People who build frontier AI expect it to change the world much faster than everyone else does. That is not a surprise, but Anthropic has now measured the size of the gap. In August 2026 it surveyed a nationally representative sample of US adults and its own staff with the same questions about what AI can do, how widely it will be used and what it will mean for jobs.
The two groups largely agree about the present and split sharply on the future. Asked when AI will match a skilled professional on eight tasks, from routine emails to a Nobel-level discovery, both ranked the tasks in nearly the same order of difficulty, and about a quarter of each group thought AI can already design, run and publish a scientific study. But 91% of staff expect a Nobel-level discovery by 2030, against 39% of the public. On their own jobs, staff expect AI to be able to do 75% of their current work by 2030 on average, against 28% for US workers, and only one US worker in 20 thinks it could do 80% or more of theirs. Staff also report far bigger time savings: 59% say a well-suited task takes a quarter of the time or less with AI, against 26% of US adults who use AI at work.
The most interesting result is what the groups share. Told that unemployment is about 4.2% and asked to forecast 2030, people who expect AI to do more of their own work predict higher unemployment, and at the same level of expected exposure, staff and the public give roughly the same forecast. US workers who expect AI to do 80% or more of their work forecast 12.2%. The disagreement, in other words, is about what AI will be able to do, not about how the economy would respond. Use matters too: regular users at work sit between the public and the staff.
For readers, this is a useful reminder of whose expectations shape the products they use. The people at Anthropic, building Claude and using agents all day, live in a different version of the near future from most of its customers, and a companion report on Anthropic's own workforce suggests their experience is not imaginary.
The limits are plain. The staff are not neutral forecasters: they work on the technology and have a financial stake in its success. Their productivity gains come from a company built around AI, which is not typical. And a survey of expectations is not a forecast of outcomes; both groups could be wrong in either direction. The note is best read as a careful view from inside the lab, set beside public opinion, not as evidence of what 2030 will look like.