Anthropic analyzed 207K research sessions: scientists lean on Claude for data analysis but barely use it for ideation

dotey · x · 2026-11-08

Anthropic released a research report on Nov 4 analyzing 207K science-related Claude sessions (Aug 2–Sep 30, 2026) plus a survey of 528 scientists.

Key findings

Data analysis dominates: 33.7% of research sessions, rising to 48.1% within Claude Code, versus 28% of scientists' research time.
The most valued tasks are least automated: Idea generation is 7.1% of research time but only 4.2% of Claude usage (2.1% on Claude Code); scientists ranked it most important (35.6/100). 70% either don't use AI for ideation or only lightly. Data collection (28.6% of time) gets just 12% of usage since it's hands-on.
Writing is over-used relative to its time share: 16% of time but 32% of usage, mostly paper drafting.

Discipline skew: CS researchers (4.3% of researchers) account for 32.7% of Claude Code research sessions; AI/ML is 1.2% vs 15.2%. Medicine/clinical, 22% of researchers, is only 2.7%.

Projected acceleration: With current AI capabilities diffused, output could at most double in some fields; if AI handled all non-physical work, math/stats/philosophy/economics could reach 16x and CS 15x, while lab-bound fields like chemistry and biology cap at 3–4x since 40% of time is hands-on.

Original post →

Anthropic analyzed 207K research sessions: scientists lean on Claude for data analysis but barely use it for ideation

More from Models