TL;DR: In a report published June 4, Anthropic disclosed that over 80% of code merged into its production codebase in May 2026 was written by Claude, not human engineers. Average engineer output has jumped 8× per quarter compared to the 2021–2025 baseline. Claude's success rate on complex, open-ended coding tasks hit 76% — up 50 percentage points in six months. The company's Mythos Preview model achieved a 52× speedup on AI training code optimization, a task where skilled humans typically reach only 4×. Buried inside the celebration: a stark warning that recursive self-improvement is arriving faster than anticipated, and that the world needs a coordinated pause mechanism before control slips.
From Assistant to Author
The shift is not incremental — it's a phase change. Before Claude Code's research preview in February 2025, AI-authored code at Anthropic sat in the low single digits. By May 2026, the ratio had inverted: Claude authors the overwhelming majority of merged code, while human engineers have moved into architectural oversight and review roles.
"The shape of stuff today is roughly: humans have ideas, and the models are able to implement, test, and evaluate them an order of magnitude faster than before," one Anthropic employee told VentureBeat.
