Anthropic Says Its AI Now Helps Build Itself — 26% of the Work

Anthropic, one of the biggest AI companies in the world and the maker of the Claude chatbot, published a blog post about how it measures the speed of AI development inside its own labs. The headline finding: Claude now takes on roughly 26% of the research and development work involved in making Claude better. Six months ago, that number was close to zero. In other words, the AI is now helping build its own successor.

Why does this matter outside a lab? Because it's an example of a feedback loop. Think of a student who starts grading their own homework, then writing next week's lesson plan, then designing the course. Each improvement makes the next improvement come faster. If that holds, you should expect new AI features to arrive more quickly and cost less, because companies won't need to hire and train as many people for the same output. It also quietly changes what AI researchers actually do all day — less hand-coding, more directing and checking.

The honest catch: this is Anthropic grading its own homework. Nobody outside the company has verified the 26% figure, and "R&D work" is a fuzzy phrase. It could mean writing routine code, running tests, and summarizing results — the tedious parts — rather than the creative leaps that actually produce breakthroughs. A quarter of the chores is not the same as a quarter of the genius. There's also a safety angle: the faster AI improves itself, the harder it gets for humans to keep watch.

What to watch next: whether other labs report similar numbers, and whether the pace keeps compounding. If it does, the practical takeaway for you is simple. New AI tools will get better faster than you're used to, and the gap between people who adapt and people who don't may widen just as quickly.