The Expectations Gap: 91% of Anthropic Employees and 39% of Americans Expect a Nobel-Level AI Breakthrough

On October 16, 2026, Anthropic published a study titled "The AI Expectations Gap." The authors — Szymon Sacher, Tess Cotter, Charles I. Jones, Anton Korinek, and Peter McCrory — surveyed a nationally representative sample of Americans alongside the company's own employees. The goal: understand how the two groups assess the future of artificial intelligence. The study uncovered a fundamental divide — not just over how fast AI will advance, but over its impact on the economy and labor market.

Eight Tasks, Two Different Futures

The survey covered eight tasks, ranging from writing routine emails to making Nobel-level discoveries. On AI's current capabilities, both groups agree: AI can handle relatively simple tasks but hasn't produced scientific breakthroughs. On future projections, however, opinions diverge dramatically.

Looking ahead to 2030: • 91% of Anthropic employees and 39% of Americans believe AI will make a Nobel-level discovery • 90% of employees and 38% of Americans expect AI will be able to design, conduct, and publish scientific research • Only 2% of employees think AI will never accomplish any of these tasks, compared to 19% to 40% of Americans

Experience using AI shapes these forecasts. Among Americans who regularly use AI at work, 56% expect AI to handle seven of the eight tasks by 2030 — higher than the overall average (41%), but far below Anthropic employees (96%).

Barriers and the Pace of Adoption

Even if AI can perform a task, that doesn't guarantee it will be used. Americans predict that by 2030, AI will be deployed in 42% of jobs for which it's technically capable. Anthropic employees expect 65%.

Americans see the main obstacles not as technological, but as human preferences and regulatory frameworks: • Customers preferring to interact with humans rather than AI • Concerns about data privacy and security • AI's insufficient reliability • Laws, regulations, and liability issues

Regular AI users at work rank these differently: 42% put privacy first (versus 37% among non-users), while customer preference becomes less critical (37% vs 42%).

Productivity and Time Savings

When estimating time savings on tasks AI is well-suited for: • 85% of Anthropic employees believe the work will take half the time or less • 59% of employees say a quarter of the time or less • Among Americans who use AI, 26% say a quarter of the time or less • Among Americans who don't use AI, that figure is 21%

Anthropic backs up these expectations with real-world data. After rolling out agentic tools internally, employees saw productivity gains: engineers tripled the lines of code shipped per month, salespeople closed more deals, and output per person rose in legal and recruiting.

Automating One's Own Job

The most striking gap concerns how much of each person's own job AI will take over: • Anthropic employees believe AI will handle 75% of their work by 2030 • American workers forecast just 28% • Only 1 in 20 Americans expects 80% or more automation

The numbers vary by industry: manual labor, manufacturing, and transportation workers expect 25% automation, while IT, engineering, and science professionals expect 33%. Even regular AI users in those fields put the figure at 40% — closer to the national average than to Anthropic employees' 75%.

Automation and Unemployment Forecasts

Both groups link expected automation of their own jobs to unemployment forecasts: those who expect greater automation also predict higher unemployment. Notably, at equal levels of expected automation, both groups produce roughly the same unemployment forecasts. In other words, differing views on AI's capabilities translate into economic expectations in the same way across both groups.