Model-Based Agentic Software Engineering
Front Matter
- •The MAGE Method at a Glance
- •The Book's Language
- •Preface
- •List of Figures and Tables
- •How to read this book
- •Acknowledgments
Part 1 — The Mindset
- 1The Printer
- 2MAGE by Example
- 3Loops and Models
- 4Why MAGE Follows from the Machine
- 5The Engineer's Seat
Part 2 — The Governed Engineering Environment
- 6The Agent Stack
- 7Models and the Semantic Gap
- 8The Governed Environment: Ex-Ante and Ex-Post
- 9Lifecycles and Runbooks
- 10Metrics
- 11When Guardrails Collide
Part 3 — The Model Zoo
- 12The Executable Zoo
- 13The Logical View
- 14The Process View
- 15The Development View
- 16The Physical View
- 17The Scenarios View
- 18The Scope of Modeling
- 19Keeping the Models in Sync with the Code — A Measurement
Part 4 — Putting It to Work
- 20Brownfield
- 21The Skills
- 22Everything Is a Transformation
- 23Training Data and the Novelty Axis
- 24Lessons Learned
- 25Validation with Agents
Part 5 — The Context
Back Matter
- •Toward a Theory of MAGE
- •Implications for Software Engineering
- •Conclusion
- •About the Author
- •Colophon: Dogfooding MAGE
Appendix A — MAGE Engineering Stacks
- AAppendix A — MAGE Engineering Stacks
- AAppendix A - 1. The provenance + fidelity stack
- AAppendix A - 2. The model-coherence stack
- AAppendix A - 3. The specification + verification stack
- AAppendix A - 4. The observe → react loop
- AAppendix A - 5. The resource-mediation stack
- AAppendix A - 6. The governance-of-governance stack
- AAppendix A - 7. The context-management stack
Appendix B — Flagship Mechanisms
- BAppendix B — Flagship Mechanisms
- BAppendix B - 1. Brief-linting
- BAppendix B - 2. Dynamic context injection
- BAppendix B - 3. Staged deploy gates (canary → smoke → promote)
- BAppendix B - 4. Resource-pressure gating (admit before, shed during)
- BAppendix B - 5. Test-serializer (N=1 flock on
dotnet test) - BAppendix B - 6. Agent registry (append-only log + marker cache)
- BAppendix B - 7. Lifecycle hooks (interpose on the agent runtime's events)
- BAppendix B - 8. Orchestrator-as-reactor over an event bus
- BAppendix B - 9. CLAUDE.md rule index (the governance document as a mechanism)
- BAppendix B - 10. Epic Definition-of-Done (Final-Opus trust-nothing re-run)
- BAppendix B - 11. Operational playbooks
- BAppendix B - 12. Self-governance (detect your own recurring issues; convert each into a tasteful control)
- BAppendix B - 13. Composed state-machine model (typed lifecycles + cross-machine invariants)
- BAppendix B - 14. Control↔substrate dependency (computed blast-radius)
- BAppendix B - 15. Coverage → model-node mapping (which invariants are actually tested)
- BAppendix B - 16. Drift & parity gates
- BAppendix B - 17. Executable source-of-truth models
- BAppendix B - 18. Formal invariant verification (temporal form → model checking)
- BAppendix B - 19. Governance graph (mechanism-interaction model)
- BAppendix B - 20. Meta-model consumption discipline (read, don't hardcode)
- BAppendix B - 21. Model-derived test-obligation census (derive what should be tested, lint the gap)
- BAppendix B - 22. Symbol-anchored traceability graph (derived edges)
- BAppendix B - 23. Synchronization model (meta-sync)
- BAppendix B - 24. PdfModel (sole PDF mutation surface)
- BAppendix B - 25. ContentValidator (input ⊆ output fidelity)
- BAppendix B - 26. Blocking semantic lints
- BAppendix B - 27. Fuzz campaigns (+ auto-coverage)
- BAppendix B - 28. Per-mutator attribution stamps
- BAppendix B - 29. Closed remediation-verb sets
Appendix C — Mechanism Catalog
Appendix D — Operator's Reference
- DAppendix D — Operator's Reference
- DAppendix D - 1. The Operator's Dashboard
- DAppendix D - 2. From Drifted Wiki to Trusted Model
Appendix E — How to Write a Skill
Index
© James C. Davis, 2026–present