The Agentic SDLC Handbook
A Guide to AI-Native Software Development for Leaders and Practitioners
Preface
Version 0.10.0 · April 2026
For Delphine, Gabriel, Laia, and Adrian — everything else is context.
Why This Book Exists
Every engineering organization is adopting AI coding agents. Almost none of them have a methodology for it.
Teams are configuring Copilot with a single instructions file and calling it “AI-native development.” Leaders are measuring success by lines-of-code generated. Individual developers are discovering that AI-assisted code needs more rework, not less – because the underlying approach is wrong.
This book provides what’s missing: a systematic methodology for building software with AI agents, from organizational strategy to the individual keystrokes. It introduces the PROSE framework – five architectural constraints that make AI agent output reliable, verifiable, and maintainable – and shows how to implement it with real tools and real workflows.
The methodology in this book produced the book itself, and it powers APM, an open-source agent package manager.
Who This Book Is For
This book speaks to two audiences:
Engineering leaders (CTO, VP Engineering, Director) who need to understand the strategic implications of AI-native development, make investment decisions, and transform their organizations. Read Part II first.
Practitioners (developers, tech leads, architects) who need concrete techniques, patterns, and workflows for working effectively with AI coding agents. Read Part III first.
Part I provides the foundational thesis that both audiences share. The closing chapter looks ahead.
How to Read This Book
You don’t need to read sequentially. Start with whichever part matches your role:
- Part I – The Foundation: The thesis – why the current approach fails and what replaces it
- Part II – For Leaders: The AI-native landscape, business case, reference architecture, governance, team structures, and transition planning
- Part III – For Practitioners: Mindset, agentic runtime machine, instrumented codebase, PROSE specification, load lifecycle, attention economy, the deterministic/probabilistic seam, multi-agent orchestration, execution meta-process, architectural patterns, anti-patterns, and primitives-as-code
- Part IV – Case Studies: APM, the handbook itself, the publishing pipeline, the growth engine
- Closing: What comes next, plus a cross-harness reference appendix
Part III is where the book gets technical: the patterns and architectural concepts for building non-trivial agent-driven systems. The book teaches them; the companion applies them. Install:
apm install danielmeppiel/genesisGenesis is a free, open-source Agent Skill — written by the same hand as the book — that ports the software architect’s role to agentic systems. Summon it in your agent with /genesis <what you want built> and it designs the architecture: which skills, custom agents, and instructions you need, and how they compose — the way a software architect lays out classes and modules for an Object-Oriented Programming (OO) system. Read the book to learn the patterns, or skip ahead and let the architect wield them on your project. Inspectable in full at github.com/danielmeppiel/genesis. Works across Copilot, Claude Code, Cursor, Codex, and OpenCode.
If you have 45 minutes and need the strategic picture, read Part I, then skim Part II – the thesis plus the business case, architecture, and governance shape of AI-native delivery. If you have an hour and start coding Monday, jump to Part III and read the chapters that match what you’re about to ship: instrumentation, PROSE specification, attention economy, or orchestration. If you want proof the methodology survives contact with real systems, the Part IV case studies stand on their own in any order. The Parts are designed to be read in any order; the chapter sequence inside each Part is the only place sequence matters.
A Note on Methodology
This handbook was produced using the same methodology and tooling it describes. The case studies document real sessions executed by the author – who also designed the PROSE methodology. This creates an inherent advantage: the author knows when to push and when to intervene in ways not fully captured by the written method. Where authorial expertise likely mattered beyond what the methodology prescribes, the case studies flag it. Treat the documented patterns as a starting point, not a ceiling.
Acknowledgments
This book exists because of the people who shaped its ideas and the people who trusted them.
François Bouterouche brainstormed the kernel vision of the technology stack with me — the conversations that became the computing paradigm at the heart of Chapter 4. Francesco Manni, my manager and mentor for four years across GitHub and Microsoft, taught me serving leadership by example; his principles and trust created the space where this work could happen. Don Syme and Peli de Halleux, from GitHub Next, cultivated ingenuity and exploration with an openness that made ideas flow freely — and trusted APM’s foundational concepts enough to carry them into GitHub Agentic Workflows. Sébastien Le Calvez built the Software Global Black Belt team and go-to-market motion that created the feedback loop between real-world practice and the methodology in these pages.
To Mondragon Unibertsitatea — for teaching curiosity, innovation, and self-starting spirit through rigorous engineering and Project-Based Learning. The methodology in this book owes more to those foundations than it might appear.
To the early adopters and contributors of APM who believed in the project and in me — Sergio Sisternes, Sébastien Degodez, and François Descamps. Open source lives or dies on the people who show up first.
License
This book is licensed under CC BY-NC-ND 4.0. You are free to read, download, and share it with attribution — as long as you share it as-is. For commercial use, translations, or adaptations, reach out.
The views and opinions in this book are the author’s own and do not represent those of his employer.
