Claude Code Was Broken for 6 Weeks. AMD Caught It in 6,852 Sessions Before Anthropic Did.

For six weeks, you thought you were writing your prompts wrong. You could feel Claude Code messing up. Refactors going sideways, files edited without being read, thinking cut mid-sentence. You re-read your CLAUDE.md, tweaked your instructions, blamed your harness. The Anthropic dashboard said everything was fine. You had your feeling against their telemetry. Guess who lost. TLDR: On April 23, 2026, the day GPT-5.5 dropped , Anthropic published a postmortem validating six weeks of user complaints