Anthropic says Claude "leads" 26% of its AI R&D work, up from 1% in March, and "collaborates" on 90%+, doing large chunks of work under close human direction

The AI you use for tasks can now complete most of them independently, freeing up human collaborators.

Anthropic announced that its AI chatbot, Claude, now leads 26% of the company's AI research and development (R&D) work, a significant increase from just 1% in March. The company defines 'leads' as Claude completing most of a task end-to-end from a high-level prompt, with human supervision. Furthermore, Claude collaborates on or performs large chunks of work under close human direction in over 90% of Anthropic's R&D efforts. These metrics were developed by Anthropic as part of a proposed framework for communicating the pace of AI development. The company also suggested measures for AI agent oversight and compute allocation to promote transparency and facilitate responses to potentially rapid AI advancements. Anthropic emphasized that Claude is not operating fully autonomously in any measured subset of its R&D work.

Anthropic's disclosure highlights a rapid integration of AI assistants into core R&D processes, moving beyond mere collaboration to leading task completion. This signifies a broader industry trend where AI is not just a tool but an active participant in innovation, capable of executing significant portions of complex work with human guidance. The metrics proposed by Anthropic suggest a move towards greater accountability and transparency in AI development, aiming to provide industry-wide standards for tracking progress and safety.

This evolution impacts how R&D teams operate, potentially accelerating project timelines and shifting human roles towards supervision, validation, and strategic direction. As AI models like Claude become more capable of end-to-end task execution, companies will need to adapt their workflows and oversight mechanisms. The focus on 'human supervision' indicates that while AI is taking on more leadership, human oversight remains critical for safety and quality control in advanced AI R&D.