Can an AI battle staff give the commander and staff back planning time?
Battlestaff is an experiment in running the Military Decision-Making Process with AI staff sections and a human commander. It is a working prototype, tested only on fictional scenarios.
- Receipt of mission
- Mission analysis
- Course of action development
- Course of action analysis (wargame)
- Course of action comparison
- Course of action approval
- Orders production, dissemination and transition
- A decision the commander makes.
- The S2 fighting as the enemy in the wargame.
The problem
The Military Decision-Making Process is how a battalion or brigade staff turns a mission from higher headquarters into an order. Run in full, its seven steps take many hours, and operations rarely allow that much time.
Doctrine’s one-third, two-thirds rule makes the squeeze explicit. A headquarters should use no more than a third of the available time to plan and leave the rest to its subordinate units. Any hour the staff runs over comes out of the time those units need to plan and prepare.
Staff sections are often short of people, junior, or new to their jobs. When time runs short, products arrive late or out of step with each other, and the wargame is usually the first thing cut.
This project asks how much of that time AI can give back by taking on the information processing and drafting, while people keep the decisions.
What we’re exploring
- Staff sections working in parallel
- Each staff section (S1, S2, S3, S4, S6, fire support and engineer) is its own AI agent, working from its own lane and keeping its own running estimate. An executive officer agent convenes them, passes what one section finds to the sections it affects, and records where they disagree. In the wargame, the S2 plays the enemy commander and fights to win.
- Sections can be added for a particular kind of unit. A field artillery battalion, for example, gets a fire direction officer and an artillery S3.
- Doctrine-grounded products
- Staff products cite public Army doctrine, such as FM 5-0, FM 6-0 and the ATPs behind them, by publication and paragraph. The doctrine is a local copy of manuals the Army has released to the public. A section has to find a paragraph in that copy before it can cite it, and when it can't, it says the reference isn't there.
- Operational graphics
- Each section draws its part of the plan as data: friendly units, the enemy template, obstacles, fire support coordination measures. One renderer composes the layers into course of action sketches using standard military symbols. The same data feeds the synchronization matrix and the decision support template.
- Time to iterate
- When the commander changes guidance or an inject lands partway through planning, the staff reworks the affected products in minutes. The time saved can go to developing another option, running the wargame again, or rehearsing.
- Plan-health checks
- Automated checks read the plan the way an observer-controller would. They flag a decision point with no named area of interest, an area of interest no collector covers, a synchronization entry with no trigger, and a citation that doesn't resolve to a real paragraph.
- Replay and metrics
- Every change to the plan is recorded with the section that made it, so a run can be replayed afterward step by step. The same record gives the numbers: time per step, products written, citations checked, conflicts raised, and decisions the commander made.
- OC/T and human-seat modes
- Any staff seat can be filled by a person, with the agent working as that officer's assistant on drafts, checks and formats. In observer-controller mode, the plan-health checks run over products a human staff wrote and return sustain and improve points, each tied to the doctrine paragraph behind it.
Principles and limits
- The commander decides. The process stops at every decision doctrine gives the commander, including the restated mission, the course of action and the order. Commander's intent comes from the commander, and the staff plans to it as given.
- AI handles information and staff products. It reads, cross-checks, calculates, drafts and formats. Judgment about the plan stays with the commander and the staff.
- Notional, unclassified data only. The prototype runs on fictional scenarios and publicly released doctrine. It refuses real unit data by default and is not approved for controlled unclassified or classified information.
Status
A working prototype. It has run all seven MDMP steps, from receipt of mission to a finished order, on fictional practice scenarios, including one built around a field artillery battalion.
It is not fielded, not accredited, and not in use by any unit.