hiring.
The fundraising fits Adcock's old profession, but the founder's task is different from the investor's. An investor can spread risk across companies. A founder concentrates it into one. Adcock has moved from evaluating management teams to building one, from advising on operational improvements to owning the cadence, and from studying technology markets to making a public claim about where one is headed.
His public communication is brisk and recruiting-heavy. He posts field clips, product diagrams, hiring calls, and praise for Figure's progress. The tone often returns to service, urgency, and national competition. At the same time, the company's most persuasive moments are quieter: a robot slowing before a turn, a camera working after the headlights go out, a human takeover becoming tomorrow's training example.
A thesis with consequences
There is a useful founder pattern inside Adcock's story. He did not discard his previous career to become an imitation engineer. He paired his experience in financing, governance, and company construction with Otis's deep autonomy background and an operations team that understands field conditions. The partnership works by specialization. So does the product. Scout does not need to own every robot if its software can become the connective layer between them.
Whether that layer earns broad adoption will depend on unglamorous proof: performance outside staged routes, secure integration with existing systems, clear human control, and behavior that survives uncertainty. The physical world does not accept a confident slide deck as evidence. It offers ruts, darkness, broken links, mismatched hardware, and consequences.
Adcock spent most of his career learning how companies absorb capital and turn it into operating capacity. At Scout, that lesson is compressed into a harder question. Can money, models, demonstrations, and military need become dependable intelligence at the edge? The answer will not arrive in a deal room. It will appear one hesitant turn, approved plan, and completed field mission at a time.