A post by Anthropic suggests that artificial intelligence (AI) systems are increasingly taking over their own development cycle, Anthropic is now delegating a growing share of AI development to AI systems themselves. Taken to its logical conclusion, this trend points toward recursive self-improvement - a system capable of fully autonomously designing and developing its own successor.
As of May 2026, more than 80% of the code merged into Anthropic's codebase was authored by Claude, up from low single digits in early 2025. Externally, the length of tasks that AI models can reliably complete on their own has been doubling roughly every four months, and key software engineering benchmarks have gone from single-digit scores to near-saturation in just two years.
From Writing Code to Running Research
Claude is not only writing code that works - it is closing the gap with humans on code quality, which many at Anthropic believe reached rough parity in 2026 and is expected to surpass human quality within the year. Beyond engineering, AI is making inroads into research itself. In April 2026, Anthropic published the first demonstration of Claude running an open-ended research project end to end, where Claude-powered agents were given an open problem in AI safety and left to solve it - proposing hypotheses, testing them, sharing findings with parallel agents, and iterating.
Still, important gaps remain. Large performance differences persist when it comes to Claude exercising judgment in choosing strategies - that is the gap between AI today and a future system capable of autonomously designing its own successor.
Looking ahead, Anthropic outlines three possible scenarios: AI capabilities plateau but today's tools diffuse broadly; AI labs compound efficiency gains with humans still setting direction; or full recursive self-improvement emerges, with AI building its successors at the speed of compute. Anthropic states it does not believe the first, most gradual scenario is likely, and expresses greater concern about the faster-moving futures.
Anthropic says it would support a slowdown or temporary pause in frontier AI development if other leading labs and countries did so simultaneously and verifiably - while acknowledging that building the trust and infrastructure needed for such an agreement would be enormously difficult.