Using AI like a senior engineer

Learn — three modes and when each is right: editor autocomplete for boilerplate, chat for explaining unfamiliar code, and a CLI agent in the repo for multi-file work. Most people only use the first two, and the third is where the real delta is.

Do — run one real task end to end with a CLI agent: give it a verifiable goal (make this failing test pass), keep the diff small and single-concern, have it write failing tests before the fix, and read every line before merging. Try one long refactor in a remote or background session while you work on something else.

Check — you can name one thing AI does badly on your codebase and what you do instead. That specific answer is the one that lands. Generic praise or generic suspicion both fail.

Ship — a short write-up of your workflow. Few people have written this honestly, so it is unusually good blog material.

Do not use it to solve practice problems. Use it to attack your solution with edge cases, review it, or run a timed mock. An AI that tries to break your code is worth more than one that writes it.