Claude, Anthropic’s AI, now leads 26% of the work involved in building the next version of Claude.
That’s not a typo. This week Anthropic released a first-of-its-kind transparency report showing exactly how much of their own research and development is done by AI systems rather than humans. The number is striking — and it’s moving fast.
Here’s what “leads” actually means in their framework. Anthropic uses a five-level automation scale. At the “leads” level, Claude can complete most of a task end-to-end from a single high-level instruction while a human supervises and steps in if needed. Think of it like a senior developer who drafts the entire solution while a tech lead reviews before shipping. Humans are still in the loop — but the AI is doing the heavy lifting.
More revealing than the 26% headline: over 90% of Anthropic’s measured R&D tasks now have AI working at or above the “collaborates” level. That’s not AI as a fancy autocomplete. That’s AI handling large chunks of real research work, with humans setting direction and making the calls that matter.
Anthropic also disclosed they’re running roughly 30,000 internal agents on their most active engineering platform. Each one executing tasks, spinning up sub-tasks, and handing results back up the chain.
None of those agents are fully autonomous. No measured task at Anthropic has zero human oversight. But the scale tells you exactly where this is heading.
What This Means Beyond Anthropic
What’s happening inside an AI research lab today is a preview of what reaches every other industry within two to three years. The same pattern — AI handling the bulk of execution while humans supervise and direct — is already arriving in software development, content production, customer operations, and data work.
The companies winning this transition aren’t replacing their teams. They’re restructuring how those teams work. Humans set objectives, own decisions, and take accountability. AI handles iteration, volume, and the parts of work that don’t require judgment.
The practical implication is stark. If your competitors adopt this model and you don’t, the productivity gap compounds quickly. Anthropic’s own numbers are evidence it’s already happening inside the most AI-dense organizations on the planet.
One more thing: Anthropic published this data specifically to give governments and the public better visibility into how fast AI development is accelerating. When an AI company starts reporting metrics on AI building AI, the pace of change is no longer a forecast — it’s a measurement.
Want to explore how AI automation could benefit your business? Let’s talk.