![[herding-cats-banner.png]] # Herding Cats in the AI Age > *"AI doesn't need more intelligence — it needs doctrine, process discipline, and quality assurance."* > — Jeep Marshall ![[herding-cats-logo.png|100]] --- ## What This Series Is I spent 26 years in the U.S. Army — airborne infantry, special operations, training brigade-level staffs through simulation exercises where planning frameworks got tested under real operational pressure every day. Then I became a Lean Six Sigma Black Belt and started building AI-integrated workflows. What I found: the AI industry keeps rediscovering, through expensive failure, what military doctrine and process engineering already solved decades ago. This series documents that convergence. Seven papers. One case study. Real evidence. No hype. --- ## The Papers ### Paper 1 — The Super Intelligent Five-Year-Old **Why AI Needs Military Doctrine and Lean Six Sigma — Not the Other Way Around** AI operates like a super intelligent five-year-old: extraordinary cognitive horsepower, zero operational maturity. Paper 1 establishes the thesis: MDMP and Lean Six Sigma provide exactly the operational architecture that AI systems need to move from impressive demos to reliable execution at scale. **Key findings:** Google/MIT research proves multi-agent systems degrade performance 39–70% on sequential tasks. UC Berkeley documents 14 failure modes with rates up to 86.7%. Gartner predicts 40% of agentic AI projects cancelled by 2027. [[Paper-1-The-Super-Intelligent-Five-Year-Old|→ Read Paper 1]] --- ### Paper 2 — The Digital Battle Staff **How Two Centuries of Military Doctrine Predicted the AI Agent Coordination Problem** From Napoleon's 1795 headquarters to SOCOM's 2026 agentic AI experiments, military doctrine solved the multi-agent coordination problem before the AI industry existed. Paper 2 maps the convergence: every major AI company independently arrived at the hierarchical orchestrator-worker architecture the military calls command and control. **Key findings:** Pentagon AI Acceleration Strategy, SOCOM agentic AI task forces, Army MOS 49B (AI/ML officer), Blue system automating MDMP, defense AI industrial base analysis (Palantir, Shield AI, Sarcos, Exia Labs, Scale AI). [[Paper-2-The-Digital-Battle-Staff|→ Read Paper 2]] --- ### Paper 3 — The PARA Experiment **How an Obsidian Vault Became a Multi-Agent Coordination Laboratory** One practitioner. One knowledge vault. 33 days. 1,768 git commits. A personal note-taking system became an accidental laboratory for multi-agent AI coordination — and 12 of 14 failure modes from UC Berkeley's MAST taxonomy appeared on schedule. **Key findings:** NATO alphabet exhausted in 5 days (60 sessions by day 7), 3.83x commit velocity increase after multi-agent introduction, zero gate violations after compliance framework deployed. [[Paper-3-The-PARA-Experiment|→ Read Paper 3]] --- ### Paper 4 — The Creative Middleman **Adobe's AI Identity Crisis and What It Means for the $257 Stock** Adobe Firefly routes your prompts to Google's Gemini, Black Forest Labs' FLUX, and Ideogram's text engine because Adobe's own native model can't render readable text. When the field test result is "BEARETIXSLUGE" vs. "HERDING CATS IN AI AGE," the middleman problem becomes impossible to ignore. **Key findings:** Head-to-head field test evidence, DOWNTIME waste analysis, SWOT from five expert perspectives, 43% stock decline analysis, and the structural tension between Adobe's growth narrative and its operational reality. [[Paper-4-The-Creative-Middleman|→ Read Paper 4]] --- ### Paper 5 — When the Cats Talk to Each Other **AI-to-AI Diplomacy and the Cross-Model Deliberation Protocol** What happens when two frontier AI systems — one trained for Constitutional AI safety, one trained for maximum truth-seeking — engage in structured dialogue? Paper 5 documents a live Claude-Grok exchange that produced the Cross-Model Deliberation Protocol (CMDP) and ran a live physics pilot demonstrating 15–20% fidelity improvement over solo model output. **Key findings:** 8-component CMDP framework, live physics pilot on dark energy, room-temperature superconductivity, and fusion net-energy timelines. Plus: Grok's independent Adobe research, field test synthesis, and what it means that two AI models with opposing philosophies converged on identical conclusions about Adobe. [[Paper-5-When-the-Cats-Talk-to-Each-Other|→ Read Paper 5]] --- ### Paper 6 — When the Cats Form a Team **Multi-Model Ensemble Decision-Making via MDMP Staff Roles** What happens when you assign four frontier AI models — Claude, Gemini, ChatGPT, and Grok — to a structured military planning staff? Paper 6 documents a live ensemble deliberation where each model filled a distinct MDMP staff role, producing higher-quality decisions than any single model alone. **Key findings:** 8-figure evidence package, convergent decisions across all four models on publication strategy, ensemble deliberation reduces single-model failure modes, structured dissent (Devil's Advocate role) catches blind spots that consensus misses. [[Paper-6-When-the-Cats-Form-a-Team|→ Read Paper 6]] --- ### Paper 6b — When the Cats Take the Same Test **Cross-Provider Experimental Design Under Identical Commander's Intent** Six AI systems received the same 3,200-word mission brief. Scores ranged from 49% to 100%. The collection process broke before the analysis started — proving the paper's own thesis about what happens without output discipline. **Key findings:** 51% quality variance from identical instructions (n=6), bimodal distribution (top 4 at 91-100%, bottom 2 at 49-56%), independent convergence on experimental architecture across four providers, provenance crisis from missing output doctrine, composite "best of six" protocol ready for formal execution. [[Paper-6b-When-the-Cats-Take-the-Same-Test|→ Read Paper 6b]] --- ### Case Study 1 — When the AI Stopped Moving Its Own Files **Applying Lean Six Sigma to AI Agent Coordination — A Session Close Transformation** A 477-line AI agent specification was replaced with a 1,160-line shell script. The result: mechanical operations became 87% faster and 1.4σ more reliable. This case study proves that AI agents should make decisions, not execute them. **Key findings:** JSON-driven architecture, gate-action coupling principle, process capability improvement from 3.2σ to 4.6σ, separation of judgment from execution as a design principle. [[Case-Study-Session-Close-Automation|→ Read Case Study 1]] --- ## Core Findings Across the Series | Research | Finding | |----------|---------| | Google/MIT (Dec 2025) | Multi-agent systems hurt sequential tasks 39–70%, help parallel tasks 80.9% | | UC Berkeley MAST | 14 failure modes across 3 categories. Failure rates: 41–86.7% | | Gartner (Jun 2025) | 40% of agentic AI projects cancelled by end of 2027 | | Cursor (Production) | Hierarchical planner-worker-judge: 1M+ lines of code per week | | MDMP vs. Flat Teams | Same agents, different doctrine: 2x output vs. 1000 commits/hour | | CMDP Pilot | 15–20% fidelity improvement when two AI systems critique each other | | Multi-Model Ensemble | 4 frontier models in MDMP staff roles converge on decisions no single model reaches | | Session Close Automation | 87% faster, 1.4σ more reliable after separating judgment from execution | --- ## Start Here New to the series? Start with [[Paper-1-The-Super-Intelligent-Five-Year-Old|Paper 1]]. Deep in AI deployment problems? Jump to [[Paper-3-The-PARA-Experiment|Paper 3]] for the live case study. Invested in Adobe (ADBE)? Read [[Paper-4-The-Creative-Middleman|Paper 4]] first. Want to understand AI-to-AI coordination? [[Paper-5-When-the-Cats-Talk-to-Each-Other|Paper 5]] is the frontier. Building multi-model teams? [[Paper-6-When-the-Cats-Form-a-Team|Paper 6]] shows ensemble decision-making in action. Want the process engineering proof? [[Case-Study-Session-Close-Automation|Case Study 1]] has the before/after data. --- ## How to Cite This Work When referencing individual papers, use the following format: > Marshall, J. (2026). *[Paper Title]*. Herding Cats in the AI Age (Paper N). Retrieved from https://publish.obsidian.md/herding-cats/ **Example:** > Marshall, J. (2026). *The Digital Battle Staff: How Two Centuries of Military Doctrine Predicted the AI Agent Coordination Problem*. Herding Cats in the AI Age (Paper 2). Retrieved from https://publish.obsidian.md/herding-cats/Paper-2-The-Digital-Battle-Staff When referencing the series as a whole: > Marshall, J. (2026). *Herding Cats in the AI Age* [Research series]. Retrieved from https://publish.obsidian.md/herding-cats/ --- ## Navigation | | | |---|---| | [[About\|About the Author]] | [[Glossary\|Glossary & Acronyms]] | | [[References\|Full References]] | [[Image-Gallery\|Field Test Gallery]] | --- ## Contact **Email:** EMAIL-REDACTED **GitHub:** [emanblue](https://github.com/emanblue) --- *Published via [Obsidian Publish](https://obsidian.md/publish). Series began February 2026.* *© 2026 Jeep Marshall. All rights reserved.* *"Herding Cats in the AI Age" is an original research series by Jeep Marshall.* ## Related - [[Index - Published]] — parent folder