I had my AI review its own code, and it found 3 bugs it had just written

Short summary

A developer discovered that asking Claude Code to review its own output as a skeptical senior engineer catches bugs the AI missed when writing. The same model writing and reviewing still finds different issues because reviewing shifts attention from intent-matching to correctness-checking. The author triages all findings before applying fixes to avoid thrashing correct code.

• •Same AI model finds bugs in its own code when asked to switch to reviewer role
• •Writing and reviewing engage different attention patterns even for AI
• •Triage findings before applying fixes to avoid false positives

Generated with AI, which can make mistakes.