Claude Self-Writes Over 80% of Its Codebase
Key Questions
What portion of Claude's codebase does it now write itself?
Anthropic reports that Claude now writes over 80% of its own codebase through recursive self-improvement.
How has Claude's performance on hard tasks changed recently?
Claude's success rate on hard tasks increased from 26% to 76% over six months.
What concerns arise from Claude's self-writing capabilities?
The development raises urgent questions around AI alignment, safety, and control as agents become more autonomous.
Anthropic's Claude now writes over 80% of its own codebase, with recursive self-improvement implications. Success rate on hard tasks jumped from 26% to 76% in six months. This challenges assumptions about AI development speed and raises urgent alignment and safety questions. Factory's Droid also demonstrates recursive self-improvement by closing the loop between production behavior and code fixes.
Sources (2)
Updated Jun 21, 2026