COA-GPT Quietly Rewrites Battle Plans

A soldier in U.S. Army uniform typing on a laptop with green code displayed on the screen

The Army is moving generative AI from the lab into daily command work, but the rush raises familiar questions about control, testing, and trust.

Quick Take

U.S. Army Pacific commanders now begin updates with a daily “data minute” focused on generative and agentic AI use cases.
The Army says the goal is faster multi-domain planning, while human judgment stays in charge.
Military research and doctrine still warn that AI can mislead users, so oversight remains critical.
New Pentagon and Army guidance pushes faster AI adoption, but only with testing, security checks, and human review.

Army Pushes AI Into Daily Command Rhythm

U.S. Army Pacific Command is using daily “data minutes” to push commanders toward AI tools that support multi-domain missions.^[1] General Ronald Clark says every commander’s update starts with a data minute, where leaders spend time on generative or agentic AI use cases.^[1] That is a clear sign the Army wants AI woven into routine planning, not treated as a side project.

The Army’s own framing is careful, but the speed of change is still the real story. Clark says generative AI is already embedded in the creation and streamlining of the Multi-Domain Pacific Command.^[1] The same reporting says the goal is to speed up AI adoption while keeping human judgment primary.^[1] For readers worried about runaway tech, that balance matters. If the Army gets it wrong, commanders could lean on software before it earns real trust.

COA-GPT Shows What the Army Wants AI To Do

The strongest sign of where this is headed is the Army’s experimental COA-GPT system, now under study at the United States Army Combat Capabilities Development Command Army Research Laboratory.^[2] The Marine Corps University journal says the system can analyze mission variables and propose a Course of Action that aligns with higher-level intent and adapts to human feedback.^[2] The same study presents AI as a tool, not a commander, which is the right safeguard.

That caution is not academic. The same journal says military doctrine should treat AI as an enabling tool and keep human judgment indispensable.^[2] The Army has also pushed a generative AI platform and claims it can speed coding and acquisition tasks.^[6] A Fort Lee Army post on the CalibrateAI pilot says generative AI can speed mission-critical work because it produces answers quickly and contains vast amounts of information.^[5] Those claims are promising, but they remain pilot-level proof, not battlefield proof.

Guidance Demands Speed, But Also Hard Limits

New Department of War strategy language shows the Pentagon wants faster AI rollout across the force.^[3] The strategy directs the Chief Data and Artificial Intelligence Office to work with vendors so the latest models can be deployed within 30 days of public release.^[3] It also calls for pace-setting projects and regular reporting from service and combatant commanders.^[3] That is a big push for speed, but it also makes the case for strict guardrails.

Those guardrails already appear in Army and Navy guidance. The Navy’s generative AI guidance warns that these tools can produce hallucinations and biased results, and it says commercial models should not be used operationally until security controls are fully investigated.^[7] Army Chief Information Officer guidance also requires authority-to-operate review for generative AI systems.^[10] In plain terms, the government is telling commanders to move fast, but only after proving the tools are safe enough for serious use.

Why Skeptics Still Have a Strong Case

Critics have a solid point: faster decision support can also create faster mistakes. The Marine Corps University article says AI must not be described as a decision maker, and commanders must validate, question, or reject AI output, especially when moral judgment is involved.^[2] The International Committee of the Red Cross says automated aids can reduce alertness and hurt situational awareness.^[11] That warning should matter to any officer who has seen how easy it is to trust a polished screen.

Broader research also supports caution. The Belfer Center says military AI systems need continuous monitoring and auditing after deployment because real-world behavior can be unpredictable.^[15] Georgetown’s Center for Security and Emerging Technology recommends training, certification, incident reporting, and risk-based deployment rules for military artificial intelligence decision support systems.^[16] That is the right conservative instinct: use new tools, but never surrender responsibility to them. Human command must stay human.