Anthropic study: AI could nearly double research speed, and up to 15x output in some fields under full automation

Anthropic published a research report on November 5, "AI for Scientific Discovery: How Researchers Use Claude", analysing 207,049 research sessions between August 2 and September 30, 2026, across Claude, Claude Code and Claude Science, the latter launched in beta on October 14. The report finds researchers most often use AI for data analysis and writing, not for the idea-generation work they themselves rate as most important. At current capabilities, Anthropic estimates AI could speed research by roughly 1.35 to 2.02 times.

Data analysis dominates; idea generation used least

Data analysis accounts for 33.7% of Claude sessions and 48.1% of Claude Code sessions, while researchers spend about 27.6% of their time on the task. Writing and communication make up 32.0% of Claude sessions, about twice the 16.0% share of researcher time. By contrast, idea generation and hypothesis formation — which researchers rank as most important — account for only 4.2% of Claude sessions and 2.1% of Claude Code sessions. Computer science accounts for 32.7% of Claude Code research sessions despite computer scientists making up only 4.3% of researchers worldwide, a usage intensity about 7.6 times their share of researchers; the formal sciences (computer science, mathematics and statistics) are all heavily over-represented in AI use.

About 2x faster today; full automation is only a hypothetical

Modelling current capabilities, the report estimates research could accelerate by between 1.35 times in the arts and 2.02 times in statistics, with computer science at about 1.96 times. In a hypothetical scenario where AI carries out all research tasks that do not require physical work, output in fields such as computer science and statistics could rise by as much as 15 times, while laboratory-based fields such as chemistry and biology would reach only about three to four times because of physical bottlenecks. Anthropic stresses the 15-fold figure is a theoretical ceiling under that scenario, not a forecast. Some 40% of surveyed researchers expect AI to be able to conduct full research projects independently by 2030; the report also warns that wider AI use could reduce human oversight and raise ethical concerns.

(Sources: Anthropic; AI Weekly; Blockchain News.)