A software engineer reflects on a moment where they let Claude handle an entire bug investigation and fix end-to-end — without ever reading the Sentry issue, the code, or the diff — and only realized it after merging. Claude happened to be correct, but the workflow would have looked identical if it had been wrong. The post argues that AI tools are valuable when they sharpen your thinking, but dangerous when they become a way to skip the cognitive work entirely. The author's new rule: if you can't explain the change, you can't ship it.