By Robbin Laird
James (Jim) Rebesco is the co-founder and chief executive of Striveworks, an Austin, Texas-based company now seven years old that works with the U.S. Army, Navy, and defense intelligence community to deploy and sustain artificial intelligence in operational environments. His background is as a PhD in computational neuroscience and applied AI, followed by years building algorithmic trading systems on Wall Street before pivoting to national security. That trajectory shapes everything about how he thinks about AI in the field: it is not enough to have a good model. What matters is keeping it running, keeping it adaptive, and measuring its impact on the lives of soldiers and warfighters
Rebesco’s path to defense AI runs through an unlikely detour. After completing his doctorate in neuroscience, he joined a startup doing electronic market making, essentially replacing the trading pit with servers and algorithms. The company went public. More importantly, it taught him a lesson that now defines Striveworks.
“It’s not so much about the model,” Rebesco explained. “Models work. But what really matters is that in this kind of very fast-paced, constantly changing environment, if you were really good at keeping the models online, keeping them running, keeping them working, keeping them adapting, that was actually the thing that set you apart.”
That insight became the founding thesis of Striveworks: solve the problem of how you keep AI models online and functioning in challenging, contested environments and do it specifically for the national security space.
Among the most prominent examples of Striveworks’ operational record is its work with the Army’s Next Generation Command and Control program, known as NGC2. It is one of several operational deployments the company has sustained across the Army, Navy, and Combatant Commands including EUCOM, INDOPACOM, and CENTCOM. Rebesco is careful to distinguish this from the long history of C2 systems that had to accommodate every legacy predecessor while somehow being better, an impossible task that NGC2 deliberately breaks from.
“They are making it very clear that AI is interstitially involved in many aspects of this command-and-control framework,” he said. “It’s really a critical enabling technology, so that commanders can make informed, accurate decisions faster. Intermeshing AI with all that is such an important development.”
What distinguishes Striveworks from the crowded field of AI vendors making similar claims is the operational status of its work. Rebesco is emphatic on this point: “A lot of people [are] doing pilots, a lot of people doing experiments. What we’re doing is operational, it’s deployed.” The headline performance metric he cites is not incremental: the system being used by the US Army enables human operators to process information 21 times faster than before. That is not an efficiency improvement. It is a step change.
One implication of our discussion is that AI in the Striveworks approach is not primarily about manpower savings. It is about force redesign. When asked to give a concrete example of how AI-enabled command and control changes what the Army can do in the field, Rebesco described the feedback loop he finds most compelling: “As we’re operationally fielding these AI-enabled solutions, [we’re]able to talk to the centers of excellence that are building out doctrine, that are building out force structures, and watching that be this huge two-way street, doctrine says we need to do this, the technology lets us do that, if we change the doctrine to be this, what does the technology need to do to cover those gaps? Seeing that iterative development in the force structure is significant.”
Rebesco does not hide his frustration with the current defense AI market. His diagnosis is sharp and worth quoting at length, because it frames both the challenge his company faces and the standard by which he believes vendors should be judged.
“Right now we’re just in a very noisy environment. Every company is at this point practically obligated to say they’re an AI company, regardless of what they do.”
His prescription is equally direct. Step one: who is actually doing stuff, and can they prove it? Not with a single laudatory officer testimonial, but with real, measurable, repeatable data—not data about what the system itself does, but about how it improves what the service needs to accomplish. “How has this improved survivability? How has this improved the core missions that the Army or the Navy or whomever needs to do?”
Step two, once operational proof exists, is demonstrating adaptability. Rebesco is clear-eyed about the adversary calculus. “China’s not sitting on their hands. They’re very smart, they’re very well resourced, they’re very thoughtful. If we’re going to depend on AI to support critical warfighting tasks, how can we assure that it stays online? How can we assure that it adapts? How can we assure that when they throw countermeasures at that stuff, we’ve got a response, and we can do it faster and cheaper than they can?”
He frames the broader failure mode with a phrase worth preserving: “Gold-plating requirements is bad, but gold-plating an AI model is also bad. You’ve got to treat it as something you never trust, always question its ability to perform, and you’re constantly making it better.”
Throughout the conversation, Rebesco returned to a theme that runs deeper than product capability: the relationship between AI tools and human judgment. His neuroscience background informs a long-term optimism about AI’s potential, the brain, he notes, operates on layers of computation that silicon has not yet begun to replicate. But his near-term view is more cautionary.
“Human agency or your ability as a human being to self-orient, to set tasks, and to try to create and solve problems is so important. With AI, if you’re a passive receptacle of that stuff, it’s not going to help you. But if the way you look at AI is, what can I do with this, what can it help me do, what do I want to do that I could never do before. It gets really exciting, really fast.”
That framing maps directly onto how he thinks about fielding AI in the Army. The technology does not replace the commander’s judgment. It compresses the time between information and decision, reduces the force footprint required to sustain those workflows, and creates the conditions under which human commanders can act faster and survive longer. The human agency is still in the loop but it operates with tools that have never existed before.
Striveworks represents a relatively rare thing in the current defense AI landscape: a company whose claims are grounded in operational deployment, not pilots or experiments. The 21x improvement in information processing is the kind of result that, if it holds up under scrutiny, signals genuine transformation rather than incremental improvement.
The deeper argument Rebesco makes that AI-enabled C2 changes not just efficiency, but force structure, survivability, and electromagnetic signature is the right argument. It is the kill web argument applied to Army ground operations: distributed, low-signature forces that can process targeting and other data to improve legacy command post configuration.
The challenge going forward is ensuring that the institutional Army moves with the technology. Rebesco described the two-way street between doctrine writers and fielded capability with evident enthusiasm. That iterative loop — technology informing doctrine, doctrine pulling technology — is precisely how transformation actually happens. It is slow, it is contested, and it requires continuity of leadership that the current environment does not always provide.
But the proof points are real, the operational deployment is live, and the conceptual framework is sound. For a defense community drowning in AI vendor claims, Striveworks is a company worth watching.