Estimated reading time: 4 minutes
Key takeaways:
- Anthropic silently degraded Claude Code for seven weeks before explaining why. Three engineering missteps quietly reduced quality while users assumed the problem was their own.
- Silence doesn’t contain speculation, it amplifies it.
- Postmortems are necessary but not sufficient: developers need real-time signal during an incident, not a detailed explanation several weeks later.
If you’ve been grumbling that Claude Code hasn’t been pulling its weight lately, you aren’t imagining it.
Anthropic has confirmed that a string of engineering missteps led to a noticeable drop in performance in its Claude Code tool.