Introduction

Coding agents change software development most when you stop treating them like better autocomplete and start treating them like bounded operators inside an engineering loop. They can inspect a repo, propose plans, edit files, run tools, write tests, and package changes. They can also waste hours if you hand them vague goals, overloaded context, and no verification contract.

This book is about making that loop reliable. The spine is simple: Spec -> Generate -> Review -> Test -> Refine -> Commit. When the loop holds, agents increase leverage. When one stage is weak, the whole workflow degrades: bad specs create thrash, weak review lets plausible bugs through, shallow tests create false confidence, and oversized context makes the agent forget what mattered.

The book is organized around that operating model. Part I: Foundations covers choosing the right agent surface, understanding harnesses, writing persistent instructions, specializing agents with skills, and controlling context and session state. Part II: Workflow Systems turns those foundations into execution modes, specification-driven development, task decomposition, process overlays, extensions and hooks, review, testing, and release discipline. Part III: Advanced Workflows and Runtime Tradeoffs handles harder runtime constraints: agents across the SDLC, multi-agent coordination, permission design, failure diagnosis, cost control, and local or self-hosted setups. Part IV: Teams, Governance, and Human Costs deals with the consequences at team scale: psychology, governance, enterprise constraints, and how to measure impact without lying to yourself.

You do not need deep AI knowledge to use this book. You do need normal engineering judgment: the ability to define the task, inspect the result, and reject work that is fast but wrong. The aim is not maximum autonomy. The aim is dependable leverage.

---