Agents propose. Commanders decide.
Agentic course-of-action generation under command: AI agents draft and argue options, deterministic code verifies every claim, a human signs the decision.
The commander needs courses of action. The staff produces them manually over days.
Manual planning is slow, unverifiable, leaves no audit trail. The estimate process becomes the bottleneck when tempo accelerates.
How a run works
Set the mission, clock, thresholds.
One planner agent per stance drafts a complete CoA.
Ten deterministic checks screen each draft.
Survivors scored against UK FACES. Commander's weights applied arithmetically, working shown.
Acceptance: two names, timestamped.
What you get
Ranked, evidence-backed recommendation with confidence, dissent and sensitivity notes. Draft CONOPS, missions-and-tasks extracts, CONPLANs. Complete audit log.
A full three-option estimate in minutes, not days.
Frontier agents inside an assurance harness nobody else in the land domain has built.
Anyone can generate options; verifying every claim at machine speed is the hard part.