Mercor finds workers want AI to handle routine tasks, not high-stakes decisions

Workers in Mercor’s new survey want help with routine work without surrendering the decisions and relationships they consider central to their jobs. The company published survey findings on November 11 showing a preference for human involvement across most task ratings—and more interest in using saved time for enjoyable work than shorter hours.

Mercor surveyed more than 200 participants from its original TaskNet study about their work, the tasks they see as core and their occupation’s future. Each received a list of tasks for their occupation, rather than a single broad question about whether they welcomed AI.

For every task they performed, respondents rated how much they would want AI to do it if it could complete the task entirely. They also rated enjoyment, setting salary aside, and how much human-AI collaboration would be needed. Workers could select tasks to hand over or keep, then explain why.

Help with the task, not a blank check

Across roughly 9,000 task ratings, 14% said no human-AI collaboration would be needed, while 27% called it essential. Most ratings fell between those endpoints. Lawyers marked collaboration essential in 38% of their ratings, compared with 19% for software developers.

The dividing line often followed responsibility and human relationships. Workers wanted to retain activities involving consequential decisions or direct care, while frequently selecting clerical work for automation. The occupational examples make that distinction concrete:

• Software developers wanted to keep mentoring junior colleagues and reviewing technical designs, while handing off software-build testing.
• Emergency physicians wanted to retain leading resuscitations but delegate keeping medical records up to date. Lawyers wanted to keep client interviews and advice about risks and options.
• Most teachers wanted AI to handle data analysis and material production, while keeping daily lessons and helping children develop emotional skills.

I would like AI to listen to my patient encounter and show me a list of tests to approve based on the discussion. I do not want it to autonomously order the tests without my approval.

An ER physician surveyed by Mercor

Saved time does not necessarily mean less work

Asked what they would most likely do with time freed by automating their preferred tasks, 38% chose more work they enjoy. Another 22% chose the same work at higher volume, 18% would move into new kinds of work, and 15% would work fewer hours. These were intended uses of hypothetical savings, not measured changes in working hours.

Occupation shaped those answers. Elementary school teachers were most drawn to doing more enjoyable work, at 51%. Emergency physicians were most inclined to increase the volume of the same work, at 36%, while dishwashers were most likely to choose new kinds of work, at 43%.

More change, more worry—and an AI-heavy sample

Mercor also found that workers in occupations with higher AI-exposure scores more often reported substantial job changes over the past year and concern about their occupation’s future. Emergency physicians reported less change and worry than their scores would suggest. The scores assess how feasible it would be to create environments where AI could practice tasks and learn from feedback.

An optional application survey introduced in mid-October drew responses from more than 4,000 workers across additional occupations. The association between exposure and reported job change held there too, with medicine again below the trend.

Mercor’s sampling warning is substantial: people on its platform are unusually interested in and engaged with AI. An earlier module found that 82% of respondents describing jobs outside Mercor already used generative AI for them, against a national figure of 45% cited by the company. Mercor expects that difference to heavily bias its sample.