Anthropic outlines research uses

Anthropic studied 207,049 research sessions on Claude and Claude Code to see how scientists use the tools. The report says researchers use Claude most for data analysis and writing. Data analysis accounts for about a third of sessions on Claude and nearly half on Claude Code. They use it less for generating ideas and gathering data, even though researchers in an accompanying survey ranked those tasks as most important for scientific discovery.

Use also varies by field. Formal sciences, including computer science, mathematics and statistics, use Claude more relative to the number of researchers in those fields. Health and lab-based physical sciences use it less. Anthropic found that AI use is higher for tasks its measure rates as more suited to AI, and lower for tasks involving physical work. That’s a useful reminder for a business owner: adoption tends to follow what the tool can handle, not necessarily what people consider most valuable.

The report’s output figures are scenarios, not predictions. Its model says that broad adoption of current AI capabilities could double research output in some fields. In a hypothetical case where AI handles all non-physical research work, output could rise as much as 15-fold in computer science and statistics, while physical bottlenecks limit the modeled gains in chemistry and biology to three or four times. Those estimates depend on assumptions about AI capability and adoption, so expert judgment and verification still matter. Anthropic’s report offers evidence to consider when planning an AI command centre, alongside decisions about where teams should check the work.