AI Software FactoryThe Guide
Menu
Browse the complete guide

Front matter

The factory in one line

Part I — Understand

Part II — Design

Part III — Build

Part IV — Prove

Part V — Operate

Part VI — Improve

Appendix — Reference

Appendix — Mission Control case studies

Appendix — Research

Book map

The AI Software Factory Guide

On this page9 sections
  1. The factory in one line, stage by stage
  2. Front matter
  3. Part I — Understand
  4. Part II — Design
  5. Part III — Build
  6. Part IV — Prove
  7. Part V — Operate
  8. Part VI — Improve
  9. Appendices (reference)

How to design, build, prove, operate, and improve an engineering system in which humans define intent and accept risk while bounded agents plan, implement, validate, and recover — and independent evidence decides what advances.

Intent → Plan → Define Agent → Execute through Harness → Apply Skills → Evaluate → Improve → Deliver Software

This eight-stage value stream is the primary reader model. The supporting architecture has six areas: Intent, Harness, Capability, Model, Trust, and Learning, surrounded by adoption. Chapter 2 teaches how the two fit together.

Read the guide front to back, or enter at the part that matches your question.

The factory in one line, stage by stage

Click a stage for a concise contract brief, then follow its links to the canonical chapters for technical depth.

  1. Stage 1 · Builder Intent
  2. Stage 2 · Plan
  3. Stage 3 · Define Agent
  4. Stage 4 · Execute through Harness
  5. Stage 5 · Apply Skills
  6. Stage 6 · Evaluate
  7. Stage 7 · Improve
  8. Stage 8 · Deliver Software

Front matter

Part I — Understand

  1. Why software engineering is changing
  2. The factory in one view
  3. First principles: trust, evidence, and authority

Part II — Design

  1. The human–agent operating model
  2. Authoritative records: from company to release
  3. Intent and specification engineering
  4. Governance, policy, and risk-proportional approval
  5. Economics, metrics, and human attention
  6. Tokenomics and factory economics
  7. Multi-repository design and coordinated delivery

Part III — Build

  1. The Agent Factory
  2. Skills as packages
  3. Control plane, orchestrator, and execution plane
  4. Durable execution: tasks, attempts, leases, and recovery
  5. Coding harnesses and agent protocols
  6. Harness engineering
  7. Development environments, sandboxes, and compute
  8. Agent architecture: loop, MCP, tools, context, and memory
  9. Data, knowledge, and semantic engineering
  10. Context engineering
  11. Models and capability selection
  12. Routing and the escalation ladder
  13. Agent and loop engineering
  14. Loop engineering patterns and defaults
  15. The 12-layer production AI agent stack
  16. Autonomous engineering workflows

Part IV — Prove

  1. Quality and evidence architecture
  2. Testing strategy for agentic change
  3. Evaluation engineering
  4. Evals as factory assets
  5. Quality contracts, proof packages, and certificates
  6. CI/CD, progressive delivery, and production verification
  7. Security: identity, secrets, threats, and supply chain

Part V — Operate

  1. The factory as a platform
  2. Observability, telemetry, and forensics
  3. Resilience, incidents, and the control tower
  4. Control surfaces, event contracts, and storage
  5. Enterprise adoption and the infrastructure landscape

Part VI — Improve

  1. Production feedback, automated review, and the agentic merge queue
  2. Governed learning
  3. Meta-loops and the closed-loop factory
  4. Mission Control as a living case study
  5. Mastering the factory
  6. Where this is going

Appendices (reference)

The v1 curriculum chapters are preserved unchanged in archive/guide-v1/.