Today’s edition of my newsletter just went out.
Rohan Paul Twitter · Rohan Paul (@rohanpaul_ai) · 2026-06-05
Rohan Paul's daily AI newsletter reports that Anthropic disclosed Claude authored over 80% of its production code in May 2026, with Claude Mythos Preview reaching 16-hour task capability and a 52x speedup on training code, while Anthropic simultaneously called for global mechanisms to slow frontier AI development.
Appears in
Extraction
Topics: autonomous-codingai-code-generationllm-capabilitiesai-safetymodel-releases
Claims
- Anthropic reports Claude generated more than 80% of the code merged into production in May 2026, up from low-single-digit percentages before Claude Code's February 2025 research preview.
- Output per engineer has risen to 8x the 2024 baseline and reliable task length is doubling approximately every 4 months.
- Claude Mythos Preview delivered a 52x speedup on AI training code and can sustain useful work on tasks lasting over 16 hours.
- Claude's success rate on open-ended engineering problems reached 76% in May 2026, a 50-point increase over six months.
- Anthropic warned its models may be approaching recursive self-improvement and called for a global framework to slow frontier AI before verification mechanisms exist.
Key quotes
Claude, generated more than 80% of the code merged into production in May, instead of human coders.
The remaining human edge is research judgment: choosing the right problem, trusting the right result, and knowing when an experiment is dead.
Anthropic also called for a global way to slow frontier AI because its own models may be approaching recursive self-improvement, where a system helps build a stronger version of itself without direct human control.