According to official sources, the feedback received by Anthropic regarding the degraded quality of Claude Code stemmed from three independent changes affecting Claude Code, the Claude Agent SDK, and Claude Cowork. Specific issues included: a decrease in intelligence caused by lowering the default inference strength from high to medium on March 4th (this setting was restored on April 7th); a bug in the caching optimization feature implemented on March 26th that incorrectly cleared old inference records, causing models to exhibit forgetting and repetition (this was fixed on April 10th); and a system prompt added on April 16th to reduce redundancy that compromised programming quality (this was revoked on April 20th). All affected models have now returned to normal, and Anthropic has reset the usage limits for all subscribers.