Anthropic has revealed that Claude, its flagship AI system, is now handling roughly 25% of the work involved in developing the company's next generation of AI models. The disclosure underscores how leading AI laboratories are increasingly turning to their own systems to accelerate research and engineering pipelines, a trend with wide-ranging implications for how frontier models get built.
The figure was shared publicly by Anthropic leadership and marks one of the most concrete data points yet on the scale of AI self-involvement in cutting-edge model development. It also raises pointed questions about the feedback loops this creates, the pace at which capabilities may compound, and what oversight structures are needed as the ratio grows. Anthropic has previously confirmed that AI is now building AI systems, but attaching a specific percentage makes the scope more tangible than earlier, broader statements.
What Claude Is Actually Doing
The work Claude contributes spans multiple areas of the development process. According to available reporting, this includes writing and reviewing code, running analyses, drafting documentation, and assisting researchers with tasks that would otherwise consume significant human hours. Separately, Anthropic has noted that Claude now handles 95% of its internal analytics queries, suggesting that automation of knowledge work inside the company runs deeper than the headline model-development figure alone implies.
Key Facts
- Claude is responsible for approximately 25% of the work building Anthropic's next AI model.
- The contribution spans coding, analysis, documentation, and research assistance.
- Anthropic's internal use of Claude for analytics already sits at 95% of queries.
- The disclosure adds concrete numbers to a broader industry trend of AI-assisted AI development.
- Questions remain about oversight, feedback loops, and how the percentage will shift over time.
The practical effect is a meaningful compression of the research cycle. Tasks that once required dedicated human researchers or engineering hours can now be delegated, reviewed, and iterated on at speeds that are difficult to match manually. Whether that acceleration is purely a productivity story or also a risk factor is a question Anthropic itself has wrestled with publicly. The company has warned policymakers about the risks posed by next-generation AI systems, even as it continues building them at pace.
"Claude is now doing a quarter of the work developing Anthropic's next AI."Anthropic, via Yahoo News
A Broader Shift in the Industry
Anthropic is not alone in leaning on its own models to build future ones. Several frontier labs have discussed using AI agents for code generation, bug detection, and experimental design. What distinguishes Anthropic's disclosure is the specificity. A 25% figure, if it holds or grows, means the company is operating more like a hybrid human-AI organization than a traditional research lab, with implications that stretch well beyond internal productivity metrics.
Anthropic's own economic research has shown how AI is already reshaping white-collar work across industries. The company now appears to be a case study in its own findings. The researchers and engineers who remain central to model development are increasingly working alongside, and sometimes through, Claude rather than independently of it.
For those tracking where this leads, the trajectory matters as much as the current figure. If Claude contributes 25% of the work today, the question is what that number looks like when the next model ships, and the one after that. The compounding effect of each generation of AI helping to build a more capable successor is precisely the dynamic that makes both the opportunity and the caution around this space so significant. Keeping up with the latest Claude AI news will be essential as Anthropic continues to share data on how its internal workflows are evolving.
