Understanding these three failure modes helps engineers diagnose perceived 'model drift' in their own coding agent setups and recognize that reasoning effort settings, prompt caching bugs, and system prompt edits can each independently tank output quality.
Anthropic's postmortem explains that user-reported quality regressions in Claude Code, the Claude Agent SDK, and Claude Cowork stemmed from three separate issues: a March 4 change defaulting reasoning effort from high to medium, a March 26 caching bug that repeatedly cleared prior reasoning every turn instead of once, and an April 16 verbosity-reduction system prompt that hurt coding quality.
All three were reverted by April 20, and Anthropic reset usage limits for all subscribers as a remedy.
