Anthropic says Claude is accelerating the development of AI systems, and internal data shows the path toward recursive self-improvement is moving faster than expected.
The numbers from inside Anthropic tell the story. Over 80% of the code merged into Anthropic's codebase is now written by Claude. Engineers are shipping 8x as much code per quarter as they were in 2024.
In March 2024, Claude could complete software tasks that take humans about four minutes. A year later, an hour and a half. Today, 12-hour tasks. If the trend holds, tasks that take a skilled person days could come into range this year. By 2027, tasks that take weeks.
Public benchmarks are saturating. SWE-bench, which tests real-world software engineering, went from low single digits to saturation in two years. CORE-Bench, which tests whether a model can reproduce published research, went from 20% to saturation in fifteen months.
Anthropic says they are not at full recursive self-improvement yet, and it is not inevitable. But Claude can already match or outperform skilled humans at executing well-specified experiments.
The remaining gap is judgment, choosing which problems to work on and which goals to pursue. That is the distance between where AI is now and an AI that builds its own successor.
