Agentic Engineering
Building agents that ship — the harness, the evals, the memory, the ops. The whole series as one volume, with The Long Bet bundled inside.
A deployed agent's capability is the product of model quality and harness quality — and the harness is the part you control. This book is the Harness Engineering series as one volume: thirteen essays on the discipline of building agents that actually ship. Evals that measure capability instead of demos. Memory that survives the session. Tools, planning, security, and ops treated as engineering surfaces, not vibes. Every chapter ends with a deployable artifact — a checklist, decision table, or playbook you can put to work Monday.
Bundled inside: The Long Bet — six essays on what survives in agentic AI, each ending in a dated, falsifiable prediction with a review already on the calendar.
Every claim in this book was traced to its source paper before publication. It is a living book: versioned like software, and every future edition is included with your copy.
Not sure yet? Read Chapter 2 as a free sample (PDF) — The Agent Evaluation Crisis, complete with its closing artifact.
Or start free: the 13 checklists, pay what you want ($0 works) — every chapter's closing artifact in one PDF.
What's inside
Nineteen chapters in five parts — 414 pages, 150+ redrawn figures, per-chapter references, and the shared notation defined once in Chapter 1.
- Part I · The Problem
- Part II · The Engine
- Part III · The Build
- Part IV · The Deployment Frontier
- Part V · The Long Bet
Appendices: the notation table, a cross-reference to the thirteen checklists (the companion PDF ships in the purchase), and the changelog.
A living book
The book is versioned like software — this edition is v2026.07. Buyers receive every future version at no cost, delivered through the store's update mechanism. Update cadence is tied to factory runs: expect a few versions a year. Corrections land in the next version and are listed in the changelog. The essays themselves are free on this site and will stay free — the book buys permanence, sequence, and a single artifact you can hand to a teammate with "read this first."
Engineers building or operating LLM agents, and tech leads deciding what to standardize. It is not for readers wanting model-training internals or a no-code tour — if that's you, the purchase will disappoint, and there's a 14-day no-questions refund either way.
The companion volume
Continual Intelligence — the continual-learning research arc in twelve essays, from the benchmark gap to open-ended AI — is the mechanisms-side companion to this book, compiled from the series at continualintelligence.xyz. It ships as its own PDF + EPUB — get it on its own (CA$39) or in the complete bundle above.
How it was made
Compiled by the Series Factory: an AI pipeline does the reading-at-scale, drafting, and typesetting; a human gates every stage; and a mandatory audit runs before anything ships — every number traced to its source, first authors verified, every figure redrawn and checked. The book carries the same "How This Book Was Made" chapter, so the process is part of the product.