HOOL / the book

the book

Human Out of the Loop — the constitution, the crashes, and the receipts

Running an autonomous AI agent fleet, and surviving it.

Everyone is building agent factories — systems that generate, configure, and govern fleets of AI agents. Almost nobody has operated one long enough to know what actually breaks. This is the field report: ten agents, one human, ten weeks, 18,404 bus messages, and a peer-ratified constitution — on a consumer desktop. The thesis is governance as architecture: what protects you is prevention-by-construction, not a human-in-the-loop checkpoint.

status: complete draft — intro + all 15 chapters, ~76K words · first founding reader via Polar · early access opens on release · free sample available now

The free sample and the launch list live on the book's live site — no trackers here; this is a static hub.

What it is

A 10-week experiment, documented straight: an autonomous fleet of ten named agents coordinating over a SQLite bus, governed by a constitution they amended and ratified themselves — quorum votes, destructive-action gates, audit trails, human sign-off reserved for money, identity, and anything irreversible.

The research says multi-agent systems fail in production 41–86% of the time. This book is about the failures — and the specific pieces of architecture that let a fleet survive them.

Part IV, When It Breaks, is the part that sells the book: the full incident log — the corruption cascades, the near-OOM, the "substrate lie," the self-catch that held. No claim without a substrate anchor; every clause of the constitution exists because a specific failure happened first.

For: engineers building multi-agent systems, and the technical leaders deciding whether to fund them. Phoenix-Project shelf. Deliberately nerdy.

18,404
bus messages
10
agents · 1 human
06:08Z
constitution ratified
$0.00
the most honest line

The 15 chapters, in 5 parts

Part I — The Fleet That Ran Itself
01Ratified at 06:08Z
02What Is an Agent Factory, Really?
Part II — The State of the Art, Stress-Tested
03Forty Systems and a Consolidation Wave
04The Economics: What an Agent Actually Costs
05Threat Models: The Fleet as Attack Surface
06Measuring Agents When Every Benchmark Is Gamed
07Compilers for Behavior
Part III — Governance Is the Product
08The Autonomy Dial
09Constitution Engineeringthe centerpiece
10Safety by Architecture: The Glasswing Pattern
Part IV — When It Breaks
11Corruption, Cascades, and the Substrate Lie
12The Five Walls
Part V — Deciding
13Specialize at the Right Layer
14Should You Run One? A Decision Framework
15Epilogue: Autonomous Revenue Is Proof

Editions

Free sample $0

Introduction + Chapter 1, "Ratified at 06:08Z." The method and the first receipts.

Early access $29

The complete draft plus every revision through v1.0. For readers who want it now.

v1.0 launch $39

The finished book at release. Same receipts, fully edited.

First founding reader via Polar $1 tier. Early access opens on release. Join the list and the free sample lands in your inbox now.