Agentic Coding

Engineering with AI coding agents — Claude Code · Gemini CLI · Codex CLI

A cross-tool, principle-first guide to building software with AI coding agents. Every chapter follows the same skeleton — Representation (how each tool models the idea), Operation (how practitioners work with it day to day), and Evolution (where the tools converge and diverge) — so a practice you learn here outlives any single release.

Claude-centered, cross-tool: Claude Code is the home lens here; Gemini CLI & Codex CLI appear as comparative context — where the tools converge and diverge. For Claude-Code-specific daily practice (sessions, CLAUDE.md, workflows, config), see the companion Handbook (in draft).

Status: v0.2.0 — content-complete (17 chapters + 6 appendices); v1.0 pending the planned version-branch infrastructure.

New here? Start with How this book is designed — the reading guide to chapter structure, freshness stamps, and the convergence/divergence model.

Chapters

Part 1 · Foundations

How an agent represents the work — its mental model, context as a budgeted resource, and prompting as specification.

Part 2 · The working loop

The day-to-day cycle — running sessions, the edit–test–commit loop, and thinking through problems with the agent.

Part 3 · Extending & delegating

Scaling a single agent — briefing documents, custom tools and skills, and parallel sub-agents.

Part 4 · Projects, teams & production

Working at real scale — starting and refactoring projects, recovering from anti-patterns, automation pipelines, team governance, and enterprise deployment.

Part 5 · Evolution & audit

Staying current — how agentic practices evolve across tools, and how to audit your own practice.

Appendices

Tracking the field

The convergence dashboard shows, per pattern, when each tool (Claude Code · Gemini CLI · Codex CLI) adopted it — so you can see which practices have settled and which are still diverging.