# Glossary & Acronyms This glossary covers military doctrine, Lean Six Sigma, AI systems, and specialized terminology used across the *Herding Cats in the AI Age* research series. Terms appear in the context in which I use them — occasionally adapted from their original domain. [[Home|← Series Home]] | [[References|Full References →]] --- ## Military Doctrine Terms **AAR — After Action Review** The Army's structured reflection process using four questions: What did we plan to do? What actually happened? Why did it happen? What do we do differently? Used in the series as the model for AI system learning loops. *See Paper 1, Section 6.* **Auftragstaktik** German military concept meaning "mission-type tactics" — give subordinates the *what* and *why*, release the *how*. Commanders discovered that AI agents perform better under Auftragstaktik than under detailed step-by-step instructions. *See Paper 2, Section 5.* **BCT — Brigade Combat Team** The U.S. Army's primary combined-arms maneuver unit, typically 3,000–5,000 soldiers. Referenced as the scale at which planning doctrine (MDMP) becomes operationally critical. **BCTP — Battle Command Training Program** Army program that trains brigade and division staffs through simulation-driven exercises. Author's primary training vehicle for 7 years. **C2 — Command and Control** The military's architecture for directing forces: hierarchical structure, standardized communication formats, clear authority chains. The AI industry independently converged on this same architecture for multi-agent systems. **Combined Arms** Military doctrine of integrating infantry, armor, artillery, aviation, and other capabilities into coordinated operations — each component amplifying the others' strengths. Analogy for multi-model AI orchestration in Papers 4 and 5. **COA — Course of Action** A possible approach to accomplishing the assigned mission. MDMP requires developing 2–3 distinct COAs, war-gaming each, then selecting based on commander's judgment. AI equivalent: generate multiple solution paths before committing. **Common Operating Picture (COP)** A shared digital display showing the positions and status of all friendly and enemy forces. AI equivalent: shared context state visible to all agents in a multi-agent system. **DOWNTIME** Lean Six Sigma mnemonic for the 8 wastes: **D**efects, **O**verproduction, **W**aiting, **N**on-utilized talent, **T**ransportation, **I**nventory, **M**otion, **E**xtra-processing. Applied to AI agent systems throughout Paper 1 and Paper 4. **FRAGO — Fragmentary Order** A brief operations order that modifies an existing OPORD. Used when the situation changes and partial orders need rapid transmission. AI equivalent: mid-session context update that modifies agent behavior without full re-planning. **G-1, G-2, G-3, G-4** Staff sections: Personnel (G-1), Intelligence (G-2), Operations (G-3), Logistics (G-4). The numbered staff system descends from Napoleon's four headquarters departments, formalized by the Prussian Great General Staff. *See Paper 2, Section 2.* **KOCOA-W** Terrain analysis framework: Key terrain, Observation/fields of fire, Cover and concealment, Obstacles, Avenues of approach, Weather. Applied in METT-TC(IT) for AI systems analysis in Paper 1. **MAST — Multi-Agent System Failure Taxonomy** UC Berkeley taxonomy of 14 failure modes across 3 categories. Research by Cemri et al., NeurIPS 2025 Spotlight. Failure rates 41–86.7% across 7 frameworks. *See Paper 1, Section 8.2; Paper 3 throughout.* **MDMP — Military Decision Making Process** Seven-step iterative planning methodology: Mission Receipt, Mission Analysis, COA Development, COA Analysis (War-Gaming), COA Comparison, COA Approval, Orders Production. Refined over 70 years. Core framework applied to AI planning throughout the series. **METT-TC** Six mission variables for analyzing any operational situation: **M**ission, **E**nemy, **T**errain, **T**roops, **T**ime, **C**ivil considerations. **METT-TC(IT)** Author's adaptation adding Information Technology as a 7th variable — acknowledging the digital battlespace in which AI agents operate. *See Paper 1, Section 2.* **MOS 49B** U.S. Army AI/Machine Learning Officer area of concentration, established December 2025. First VTIP (Voluntary Transfer Incentive Program) window opened January 5, 2026. *See Paper 1, Section 8.7; Paper 2.* **OAKOC** Terrain analysis framework: Observation, Avenues of approach, Key terrain, Obstacles, Cover and concealment. Used in MDMP Mission Analysis. Automated in the Blue system (Exia Labs). *See Paper 2, Section 5.* **OPORD — Operations Order** The five-paragraph operations order: Situation, Mission, Execution, Sustainment, Command and Signal. The Army's standardized format for directing military operations. AI equivalent: the prompt that launches a complex multi-agent workflow. **ROE — Rules of Engagement** Constraints on the use of force defining who can act, against what targets, under what conditions. Applied as the AI Off-Ramp model in Paper 1, Section 9.3: Weapons Hold (read-only), Weapons Tight (rule-bound execution), Weapons Free (full autonomous within boundaries). **SOCOM — Special Operations Command** U.S. military command overseeing special operations forces. Issued Request for Information for agentic AI demonstrations, April 2026. *See Paper 2, Section 4.* **SOF — Special Operations Forces** Elite military units (Army Special Forces, Rangers, Delta Force, Navy SEALs, etc.). Identified in Paper 2 as the natural proving ground for agentic AI due to small teams, high autonomy, and adaptive doctrine. **WARNO — Warning Order** Preliminary notice of a forthcoming order. Issued immediately upon mission receipt to enable parallel preparation. AI equivalent: early task alerts that allow multi-agent systems to begin staging resources before the complete plan exists. --- ## AI Systems Terms **A2A — Agent2Agent Protocol** Google's April 2025 protocol (donated to Linux Foundation) standardizing how AI agents communicate with each other. Handles horizontal agent-to-agent relationships. Complements MCP. *See Paper 1, Section 8.4.* **AAIF — Agentic AI Foundation** Linux Foundation directed fund announced December 2025 housing three founding projects: Anthropic's MCP, Block's goose framework, and OpenAI's AGENTS.md standard. Platinum members: AWS, Anthropic, Block, Bloomberg, Cloudflare, Google, Microsoft, OpenAI. **Agent Card** JSON metadata document in the A2A protocol that describes an AI agent's capabilities, enabling other agents to discover and delegate to it. **Agentic AI** AI systems that pursue goals through sequences of actions, making decisions about which tools to use and steps to take — rather than simply responding to a single prompt. Distinct from chatbots. **Blue** MDMP automation system built by Exia Labs. Deploys specialized AI agents for each MDMP phase. Being tested with the 101st Airborne Division and Washington Army National Guard. *See Paper 2, Section 5.* **CMDP — Cross-Model Deliberation Protocol** Eight-component framework developed through the Claude-Grok AI-to-AI exchange documented in Paper 5. Components: independent generation, blind critique, revealed-identity critique, human synthesis, live fact-check, probability distributions, training prior disclosure, open publication. **Context Window** The maximum amount of text (measured in tokens) that an AI model can process in a single session. A key constraint in multi-agent system design — agents accumulate "context pollution" over long runs. **DMAIC** Lean Six Sigma improvement cycle: **D**efine, **M**easure, **A**nalyze, **I**mprove, **C**ontrol. Applied to AI system improvement throughout Paper 1. Extended as Quality 4.0 in AI-accelerated form. **Gastown / GUPP** Steve Yegge's fourth-generation agent orchestration framework. GUPP = Gastown Universal Propulsion Principle: sessions are ephemeral, workflow state lives externally in Git, mission persists across agent restarts. *See Paper 1, Section 8.3.* **Hallucination** When an AI model generates confident but factually incorrect output. A quality defect in QASA/LSS terms. Addressed in the CMDP through bilateral fact-checking rounds. **LLM — Large Language Model** Foundation AI model trained on large text datasets (GPT-4, Claude, Gemini, Grok, etc.). The individual "agents" in multi-agent systems are typically LLMs with tool access. **MAST Failure Categories** Three categories from UC Berkeley's taxonomy: (1) Specification & System Design — 37% of failures, (2) Inter-Agent Misalignment — 31%, (3) Task Verification & Termination — 31%. 68% of failures occur before execution begins or after it ends. **MAS — Multi-Agent System** A system where multiple AI agents collaborate (or fail to collaborate) on a shared objective. The primary subject of this research series. **MCP — Model Context Protocol** Anthropic's late-2024 protocol (donated to Linux Foundation December 2025) standardizing how AI agents connect to tools, data sources, and external context. 97M+ monthly SDK downloads, 10,000+ active servers by early 2026. *See Paper 1, Section 8.4.* **Orchestrator-Worker** Multi-agent architecture where an orchestrator agent decomposes tasks and delegates to specialized worker agents. The independent convergence of every major AI company (Anthropic, OpenAI, Google, Microsoft, Cursor, Gastown) on this architecture mirrors the military's hierarchical C2 structure. **PARA Method** Productivity organizational system by Tiago Forte: **P**rojects, **A**reas, **R**esources, **A**rchives. The organizational framework underlying the Obsidian vault documented in Paper 3. **PKM — Personal Knowledge Management** Systematic approach to organizing, capturing, and retrieving personal information and knowledge. Obsidian is the PKM tool used in this series. **Quality 4.0** The convergence of AI/ML with Lean Six Sigma and quality management methodologies. Gartner projects 50%+ of LSS organizations will incorporate AI tools by 2026. **QASAS — Quality Assurance Specialist, Ammunition Surveillance** The oldest federal civilian career program (est. 1920). Provides the model for AI quality assurance throughout the series: dedicated, trained specialists who operate independently of the production chain. *See Paper 1, Section 5.* **RAG — Retrieval Augmented Generation** Technique where AI systems retrieve relevant documents before generating responses, improving accuracy. Used in Blue (MDMP automation) and the CGSC wargaming experiment. **Token** The basic unit of text that LLMs process. Roughly 0.75 words. Token budgets determine how much context an agent can consider. Token efficiency is a key metric in multi-agent coordination — hybrid systems burn 5x tokens per successful task vs. single agents. --- ## Organizational Acronyms **ADBE** — Adobe Inc. (NASDAQ ticker). Subject of Paper 4. **APL** — Applied Physics Laboratory (Johns Hopkins). Runs GenWar initiative. **CDAO** — Chief Digital and AI Officer. Pentagon position overseeing AI Acceleration Strategy. **CGSC** — U.S. Army Command and General Staff College, Fort Leavenworth, Kansas. **CSIS** — Center for Strategic and International Studies. Published the Napoleonic staff analysis in Paper 2. **DIU** — Defense Innovation Unit. Pentagon organization working with commercial technology companies on defense applications. **DUI** — Defense Unmanned Intelligence. (See SOCOM context.) **GenAI.mil** — Pentagon's enterprise-wide generative AI platform, Pace-Setting Project 1. Reached 1.1M unique users across 5 of 6 military branches. **SOFWERX** — SOCOM's innovation arm. Coordinates industry experimentation for special operations. --- ## Key Researchers and Sources **Cemri, Mert & Pan, Melissa Z.** — Lead authors, UC Berkeley MAST taxonomy (NeurIPS 2025 Spotlight). *"Why Do Multi-Agent LLM Systems Fail?"* **Jensen, Benjamin & Strohmeyer, Matthew** — CSIS Futures Lab. Authors of *"Agentic Warfare and the Future of Military Operations."* **Jones, Nate B.** — AI strategist, former Head of Product at Amazon Prime Video. AI News & Strategy Daily. Synthesized Cursor and Gastown production data. **Kim, Y. et al.** — Google Research, Google DeepMind, MIT. *"Towards a Science of Scaling Agent Systems."* (arXiv:2512.08296, December 2025.) **Narayen, Shantanu** — Adobe CEO. Announced Google Cloud partnership and Gemini integration at Adobe MAX October 2025. **Pan, Jonathan** — Exia Labs co-founder, Army veteran. Builder of Blue MDMP automation system. **Verma, Anushree** — Gartner Senior Director Analyst. Published 40% cancellation prediction for agentic AI projects. **Yegge, Steve** — Ex-Google, ex-Amazon. Built Gastown (4th-generation agent orchestration). *"Welcome to Gas Town."* --- [[Home|← Series Home]] | [[References|Full References →]] | [[Image-Gallery|Field Test Images →]] ## Related - [[Index - Published]] — parent folder