How One NATO IT Leader Leans Into AI's Greatest Strength: Faster Decisions Under Imperfect Conditions

Certainty is too slow for modern combat. Today’s military operations prioritize fragmented intelligence processed in real time over fully verified datasets that arrive after the window to act has closed. In contested environments, connectivity is fragile and often deliberately disrupted, leaving commanders to work with incomplete information by default. Defense leaders face a calculated tradeoff between speed and certainty, and speed is winning.

Dominique Luzeaux is a digital transformation specialist currently serving as a Special Advisor to the Supreme Allied Commander Transformation at NATO. His career includes high-level director roles within the French Ministry of Defence, including at the Defense Digital Agency and the Direction générale de l'armement. Luzeaux’s background as the former Chairman of the French chapter of the International Council for System Engineering demonstrates his deep expertise in large-scale systems.

“Digital technologies are not magic. They don’t replace human decision-making, but they allow us to process information faster and provide courses of action so decisions can be taken in time, which is often more important than taking the perfect decision too late," said Luzeaux. He called the principle at the heart of this new reality “time-boundedness,” a calculated balance between the completeness of a dataset and the speed required for a decision to be relevant.

• Good enough is great: Success in this environment requires accepting constraints that would have disqualified systems a decade ago. Defense organizations must now design for incomplete data and unreliable connectivity as baseline assumptions, not edge cases. "New digital technologies allow us to work on data that is not complete or exhaustive," Luzeaux said. "Of course, you have to be careful of hallucinations, but these technologies bring new ways to deal with massive amounts of data when it's available." The shift represents a fundamental break from traditional military intelligence doctrine, which prioritized verification and completeness.

• The connectivity problem: "Even with the most advanced technologies, we still need connectivity," he said. "This matters most for defense applications because connectivity can be severely disrupted or denied depending on the operational context, whether hostile or non-hostile." Adversaries target networks as a primary attack vector, and natural terrain can create dead zones even in non-hostile environments. This forces a design paradox: systems must be sophisticated enough to handle complex data processing, yet resilient enough to function when networks degrade or disappear entirely.

For Luzeaux, the answer is not replacing human decision-makers, but augmenting them. This represents a deliberate design philosophy, particularly important given the global debate over AI's role in military operations. While Luzeaux champions augmentation, he acknowledged exceptions where full automation is the only viable approach, such as the Aegis combat system, where threats must be neutralized in seconds. For most other decisions, the goal is to support commanders and give them back their most valuable resource: time.

• Filtering, not flooding: "The real job of AI here is to reduce information acquisition time, leaving more relevant time for the human to make a decision," Luzeaux said. "These systems help by providing not all the available information, but only some of the information." AI prioritizes and filters data so commanders can focus cognitive resources on analysis and decision-making rather than information gathering.

• The power of the plural: "I always use the plural form when I mention 'courses of action' for an important reason," he noted. "We must design systems that provide the final decision-maker with several possibilities, not just a single answer they have to agree upon." When AI provides a single recommendation, decision-makers face a binary choice: accept or reject. Multiple courses of action allow commanders to evaluate tradeoffs, apply judgment based on context the system cannot see, and maintain ownership of the ultimate decision.

Trust is the primary barrier to adoption. Technology advances faster than humans become comfortable with using it. Luzeaux recalled how military photo-interpreters resisted the first multispectral satellites 30 years ago, overwhelmed by the volume of imagery and reluctant to trust basic algorithms to help process it. The pattern repeats with every generation of technology.

• A sense of control: "It comes down to trust and a sense of control over the system," Luzeaux explained. "When everything is automated, you lose that feeling of control and you have to simply trust. That feeling is especially important to understand when you are in stressful or life-endangering conditions." He compared it to riding in an autonomous vehicle: even if the technology is statistically safer, passengers resist because they cannot intervene. Systems must preserve human control at critical decision points to earn the trust required for adoption.

Looking forward, Luzeaux sees the demand for accountability as a key factor driving the technology's maturation. "The most important issue in technology now is trust, and the ability to guarantee it," Luzeaux said. "Trust is a two-sided thing. I have to trust the machine, but the machine has to do everything possible so that I can trust it." This reciprocal relationship will define how military and enterprise organizations design future data and AI systems. The systems that succeed will not be the fastest or most sophisticated, but the ones that earn trust by preserving human agency while delivering speed.