Josh Giegel built rocket engines at SpaceX and scaled Virgin Hyperloop from a garage startup to a 300-person company. Now he runs Gambit, a company racing to solve what may be autonomy’s most stubborn unsolved problem: getting different kinds of autonomous machines to actually work together on the battlefield and beyond. While defense has driven early adoption, Giegel built Gambit’s architecture to be domain and market agnostic from day one, as relevant to a warehouse robot fleet as a swarm over a contested battlefield.
The Problem Ukraine Made Impossible to Ignore
Giegel joined SpaceX in 2009 at age 23 to help build the engines that now land rockets routinely. Then he spent seven years at Virgin Hyperloop, eventually serving as CEO. When he left at the end of 2021, he started looking for the next frontier. He found it in the explosion of robotic and autonomous systems entering the world with no shared language to coordinate their actions.
Specifically, he noted, two forces converged in early 2022 to shape Gambit’s founding thesis. The Ukraine war demonstrated how a smaller force could hold off a much larger one using unmanned systems at scale. Conversations with customers ranging from the Defense Innovation Unit (DIU) to operators with firsthand Ukraine experience revealed the same pattern everywhere: systems worked in isolation under many-to-one control, rather than true one-to-many coordination.
Giegel saw an industry chasing drone light shows and pre-scripted waypoint missions while ignoring the harder question. “There was not a solution that can go out there and allow these systems to work together,” he explained.
That wake-up call pushed him to found Gambit, a term that means to take a risk to gain a strategic advantage. He teamed up with co-founder Ben Richardson, whose background at the DIU paired technical vision with real government procurement experience. Together they put their money on a market that barely existed three years ago. “You don’t need collaborative, adaptive autonomy if you don’t have multiple systems in existence,” Giegel said. Giegel and Richardson bet big that once multiple systems existed in the field, operator would need heterogenous platforms to work as autonomous teams.
Resilient Adaptive Intelligence, Not Just Coordination
According to Geigel, the robot teaming challenge really involves two intertwined problems: adaptive collaboration across different form factors plus resilience against electronic warfare like spoofing and jamming. Giegel pointed to failed drone light shows, where a single glitch can send hundreds of units crashing simultaneously, as brittleness in action.
Gambit approaches reliability the way military units approach missions rather than the way single vehicles are engineered. “If I have to do a mission at 9:00, I have to do a mission at 9:00,” regardless of whether or not an individual platform fails, he explained.
He compared a coordinated fleet to an octopus, where each tentacle keeps working even if another loses power. If one drone loses battery, drops communications or shuts down, the rest adapt to still accomplish the mission instead of leaving the human operator to manage the gap. Or better yet, an autonomous ground vehicle may be able to pick up the slack.
Giegel drew a line between what most companies call “coordinated autonomy” and what Gambit calls “adaptive intelligence.” Coordinated autonomy typically means deconflicting pre-planned routes among systems, an approach that breaks down the moment reality diverges from the plan. He offered a scenario where ground and air units are supposed to strike simultaneously, but the ground vehicles hit an unexpected trench. Under simple scripting, timing falls apart and the element of surprise disappears.
“Adaptation means our software constantly looks at the heartbeat of all the different systems,” Giegel said. Real-time replanning can keep hundreds of unmanned assets synchronized even as conditions shift. He argued that this capacity to react and adapt at speed and scale is what will enable success for advanced militaries and civilization more broadly. Organizations that can learn, field and scale new tactics will be the ones that win.
Inside the Physical AI Stack
Gambit’s architecture rests on what Gambit calls the “Physical AI Stack,” built around three layers. The interface layer, named ALIEN for Adaptive Learning and Interaction Engine, lets operators issue natural language commands, working alongside large language models like ChatGPT to translate human intent into machine action. The autonomy core sits on each vehicle itself rather than in the cloud, sharing a common “world model” so every platform understands the environment collectively even if only one carries a particular sensor like radar. A third layer runs real-time analytics, including object detection and computer vision, directly on the edge device.
Because vision-language-action models struggle with team dynamics, Gambit trains its systems using reinforcement learning inside simulation environments. It essentially teaches robots to play a game until they develop reusable behaviors. Giegel likened the resulting library to the fictional program downloads in The Matrix: “Tank, I need a helicopter, bro.” Where does that helicopter program come from? Who trains it? “That’s the stuff that we’re doing,” he says.
Those behaviors become “skills,” composed into workflows using natural language or a graphical interface, then deployed on hardware ranging from a Raspberry Pi to Nvidia chipsets. Incredibly, customers can build new capabilities without calling a Gambit engineer. “You don’t have to pick up the phone and call a Gambit employee to create a new capability,” Giegel said. The model lets defense units field new tactics almost overnight. He even envisions a marketplace where operators create and sell skills to each other, turning tactical innovation into something closer to a crowdsourced gig economy. This same skill-sharing marketplace concept applies to industrial customers, not just defense units.
Winning Trust Where Failure Costs the Most
Defense adopted Gambit’s technology early because it faces existential consequences for failure and already runs the infrastructure to test unproven systems rapidly. Giegel pointed to Gambit’s work with the Air Force Research Laboratory and exercises like Task Force Raptor, where Gambit demonstrated capabilities that evolved into a “red team as a service” product line used across multiple test ranges. The company is also building counter-UAS defeat capabilities and electronic warfare tools that spoof enemy sensors, both of which demand the same adaptive orchestration in which Gambit specializes.
The numbers reflect the acceleration. The Pentagon’s Defense Autonomous Warfare Group budget jumped from roughly 220 million dollars last year to a proposed 55 billion dollars next year, a scale that could rival the Marine Corps and may eventually warrant its own senior command. Gambit’s own contract count has grown from a handful to more than a dozen active engagements.
While defense adopted the technology first, it is not defense-exclusive. Commercial sectors are catching up fast, and the underlying coordination problem looks identical whether the “mission” is a strike package or a pipeline inspection. In the energy sector, for example, operators are not just flying drones but running crawlers and ROVs alongside them, often on the same job at the same time. That is the same heterogeneous, multi-domain challenge Gambit built its coordination layer to solve.
A Company Built for the Robots Still to Come
Gambit’s wager, that heterogeneous fleets of drones, ground robots and maritime systems need a shared adaptive brain rather than isolated remote controls, looks increasingly prescient as the Pentagon commits tens of billions toward exactly that vision.
So where is this all going? Geigel once expected air and land integration to dominate. Instead, maritime and air coordination has taken the lead, driven partly by concerns that surface ships and submersibles deploying unmanned systems could become “the battleships of the 30s,” vulnerable to newer platforms the way battleships gave way to aircraft carriers. He also expects fixed-wing aircraft to gain ground over the ubiquitous quadcopter, as operations extend across longer distances and water-heavy theaters. He pointed to programs like the LUCAS (Low-Cost Uncrewed (or Unmanned) Combat Attack System) program as early signals.
He anticipates an accelerating arms race between distributed electronic warfare capabilities and the counter-systems built to defeat them, along with a shift in doctrine as aerial fleets replace ground-based defenses for covering large areas. Giegel noted that tactics suited to a foreign battlefield look very different from protecting a crowded stadium in Los Angeles, where population density rules out gun-based counter-drone systems entirely. On that note, however, these same adaptive behaviors, graceful degradation, edge autonomy, skill libraries, transfer directly to industrial inspection, logistics, and energy infrastructure.
Clearly, there is more work to be done. And so Gambit is hiring across engineering and business development as demand grows. Giegel plans to attend upcoming industry events including Fed Supernova. As he puts it, the real advantage comes when “the laptop closes” and the mission keeps running anyway. For Gambit, that’s one thing you can bet on.