The Companion Paper

Appendices


This book has a companion academic paper: The Architecture of Intent — A Framework for Designing Delegated Systems. It is an arXiv-format distillation of the same framework, ~15,000 words and 34 pages, written for a different audience and a different reading mode.

This appendix tells you what is in the paper, who it is for, and how to read it alongside the book.


Where to find it

The paper lives in this repository at paper/architecture-of-intent.pdf, with the editable Markdown source at paper/architecture-of-intent.md. A pandoc + xelatex toolchain compiles one to the other; the build instructions are in paper/README.md.

Two companion teaching decks accompany the paper — a PowerPoint at paper/architecture-of-intent.pptx and a self-contained HTML deck at paper/architecture-of-intent.html. Both are 19 slides, generated from a shared content source.


Who the paper is for

The paper has a different reader than this book.

  • The book's reader is a tech lead, staff engineer, or platform-team member who is on the hook for an agent system going to production. They read the book to make their next decision better.
  • The paper's reader is a researcher, a reviewer, a conference attendee, or a senior practitioner evaluating the framework for adoption. They read the paper to decide whether the framework is worth the larger investment in the book.

The paper assumes more academic context (familiarity with SAE J3016, Shavit & Agarwal 2023, MAST, Building Effective Agents, the OWASP LLM Top 10) and less operational detail. The book assumes the inverse.


What the paper covers, and where the book covers it

The paper is structured in seven sections plus two appendices. The mapping below tells you which book chapter elaborates each section, so a paper reader can use the book to dig deeper on any one topic.

Paper sectionWhat it coversBook chapters that elaborate
§1 IntroductionThe judgment gap; the framework's central claim; the three novel contributionsPrologue; Introduction; What is the Architecture of Intent?
§2 Prior work and lineageEight bodies of standing literature the framework operates withinReading List & References
§3 The frameworkThe four load-bearing elements: intent, archetypes, dimensions, failure taxonomy, SDD; introduced with the framework canvas (Figure 1)The framework on one page; A Miniature Pilot
§3.1 Intent as a design surfaceThe intent / implementation / requirements / policy distinctionsIntent vs. Implementation
§3.2 ArchetypesThe five archetypes; the selection tree (Figure 2); composition as a first-class design surfacePick an Archetype; The Archetype Selection Tree; Composing Archetypes; Intent Archetype Catalog; per-archetype pages under frame/archetypes/
§3.3 Four dimensions of calibrationAgency, autonomy, responsibility, reversibility; the orthogonality argument (Figure 3); spec-clause mappingCalibrate Agency, Autonomy, Responsibility, Reversibility; Four Dimensions of Governance
§3.4 The fix-locus failure taxonomyCat 1–7, with Cat 7 (Perceptual) detailedFailure Modes and How to Diagnose Them; Computer-Use Agents
§3.5 Spec-Driven DevelopmentSDD as the executable protocol; the canonical spec templateAll of Part 2 — The Spec; especially The Canonical Spec Template
§4 Application to AI agent systemsAgentic development lifecycle; capability boundaries via MCP; coding agents; computer-use agentsThe Agent; Least Capability; The Model Context Protocol; Coding Agents; Computer-Use Agents; Designing an AI Coding Agent
§4 Composition with DevSquadPhase-by-phase mapping into the DevSquad 8-phase agentic delivery cycleMapping the Framework to the DevSquad 8-Phase Cadence; Co-adoption with DevSquad Copilot
§5 DiscussionApplicability boundary; complementarity with MAST; generalization beyond AI agentsMulti-Agent Governance; Failure Modes §"How this taxonomy relates to the empirical literature"
§6 LimitationsPosition-paper scope; what the framework does not doIntroduction §"Honest scope"; Signs Your Architecture of Intent Is Degrading
§7 ConclusionThe framework's reach; future workIntroduction
Appendix APaper → book mapping (the inverse of this page)This appendix
Appendix BMapping the framework to Microsoft DevSquad CopilotMapping the Framework to the DevSquad 8-Phase Cadence; Co-adoption with DevSquad Copilot

Reading modes

Read the paper first if you are evaluating the framework. The paper is shorter, more compressed, and structured for a reader who needs to decide whether the larger investment in the book is worth their time. It states the framework's commitments and contributions narrowly; it does not give you the working artifacts.

Read the book first if you have decided to adopt the framework. The book gives you the spec templates, the worked pilots, the patterns, and the rituals (Intent Design Session, Discipline-Health Audit) you actually run. The paper is the executive summary of why those artifacts have the shapes they do.

Read both if you are building the framework into an organization. The paper anchors the conversations you'll have with stakeholders evaluating the framework; the book anchors the conversations you'll have with the team building against it. Treat the paper as the citation, the book as the manual.


Honest scope of the paper

The paper claims novelty for three things only — the orthogonality operationalization of agency and autonomy; the fix-locus framing of the failure taxonomy; and the Cat 7 (Perceptual) category. It explicitly does not claim novelty for SDD, archetypes-as-concept, the four dimensions individually, or Cat 1–6.

The same accounting applies to the book. When you cite the framework, cite what is novel as novel and the rest as synthesis. The ratio is documented in the framework changelog under the v1.0.0 entry.


Framework version

This appendix and the rest of the book reflect framework v2.4.0 (2026-05-10). The paper reflects the same framework version. Both move together — the framework's load-bearing commitments are versioned, and a change in either artifact that touches a load-bearing commitment bumps the framework version. See CHANGELOG.md at the repository root for the versioning convention and the release history.