Speed wins in modern conflict, and A.I. will help America
War Secretary Pete Hegseth put it plainly in his recent memo on military AI transformation: “speed wins.” In an era of unprecedented technological acceleration and multi-polar risk, the winners must create a truly “AI-first military”. And that requires overcoming challenges unique to delivering AI to military commanders and warfighters.
The science-fiction nature of current and emerging AI capabilities is already our reality. State actors are already using frontier models for complex cyber espionage. Even in the last year, it was widely assumed that AI-driven development of large, complex projects like web browsers was years away. But several teams did this last week. Days ago, the small open-source project OpenClaw enabled millions of AI agents to collaboratively navigate vast amounts of personal data.
The future is now, just as it has been several times over my 30-year career at the leading edge of technological innovation. Building-changing products during the last three technological transformations has reinforced a core truth: when the status quo changes, the winners are those who master and deploy the new technology fastest. Speed wins.
To maintain our edge, the Pentagon must address three novel challenges.
First, as AI frontier model capabilities advance on a monthly and weekly basis, they create an enormous attack surface and face never-before-seen challenges around consistency and validation.
We must contend with the rapid evolution of state-of-the-art models, a challenge Secretary Hegseth directly addresses. We must be both aggressive and deliberate about the infrastructure we build, ensuring we do not lock it inside proprietary walled gardens, nor grant unnecessarily restrictive Authorizations to Operate. As companies deliver commercial model improvements in real time, the military must have access to them and know they are safe to use.
Model training presents a major exposure risk for adversaries to exploit. Enabling more rapid adoption of state-of-the-art AI requires a matching investment in true observability and validation of both models and applications in secure and operational environments. This challenge is fundamentally different from prior data cleanup and requires strong, ongoing leadership and planning.
Second, while many Americans have experienced large language models through chat, their true potential lies in agentic autonomy. “Weaponizing learning speed,” as Secretary Hegseth envisions, requires an entirely new level of intentional knowledge sharing and experimentation.
We can start the almost unimaginable quantity of text and data our military members must read, process, produce, and interact with daily. Existing and emerging AI products are extremely capable of synthesizing and manipulating that text, but less equipped to contextualize it and predict its importance.
Agentic AI enables radically more efficient ways to get the right information to the right leader at the right moment. Only through tight collaboration between the military and developers can we harness these new capabilities, build the best possible products and experiences, and ensure they’re made securely, and broadly, available at outposts across the globe.
Finally, as important as text and factual understanding are, our warfighters operate in the physical world. Leading AI researchers have argued that for AI to reach that next level of integrated intelligence, it needs a so-called “world model,” an internal representation of how the physical world works. An advanced LLM knows today what the word “tank” means in several different contexts, but it doesn’t understand that a 60-ton vehicle will sink in a marsh.
“World models” will be the AI topic of 2026 as multiple research efforts compete. While we wait for these models to mature, we have opportunities to combine cutting-edge AI with military simulation and wargaming. The United States needs to build systems that pair the reasoning capability of today’s AI agents with high-fidelity, planet-scale simulations.
This is the missing link in the current tech stack that enables “AI-Native Warfighting.” By integrating a planning platform with a rigorous physics-based simulation, we will create a feedback loop that allows AI agents to learn about the impact of decisions in the physical world. Those integrated agents can then run thousands of iterations of a plan, able to test fuel consumption against terrain, examine line-of-sight against topography with pinpoint accuracy, and assess proposed movement rates against weather. We can then transform all this simulation and learning into actionable knowledge for leaders, planners, and commanders at every level.
The Pentagon is spot-on that speed wins. After all, the history of technological revolutions shows that what separates a fad from a keystone of society is whether an industry can build new systems and create new opportunities faster and smarter than its competitors. This has been true in consumer technology since the invention of the telephone, and it’s true in defense today.
America’s military needs the right AI. By directly addressing the three unique challenges to becoming AI-first – fast model adoption and validation, true partnered development of future capabilities, and rapidly evolving spatial simulation – we ensure America’s commanders have what they need to be their best: planning, testing, and executing at superhuman speeds, without sacrificing the uniquely human decision-making capabilities that guarantee success.
Cory Ondrejka is Chief Technology Officer at Onebrief.