Anthropic employees expect artificial intelligence to handle roughly 75% of their work by 2030, while the broader US public puts that figure at just 28% — a wide divergence documented in the AI developer's own research. The people building advanced AI think it will take over most of their jobs within a few years; the rest of the country is considerably less convinced.

The findings appear in "The AI Expectations Gap," a report published by Anthropic on October 16, 2026. It compares how the company's own employees see AI's trajectory with how the general public views the same technology. Because both groups were asked the same questions, the report works as a single side-by-side measure of how far insider and public assumptions have drifted apart.

The numbers behind the gap

Anthropic surveyed 1,334 of its staff members internally in August 2026, then set their answers against responses from US adults. The headline split is the 75%-versus-28% figure on how much work AI will handle by 2030. Neither number claims to measure what AI will actually do by then — both are expectations, which is what makes the 47-point distance between them the report's central data point.

More than half of Anthropic employees expect at least 80% of their tasks to be automated. Among the broader US workforce, just 5% share that view.

Some 91% of Anthropic staff believe AI will produce a Nobel-level discovery by 2030. Only 39% of US adults agree. That question concerns scientific capability rather than jobs, and it shows the same insider-versus-public pattern.

Expectations also diverge on the labor market. Anthropic staff predict US unemployment will exceed 12% by 2030, while the general public forecasts a rate near 8%. On growth, the median expectation among Anthropic employees is 17.1% GDP growth for 2030; US adults put the median at 5.4%.

Productivity perceptions differ as well. Among US adults who use AI at work, 26% say AI assistance lets them finish tasks in 25% of the usual time. Among Anthropic staff, that share is 59%.

Inside the building

Internally, AI agents were reported to have worked six times the hours of human employees by August 2026. The share of junior technical hires at Anthropic reportedly fell from 27% to 9%. Separately, AI is currently managing 51% of frontier model-building tasks, with forecasts suggesting this will escalate to 80% by the end of 2027 and 95% by 2030.

Those figures sketch the operational backdrop to the survey: this is a company that, by its own accounting, already relies on AI for the majority of its frontier model-building work.

What this means

For policymakers, the unemployment gap is the one that stands out. A forecast above 12% is a very different planning scenario from one near 8%.

The 17.1%-versus-5.4% growth split captures the tension neatly. One number describes a historic economic boom. The other describes a solid but recognizable economy.

The internal projections also come with dates attached — 80% of frontier model-building tasks by the end of 2027, 95% by 0 — which makes them the report's most directly checkable claims and a yardstick for the insider outlook as 2030 approaches.