In brief
Anthropic says Claude now authors more than 80% of the code merged into the company's codebase.
The AI startup says engineers are shipping roughly eight times more code than they did in 2024.
Anthropic argues AI is already helping build future AI systems and could eventually contribute to designing its own successors.
AI has become so effective at writing code and researching that the biggest constraint on developing new AI systems may now be the humans overseeing them, according to a new study by Anthropic.In its report “When AI Builds Itself,” published Thursday, Anthropic argued that Claude is already helping build future AI systems by writing code, running experiments, and assisting with research—a trend the company says could eventually lead to recursive self-improvement, where AI systems help design their own successors.Claude now authors more than 80% of the code merged into its codebase, Anthropic said, and has helped engineers increase code output roughly eightfold since 2024.“Before Claude Code launched in research preview in February 2025, this number was in the low single digits,” Anthropic wrote, adding that the shift also shows up in the amount of output per engineer. “Lines of code merged per engineer per day stayed constant through Anthropic’s first four years (2021-2024), then began to climb upward in 2025 when Claude began to run code rather than just suggesting it for an engineer to copy and paste.”Anthropic said the future could unfold in several ways: AI progress could slow, humans could remain in charge while AI automates much of the work, or AI systems could eventually begin improving their own successors.“Taken far enough, and given enough compute, that trend points to an AI system capable of fully autonomously designing and developing its own successor,” Anthropic wrote. “This is called recursive self-improvement. We are not there yet, and recursive self-improvement is not inevitable. But it could come sooner than most institutions are prepared for.”The company said it's too early to know which outcome is most likely, but argues that AI is already helping build AI, and acknowledged that lines of code are an imperfect measure of productivity.
