Mission Planning & COA Analysis
Military mission planning requires simultaneous analysis of terrain, weather, threat disposition, force availability, rules of engagement, and risk — across multiple courses of action. AI dramatically compresses planning timelines and surfaces options human planners might miss entirely.
The Military Decision Making Process (MDMP) for a brigade-level operation takes 6+ hours under ideal conditions and can stretch to 24+ hours for complex operations. AI compresses the COA development and analysis phases — which consume the majority of MDMP time — from hours to minutes, enabling decision cycles that outpace the adversary's planning tempo.
Key Context
The Penalty Stakes
- OPSEC classification of plans: Operation plans and orders are classified at levels corresponding to their operational sensitivity — often SECRET or above. No planning data can be processed in commercial cloud environments.
- WARNORD/OPORD classification: Warning orders and operation orders carry classification based on the sensitivity of the operation. AI systems that draft these documents must operate at the classification level of the output.
- Force disposition sensitivity: Order of battle data — unit locations, strengths, equipment — is classified and operationally critical. AI that accesses this data for planning support must be hosted within the appropriate classified network.
- Commander's intent protection: The commander's intent, prioritized objectives, and decision points are among the most sensitive planning elements. Compromise of this information provides adversaries predictive capability over friendly operations.
AI Performance vs. Rule-Based Systems
| Metric | Rule-Based | AI-Driven | Source |
|---|---|---|---|
| Mission analysis | 2–4 hours | 30–60 min | Automated METT-TC analysis, IPB synthesis |
| COA development | 2–4 hours | 10–20 min | AI generates initial COA options from constraints |
| COA analysis / wargaming | 2–6 hours | 30–60 min | Simulation-assisted wargaming with AI red force |
| COA comparison | 1–2 hours | 5–10 min | Automated scoring against commander's criteria |
| Orders production | 2–4 hours | 30–60 min | AI drafts OPORD from approved COA |
| Total MDMP (brigade) | 12–24 hours | 2–4 hours | 5–6x compression; human validation at each gate |
Business Impact
Army Research Laboratory's COA-GPT (February 2024) demonstrated LLM-generated initial courses of action in seconds — compared to 2–4 hours for human planning staff. Accepts mission text and imagery; outputs multiple strategically aligned COA options simultaneously with real-time commander feedback loops.
AUSA analysis (2024) projects AI/ML will speed MDMP 'by an order of magnitude,' enabling replanning during execution — currently impossible at pace under manual MDMP. Brigade-level planning cycle: 12–24 hours → 2–4 hours with AI assistance at each MDMP phase gate. Harvard Belfer Center (2024) further assesses that agentic AI tools could compress Joint Operation Planning Process (JOPP) timelines from days to hours at the operational level.
Infrastructure Requirements
NEXUS Foundry fine-tunes LLMs on military doctrine, planning formats, and operational data — producing a model that understands METT-TC, MDMP structure, and tactical terminology. Commercial LLMs produce generically formatted plans without doctrinal fidelity. NEXUS OS operates within classified networks — SECRET and above — where planning data lives, so no planning data transits commercial infrastructure at any stage. NEXUS OS integrates with order-of-battle databases, weather systems, and terrain analysis tools within the classified environment, and interfaces with modeling and simulation environments (OneSAF, JCATS) to enable AI-assisted wargaming within classified networks. AI-generated planning products are formatted to current doctrinal standards — OPORD five-paragraph format, matrix formats, synchronization matrices — producing review-ready output rather than raw model text. As intelligence updates arrive during planning, NEXUS OS immediately reflects updated threat and terrain assessments in COA analysis, ensuring planning products remain current throughout the planning cycle.
- COA-GPT (v1 & v2) — Army Research Laboratory; published Feb 2024 (NATO ICMCIS); COA options generated in seconds vs. hours; GPT-4 Turbo + StarCraft II evaluation environment.
- DARPA CAML (Competency-Aware ML) — Active R&D program; ML systems that assess own performance and communicate confidence, enabling trust-calibrated autonomy.
- OpenAI DoD Contract — Signed Dec 2024; LLM infrastructure for planning, analysis, and decision support across DoD.
- Scale AI DoD Deal — Multi-million (Mar 2025); data and AI tools for planning intelligence; expands DoD LLM planning capability.
- Army CGSC AI Wargaming Integration — Fort Leavenworth; active since 2026; AI-generated COAs validated vs. human-generated in structured PME scenarios.
- Harvard Belfer Center (2024): Agentic AI tools could compress Joint Operation Planning Process (JOPP) timelines from days to hours at the operational level.