Anthropic's Survey Finds a Huge Gap Between How the Public and Its Own Staff See AI's Future
Anthropic surveyed about 11,000 US adults and 1,334 of its own staff. They agree on what AI does today, but not on how fast it will improve or how much of the work it will take over.
Anthropic has published a study with a blunt title, “The AI expectations gap.” It compares what ordinary Americans and Anthropic’s own employees expect from AI by 2030. The main finding: both groups mostly agree on what AI can do today. They disagree sharply about how quickly that will change.
The numbers are striking. By 2030, 39 percent of US adults expect AI to make a Nobel-level discovery, against 91 percent of staff. For any given task, between 19 and 40 percent of adults say AI will never do it; at most 2 percent of staff say that. Adults expect AI to handle 42 percent of the tasks it could do by 2030, staff expect 65 percent. Staff expect AI to do about 75 percent of their own working hours by then, while US workers on average say 28 percent. Inside the lab, staff estimate AI already does a median 51 percent of the work of building a frontier model, rising to 95 percent by 2030.
The study also links expectations to worry. US workers who expect 80 percent or more of their work to be automated forecast 12.2 percent unemployment, while those expecting less than half forecast 7 to 8 percent. The authors stress this is an association, not proof of cause. The survey data also feeds an economic scenario model, whose median results for 2030 GDP growth range from 5.4 percent (typical US adult) to 17.1 percent (typical staff member).
How to read this: the report comes from a company that sells AI, surveyed its own enthusiastic employees, and says so. The US sample is an opt-in online panel of 10,980 people (Morning Consult, August 2026), the staff poll is unweighted, and the website quiz attracted people already interested in AI. Everything here is self-reported expectation, not measurement, and the GDP figures are scenario outputs, not forecasts. Still, the gap itself is useful: people closest to the technology are very bullish, and you can decide how much to discount for that.
What this means for you: nobody, including the people building it, knows the real pace. If you are new to AI, the sensible reaction is curiosity, not panic or dismissal: try the tools on your own tasks and see what saves time. In the survey, 29 percent of AI-using adults said it saved no time or took longer, which is a useful reminder that results vary by task. If you are further along, treat headline predictions, ours included, as inputs rather than answers.
Sources
Source: https://www.anthropic.com/research/ai-expectations-gap
Anthropic Lets an Outside Watchdog Work Inside Its Walls
Anthropic has given six staff from the nonprofit METR employee-like access to its models and safety work. METR can publish what it finds, though Anthropic may redact a few defined categories.
