Developer Drops Model Arbitration as AI Code Gets Good Enough

Developer Sauers says about 85% of his code was once generated via an 'arbiter' system since Gemini 2.5 Pro, but now only 5% goes through arbitration review because models write code well enough—though he still vets security-sensitive cases.

2026-09-05 ~ 2026-09-05 · 3 related posts

• Dev: arbiter system wrote 85% of my code since Gemini 2.5 Pro — now it's down to 5% — Sauers_ · 2026-09-05
• Only 5% of my code goes through arbitration review, dev reveals — thkostolansky · 2026-09-05
• Dev drops AI code arbitration: models are just better at writing code now — Sauers_ · 2026-09-05