Anthropic made an announcement on Thursday that would have sounded absurd a year ago: Claude, the company's own chatbot, is now helping develop the next, more intelligent version of itself. And unlike the usual startup hype, this one came with actual numbers, an explanation of what "helping" means, and a warning about where it all leads. Claude is leading 26% of Anthropic's model research and development — and the company wants the whole industry to report the same metric publicly.
01.What the numbers actually say
Two statistics carry the story. First: Claude leads 26% of Anthropic's model research and development, which the company says means the model can complete most of a given task "end-to-end from a high-level prompt" while remaining under human supervision. Second: about 90% of the company's R&D is done "in collaboration" with Claude — the model handles "large chunks of work under close human direction". The distinction matters. Leading a task is not the same as doing everything, and Anthropic is careful to state the model is not yet working completely autonomously.
The pace, though, is what made me sit up. The portion of work Claude leads was zero in February. Six months later, in August, it hit a quarter of all research and development. That is a steep curve by any standard, and Anthropic itself flagged where it points.
02.30,000 agents inside one AI lab
Anthropic also disclosed that approximately 30,000 AI agents were doing research and engineering work at the company as of August. These aren't chatbots answering questions — they're automated workers running experiments, writing code and completing engineering tasks, with oversight systems tracking how often agent misbehaviour gets caught by monitoring. The company says these oversight measures are crucial, and it has committed to setting up external third-party evaluators embedded within the company to monitor safety efforts.
If you've been following this space, you know the timing is not accidental. An Anthropic researcher resigned last week with a dire public warning about the technology's threats. CEO Dario Amodei has been leading the call for AI development to actually slow down. Publishing these metrics is the company putting numbers behind the debate.
03.Recursive self-improvement — the technical term you should know
The phrase that matters here is recursive self-improvement: a model's ability to autonomously build its successor. Anthropic was blunt in its blog post — "Models accelerating their own development could make it more challenging for humans to understand or control these systems." Their argument for publishing the numbers is that sharing these metrics helps the world understand how close leading labs are to that threshold.
The company also urged other AI developers to report similar metrics regularly, using a public methodology so the numbers can be compared over time — and potentially across labs. Whether OpenAI and Google take up that invitation is now the interesting question.
Honestly, what surprised me most wasn't the 26% figure itself. It's that Anthropic chose to volunteer it. Companies usually publish numbers that flatter them; this one invites scrutiny of exactly the thing safety researchers are most worried about.
04.Why this lands in the middle of the slowdown debate
This announcement comes at a strange moment for the industry. Amodei, Sam Altman, Elon Musk and other tech leaders have voiced support for slowing development, while President Trump and other figures have pushed back, arguing a US slowdown hands the advantage to China. Meanwhile, Anthropic is expected to beat OpenAI to public markets with a blockbuster IPO later this year — a company calling for caution while preparing one of the biggest listings in tech history.
For Indian developers and AI-curious professionals, the practical read is simpler: the tools you use every day are increasingly built by the tools themselves. The gap between "user" and "builder" is shrinking, and the people who benefit most will be those who learn to direct AI agents well rather than just chat with them. Working with agents — not just prompts — is the skill to build this year, and it's a core part of our GPT-6 Astra — Zero to Hero Guide if you want a structured head start.
Key Takeaways
- Claude now leads 26% of Anthropic's R&D, up from zero in February
- 90% of R&D happens in collaboration with Claude, always under human supervision
- About 30,000 AI agents were doing research and engineering work at Anthropic as of August
- Anthropic published the metrics to track progress toward recursive self-improvement — AI autonomously building its successor
- The company called on other labs to report the same numbers publicly
Frequently Asked Questions
What did Anthropic reveal about Claude building Claude?
Anthropic says Claude now leads 26% of its model research and development, up from zero in February.
How much of Anthropic's R&D involves Claude?
About 90% of the company's R&D happens “in collaboration” with Claude, with humans supervising throughout — and around 30,000 AI agents were doing research and engineering work at Anthropic as of August.
Why did Anthropic publish these numbers?
To measure how close the industry is to recursive self-improvement — AI autonomously building its own successor.
What This Means for You
If you're a student or working professional in India, here's my honest take: you don't need to worry about AI building itself this week. You need to notice what it means for your career. When one AI lab runs 30,000 agents doing real engineering work, "knowing how to code" is quietly becoming "knowing how to direct agents that code". The engineers and founders I speak to are already hiring for exactly that shift. The good news is that this skill is learnable — prompting, agent workflows, cost optimization and safety are all covered in the GPT-6 Astra — Zero to Hero Guide (₹149 digital edition, or ₹1,999 with the printed book delivered in India). The AI building itself is the headline; you building yourself alongside it is the actual story.