Anthropic: Claude Now Writes Over 80% of Merged Code at Company
Anthropic published a report titled "When AI Builds Itself" on Thursday, claiming that its AI model Claude now authors over 80% of the code merged into the company's codebase. Since early 2025, engineers have been shipping roughly eight times more code per day than in 2024, driven by Claude's ability to run code autonomously rather than just suggesting snippets. The company suggests that AI is already actively building future AI systems by writing code, running experiments, and assisting with research. This trend could potentially lead to recursive self-improvement, where AI systems design their own successors. However, Anthropic cautions that this outcome is not inevitable, as Claude still lacks research judgment in selecting the right problems. The report coincides with Anthropic's increasing focus on AI autonomy ahead of its planned IPO, and similar moves by competitors like OpenAI with GPT-5.5 and Google with Gemini Spark. Anthropic notes that humans will increasingly shift towards oversight and validation roles, while AI accelerates software development and potentially revolutionizes other scientific fields.
Key facts
- Claude writes over 80% of merged code at Anthropic, up from low single digits in 2024.
- Engineers now ship ~8x more code per day compared to 2024 due to Claude's autonomous coding.
- Anthropic warns recursive self-improvement is possible but not inevitable yet.
- Claude lacks research judgment to choose the right problems independently.
- Report aligns with Anthropic's IPO preparations and broader AI industry trends.