AI
Claude Code
A terminal agent that behaves like a careful colleague.
- Score
- 5.0
- Status
- Still using
- Reviewed
What I tested
Long-running tasks in a real repository: dependency upgrades, migrating an API surface, and the kind of work that is too big for a single editor suggestion but too boring to do by hand.
What worked
- It plans before it edits, and the plan is usually the part that catches my misunderstanding rather than the diff.
- Working in the terminal means it uses the same tools as CI. What it runs locally is what will run in the pipeline.
- It reads before it writes. The good news is that this is why it is slower, and the slowness is usually correct.
What did not
- It can over-reach on a vague instruction and touch files you did not ask it to touch. Narrow prompts are not optional.
- Long sessions drift. Without a written task list in the repo, the thread of the work gets lost.
Verdict
The agent I trust most with work I do not want to do myself, provided I give it a tight brief and review everything it produces.