A comprehensive guide to AI-native software development for engineering leaders and practitioners.
Read online | Download PDF/EPUB
By Daniel Meppiel — Software Global Black Belt at Microsoft, creator of APM and the PROSE framework.
Part I: The Foundation — the agentic SDLC thesis
Part II: For Leaders — strategy, ROI, governance, team structures
Part III: For Practitioners — techniques, workflows, tooling, patterns
Closing: What Comes Next
15 chapters. 225+ pages. From "why AI-native development isn't just using Copilot" to the PROSE specification framework, multi-agent orchestration, and enterprise-scale governance.
This handbook practices what it preaches. It was produced using an AI-native editorial pipeline — the same agentic methodology the book teaches. Every claim, framework, and recommendation in these pages was shaped through a process that demonstrates the Agentic SDLC in action.
The agents didn't work in a vacuum. They were grounded in the author's IP and field experience:
- The PROSE framework — a specification methodology for writing AI agent instructions, developed through building APM
- APM architecture and design — lessons from building and maintaining a developer tool under the
microsoftorg (700+ stars, real-world adoption) - Enterprise adoption patterns — drawn from strategic Agentic SDLC conversations with enterprise customers, GitHub Copilot adoption workshops, and AI-native development hackathons
- Reference architecture research — cross-vendor analysis of how AI-native development actually works at scale
The agents amplified a signal that already existed. They didn't manufacture one from noise.
11 specialist AI agent personas, each with a distinct role and editorial mandate. Defined as PROSE agent specifications and orchestrated via the handbook-panel skill:
| Agent | Role | What They Do |
|---|---|---|
| Chief Editor | Narrative Architect | Owns voice, arc, and flow. Cuts bloat. Enforces consistency across all 15 chapters |
| C-Suite Strategist | Executive Writer | Drafts the Leaders block. Frames everything in business outcomes and competitive moats |
| Practitioner Authority | Technical Writer | Drafts the Practitioners block. Ensures every technique is battle-tested, not theoretical |
| CTO Proxy | Skeptical Reader | Reviews leader chapters asking "so what?" and "prove it" — rejects fluff |
| Dev Lead Proxy | Impatient Reader | Reviews practitioner chapters asking "can I use this Monday?" — rejects theory without code |
| Market Analyst | Competitive Intel | Validates vendor claims against market reality. Enforces multi-vendor objectivity |
| Platform Strategist | Architecture Analyst | Evaluates reference architectures. Flags shipped vs. aspirational capabilities |
| Fact & Ref Checker | Claims Auditor | Hunts unverified claims, unsourced statistics, and assertions presented as fact |
| Illustrator | Visual Strategist | Identifies where diagrams accelerate comprehension. Specs Mermaid diagrams |
| Thought Leadership | Voice Reviewer | Ensures the author's voice stays distinctive — not generic "AI best practices" |
| Publishing Advisor | Distribution Strategist | Advises on publishing path, format decisions, and launch sequencing |
Each chapter went through a multi-stage pipeline where agent personas were instantiated as parallel fleets — the same persona running independently on different chapters simultaneously, then results synthesized:
┌─────────────────────────────────────────────────────────────┐
│ Stage 1: DRAFTING │
│ │
│ C-Suite Strategist ──→ Ch 2, 3, 4, 5, 6, 7 (parallel) │
│ Practitioner Auth. ──→ Ch 8, 9, 10, 11, 12, 13, 14 │
│ │
│ Each chapter injected with: │
│ • PROSE framework spec • APM design notes │
│ • Enterprise field notes • Workshop observations │
├─────────────────────────────────────────────────────────────┤
│ Stage 2: ADVERSARIAL REVIEW │
│ │
│ CTO Proxy fleet ──→ reviews all Leader chapters │
│ Dev Lead Proxy fleet ──→ reviews all Practitioner chapters │
│ │
│ "So what?" / "Prove it" / "Can I use this Monday?" │
├─────────────────────────────────────────────────────────────┤
│ Stage 3: SPECIALIST AUDIT (parallel) │
│ │
│ Market Analyst ──→ vendor balance + claim validity │
│ Fact Checker ──→ unverified claims + stat validation │
│ Illustrator ──→ visual opportunities + diagram specs │
├─────────────────────────────────────────────────────────────┤
│ Stage 4: INTEGRATION REVIEW │
│ │
│ Chief Editor ──→ full-manuscript arc, voice, bloat check │
│ Thought Leadership ──→ author voice + positioning │
│ Both proxies ──→ final "would I read this?" check │
├─────────────────────────────────────────────────────────────┤
│ Stage 5: AUTHOR CHECKPOINT │
│ │
│ Author reviews all agent outputs, accepts/rejects/rewrites │
│ Agent recommendations ≠ final content │
│ Every word published is author-approved │
└─────────────────────────────────────────────────────────────┘
The review artifacts are in the repo — raw, unedited:
handbook/reviews/ch*— chapter-by-chapter editorial reviews from CTO Proxy, Dev Lead Proxy, and Chief Editorhandbook/reviews/fact-check-*— structured claim audits (Verified / Qualified / Unverified)handbook/reviews/visual-audit-*— diagram specifications and visual density analysishandbook/reviews/integration-*— full-manuscript consistency reviews
Because a book about AI-native development that hides its AI-native process would be dishonest.
The agents are the infrastructure. The methodology, opinions, and 14 years of field experience — from CERN to GitHub to founding an AI startup to Microsoft — are the author's. Every reader can inspect the full pipeline, see exactly how the editorial process worked, and judge the output on its merits.
This handbook is built with Quarto and uses APM to manage the agent team:
apm install # Installs the handbook-agents panel
apm compile # Compiles agent instructionsThe agent team is distributed as the handbook-agents APM package.
The PDF and EPUB are served from the gh-pages branch via GitHub Pages. Kit.com email links point to stable URLs that always resolve to the latest version — no per-release URL updates needed.
# 1. Build PDF/EPUB locally (requires Chromium for Mermaid diagrams)
./scripts/build-downloads.sh
# 2. Publish to gh-pages and tag the version
./scripts/publish.sh # push PDF/EPUB to gh-pages + create git tag
./scripts/publish.sh --dry-run # preview without changing anythingVersion is set in _variables.yml. Build metadata (date, commit hash) is stamped automatically by the pre-render hook.
The content of this book (prose, diagrams, images) is licensed under CC BY-NC-ND 4.0. You are free to share it with attribution.
Build tooling and scripts are MIT licensed.
For commercial use, translations, or adaptations, please reach out.