Let another pair of eyes question the work
Self-checking helps; a fresh review asks different questions.
Getting startedAbout 2 min read
The explanation can sound like the intention
After an agent finishes a change, asking it to check the result is useful. It can find mistakes and run relevant tests. But a confident explanation from the same conversation is not an independent review.
The conversation already contains the reasoning that led to the implementation. It may keep following that reasoning instead of questioning whether the chosen solution fits the person or preserves the right information.
A fresh conversation needs evidence too
For an important change, I can ask a reviewer that did not author the solution to inspect the actual changes, the intended behaviour and the relevant tests. Give it permission to question the approach, not just approve the description.
A new chat does not automatically guarantee independence or correctness. If it is handed only the author’s summary, it may inherit the same assumptions. It needs enough source and runtime evidence to reach its own view.
Ask about a consequence
A useful request is specific: what happens if this person edits again before saving finishes, or opens a record that belongs to someone else? Ask for the location of a problem, a reproducible sequence and what remains uncertain.
Review, tests and my own use of the app answer different questions. I want the pieces to inform a decision, rather than turn agreement between agents into proof.