Developer-first · field-tested · no hype
The craft of engineering AI coding agents.
Prompting, agent harness, and Claude Code / Codex skills, tested in production, with the judgment a generic AI answer can't give you.
16 chapters
5 parts
★ Claude Code · Codex
THE CURRICULUM
One running project, end to end.
Part 0 · Get it running
Mindset + your first agent
Part 1 · Make the agent obey (foundations)
1 Prompts as contracts 2 Hand-write a minimal agent loop 3 Context engineering Part 2 · Claude Code / Codex skills, hands-on
4 Your first SKILL.md 5 Skills, hooks & slash-commands 6 Subagents & orchestration (and when not to) 7 The Codex side Part 3 · From running to trustworthy
8 Evals 9 Observability 10 Error handling & reliability 11 Guardrails & sandboxing 12 Cost & latency Part 4 · Ship to production
13 Deployment shapes 14 Versioning, rollback, SLA 15 Capstone: ship & publish WHY THIS, NOT A FREE COURSE
What makes it different.
Tested in production
Error handling, evals, cost, the failure modes that only surface once you ship. The parts the free courses stop short of.
Vendor-neutral
Which tool actually holds up, and when, not whichever one's docs you happened to land on.
Which-to, not how-to
AI can autocomplete the code. It can't tell you what's worth building or whether to trust the result.
Start where it gets real.
Chapter 0 — install an agent and run one real task end to end.
Open Chapter 0 →