Anthropic has published the largest qualitative study of AI-at-work conducted to date, based on open-ended interviews with 81,000 Claude users across 159 countries and 70 languages over a single week in December 2025. An AI interviewer conducted adaptive, multi-turn conversations with each participant; AI classifiers then tagged every response across multiple dimensions.
The headline number is the productivity self-assessment: an average of 5.1 out of 7 across the user base. More interesting than the aggregate is the shape of the gain. 48% of users who explicitly mentioned productivity credited 'scope expansion' — doing new kinds of tasks they previously couldn't attempt — as the dominant benefit. 40% cited raw speedups on existing work. The two groups are not the same people, and the distinction matters for thinking about where AI's economic impact will actually land.
The distribution of gains is counterintuitive. Users in the highest- and lowest-paid occupations reported the largest productivity impacts, with middle-income workers reporting relatively smaller effects. High-wage knowledge workers benefit because their work is largely text and judgment; low-wage workers benefit because AI unlocks capabilities previously gated behind formal education. Middle-income workers, whose jobs often include more routinized procedures and regulated workflows, report the smallest gains.
The displacement concern is the most cited tension. About one in five respondents expressed active worry about AI-driven job displacement. Critically, concern correlated with usage intensity — the users experiencing the largest speedups from AI were also the most likely to worry their role would be automated. Anthropic frames this as a signal the company itself has to take seriously: the people most productive with its product are also the ones most anxious about what it means for them.