A research report titled “The AI expectations gap” published by Anthropic on October 16, 2026, painted a deeply polarized picture of the future of artificial intelligence. Survey data collected in August 2026 from 10,980 U.S. adults (conducted by polling firm Morning Consult) and 1,334 personnel working at Anthropic revealed a massive gap in confidence regarding the pace of technological progress between those directly building models and the general public.

This gap is most apparent in the capability to solve complex scientific challenges. As many as 91% of Anthropic personnel believe AI will be capable of producing a Nobel Prize-level scientific breakthrough by 2030, compared to just 39% among U.S. adults. Even over a shorter timeframe by 2027, roughly 90% of lab staff expect AI to independently design, execute, and publish scientific research, compared to a 38% forecast by the general public.

Job threat levels and time-saving efficiency

When evaluating the impact of AI on their own job roles, the divergence in perceptions becomes even sharper. On average, Anthropic employees estimate that AI could handle up to 75% of their current workload by 2030; more than half of lab personnel believe that figure will reach 80% or more. In contrast, American workers forecast AI taking over only about 28% of their work time, and a mere 5% of workers believe 80% of their job could be automated.

The difference in confidence stems partly from hands-on experience with agentic tools. Among frequent AI users in the U.S., only 26% reported that relevant tasks were completed at least four times faster (taking one-quarter of the time or less). At Anthropic, the share of staff reporting similar time savings reached 59%, reflecting a substantial leap in productivity when automated workflows are incorporated into daily operations.

Similarities in labor market forecasts

Despite the wide divide over technological capabilities, both groups share a similar logic regarding the labor market. Both tech employees and the public project that overall unemployment will rise as the share of tasks displaced by AI increases. Specifically, workers who expect 80% or more of their jobs to be automated also projected that the U.S. unemployment rate in 2030 would reach 12.2%, surpassing the peak of the 2008 global financial crisis.

It should be noted that these figures reflect the expectations and subjective sentiments of survey respondents rather than a definitive economic forecasting model. AI researchers often tend to linearly extrapolate model capabilities observed in the lab, whereas deploying technology into real-world operations faces barriers involving legal regulations, data security, and consumer adoption habits. Are the engineers on the front lines of the AI wave truly anticipating the future, or are they falling into the trap of their own techno-optimism bias?