# Anthropic Index Classifies 26% of August AI R&D Work as Claude-Led

By Simon Yoon

Canonical URL: https://www.tokenpost.com/news/technology/33758
Published: 2026-10-17T05:23:47.000Z
Updated: 2026-10-17T05:23:47.000Z
Section: Technology

> More than 90% of measured work involved AI collaboration or greater automation, but no measured category was fully autonomous.

Anthropic’s August index classified 26% of its measured artificial intelligence research and development work as led by Claude, while no measured work category was fully autonomous.

The index’s “AI leads” category covers tasks where AI completes most of the work from a high-level prompt while a person supervises. Claude was in that category for less than 1% of measured work in February. More than 90% of the August work fell into categories involving AI collaboration or greater automation.

Anthropic based its July measurement on weekly samples covering 20% of staff in model research and development departments. Claude agents reviewed internal work records and identified about 15,000 tasks. The tasks were organized into a tree with 542 nodes.

A model then rated the tasks’ automation levels. Its ratings matched staff assessments exactly 59% of the time and were within one level of those assessments 97% of the time.

A separate poll of 1,334 Anthropic staff, conducted Aug. 9-12, asked about AI’s role in building frontier models. The median estimate was that AI performed 51% of the work involved in building an early-2026 model. Staff’s median expectations were 80% by the end of 2027 and 95% by the end of 2030.

The poll figures are staff estimates, not measured results or a company forecast. They also cover a different question from the index: expected AI involvement in building a frontier model, rather than the share of Anthropic’s broader research and development tasks classified as AI-led. Comparing results across AI labs requires common measurement methods and independent checks.