Skip to content
TerminalPoet
Go back

Notes on working with AI coding agents

Placeholder content. It shows how an opinion piece with a table renders.

What works

  1. Give the agent a way to check its own work: tests, a linter, a build.
  2. Keep project instructions in the repo, not in your head.
  3. Review the diff, not the transcript.

What doesn’t

If you can’t describe what done looks like, neither can the agent.

TaskAgent-friendly?Why
Dependency upgradesYesThe build tells you when it’s done
Visual designPartlySomeone still has to look at the page
Naming thingsRarelyStill the hardest problem

Share this post on:

Previous Post
TIL: pnpm blocks dependency build scripts
Next Post
Markdown style guide