A U.S. Air Force cyber officer says the Department of Defense should treat AI agents as subordinate commanders rather than simple tools.

He runs a personal project in a Maryland basement where dozens of AI agents operate on consumer‑grade graphics cards and Mac minis, a setup he has overseen since 2023 under a mission‑command framework.

The officer argues that current reliance on closed frontier models centralizes planning and conflicts with the decentralized execution principles of mission command, and that a June 2026 national security memorandum has directed a rewrite of Directive 3000.09 to require autonomous systems to respect the chain of command.

He defines an agentic warfighter as a model combined with scaffolding and commander‑directed intent, noting that recent Model Evaluation and Threat Research data show task completion times extending beyond 16‑hour horizons, indicating rapid capability growth.

He points to open‑weight models such as Alibaba Cloud’s Qwen3.6, which can be run locally on modest hardware, fine‑tuned to remove vendor refusals, and deployed on rented infrastructure to mitigate risk of hardware loss.

The officer proposes creating small pilot units staffed by officers and non‑commissioned officers to evaluate AI‑driven missions, with each unit preserving model versions, mission orders, and human interventions for reporting to Congress.

He warns that without clear command authority, the Pentagon could cede control of critical systems to unmanaged AI, and stresses that competition will be decided by who can effectively command AI rather than who uses the most AI.