An automated task tells you less about a job than you might think. Anthropic economist Peter McCrory discusses that gap with Harvard’s Jason Furman, drawing on observed use of Claude. He describes a pattern of AI helping skilled workers rather than replacing every part of their jobs.
McCrory says the data discussed doesn’t show an occupation where Claude automates every task. That finding concerns the work people currently give the model; it doesn’t prove job losses won’t happen. Helping one person can also replace work that another person used to do. The effect depends on whose job you follow.
So we’d start by naming the task and its place in the day. If software drafts a reply, someone may still need to resolve the complaint behind it. You can measure the minutes saved without pretending the whole role has disappeared. Then check who inherits the review and the awkward cases. A useful change should make those responsibilities easier to see.
Peter McCrory, Anthropic, in conversation with Jason Furman. Published by the Institute of Politics at Harvard Kennedy School.
Original interview published September 23, 2026. Commentary added September 28, 2026.
