Command fails → you alt-tab to ChatGPT → paste error → read fix → type it. 20x/day. ai-fix kills that loop. One word: $ pip install flash-attention # fails $ ai-fix ✓ Applying fix: pip install flash-attention --no-build-isolation Reads shell history, sends to Claude Haiku (~$0.0003/fix), applies fix. GPT-4o-mini fallback. pip install ai-fix