Recap and Practice
Key takeaways
- A good workspace gives the AI context, a fast feedback loop, and cheap undo — the setup is the multiplier, often more than the model itself.
- A context file (AGENTS.md / CLAUDE.md) is the highest-leverage thing you can write: concrete rules, loud don'ts, and pointers to examples.
- Make version control a reflex — small, read the diff, for risky work — so nothing the AI does is truly irreversible.
- Keep the loop tight and let the close it: edit, run, read the failure, fix. Feed diagnostics (, lint, typecheck) back to the AI.
- Cage every paid with a hard cap and a throttle, keep secrets out of the repo, and use a harness to make safety checks something the AI can't skip.
Try it
Open whatever project you're working in (or scaffold an empty one) and write a first-draft AGENTS.md by hand: what the project is, the tech stack, the exact commands to run dev/test/lint, three real conventions, and at least two loud don'ts (e.g. "never read or print .env," "don't add dependencies without flagging it"). Then ask your AI tool to make a trivial change and watch whether it respects the file. Each time it does something you didn't want, add or sharpen one line. You're tuning the room, not just the code.
of the chapter
Read my project and help me write a concise AGENTS.md context file for it.
First, inspect the repo and tell me:
- the tech stack and versions you detect
- the commands for install / dev / test / lint / build
- the layout and where the entry point is
Then draft an AGENTS.md with these sections: What this is, Tech stack,
Commands, Conventions, and Don'ts. Be specific and concrete (name the
exact libraries and rules, not "follow our conventions"). State the don'ts
as loudly as the do's, and point to one existing file as a style example.
Keep it under ~40 lines — I'll prune from there.