The Great AI Expectations Gap

Overview

A new survey from Anthropic highlights a stark divide in how the American public and AI insiders perceive the trajectory of artificial intelligence. Published on October 16, 2026, the research reveals that while both groups agree on current capabilities, they fundamentally disagree on the pace of future progress and its economic implications.

The study, conducted by economists Szymon Sacher, Tess Cotter, Charles I. Jones, Anton Korinek, and Peter McCrory, compares a nationally representative sample of US adults against Anthropic staff. The findings suggest that those closest to the technology anticipate a far more rapid and disruptive transformation than the general public.

"Forecasts vary widely and are fiercely debated," the authors note. "This note presents new survey evidence on these questions, focusing on how beliefs about AI capabilities translate to expectations about adoption and labor market outcomes."

Diverging Timelines for AI Capabilities

The survey asked respondents to estimate when AI would match skilled professionals across eight distinct tasks, ranging from writing routine emails to making Nobel-level scientific discoveries. Both US adults and Anthropic staff largely agreed on what AI can achieve today, ranking tasks by difficulty in nearly identical orders.

However, the divergence becomes pronounced when projecting into the future. By 2027, approximately 90% of Anthropic staff expect AI to be capable of designing, conducting, and publishing scientific studies. In contrast, only 38% of the general public shares this optimism for the same timeframe.

The gap widens significantly by 2030. While 91% of Anthropic staff believe AI will be able to make a Nobel-level discovery within four years, only 39% of US adults hold that view. Furthermore, between 19% and 40% of US adults suggest that certain complex tasks will never be achievable by AI, whereas no more than 2% of Anthropic staff agree with such finality.

Frequency of use appears to influence these outlooks. Among respondents who use AI at work regularly, 56% expect AI to handle at least seven of the eight listed tasks by 2030. This figure is substantially higher than the 41% seen among all US adults and the 28% among those who have never used an AI tool, yet it still falls well short of the 96% expectation held by Anthropic staff.

Adoption Rates and Productivity Gains

Beyond raw capabilities, the survey examined how widely AI would be adopted in the workplace and the resulting productivity gains. US adults project that AI will be utilized for 42% of eligible work tasks by 2030. Anthropic staff, however, anticipate a much faster diffusion, expecting AI to handle 65% of applicable tasks within the same period.

When it comes to time savings, the disparity is equally sharp. Among US workers who currently use AI at their jobs, 26% report that AI reduces the time required for a task to a quarter of the original duration or less. Among Anthropic staff, that figure rises to 59%.

These differences in expected adoption and efficiency underscore a fundamental disconnect in how each group envisions the integration of AI into daily economic activities. The general public tends to view adoption as a slower, more gradual process constrained by external factors, while insiders foresee a near-instantaneous scaling of utility.

Workplace Automation and Unemployment Concerns

One of the most striking findings concerns the impact of AI on individual jobs. Over half of Anthropic staff expect AI to be capable of performing at least 80% of their current work by 2030. Comparatively, only 5% of US workers believe their own jobs will face such a high degree of automation. On average, staff members expect AI to handle 75% of their work, versus 28% for US workers overall.

Despite these differing expectations about personal job security, both groups reason about the broader labor market in a remarkably similar way. Individuals who anticipate that AI will automate more of their own work consistently forecast higher overall unemployment rates, regardless of whether they are staff or members of the general public.

Crucially, at any given level of expected personal automation, both groups predict approximately the same unemployment rate. This suggests that while insiders are more optimistic about AI’s technical potential, their economic reasoning regarding displacement aligns closely with that of the wider population.

Barriers to Adoption

The survey also explored what might hinder the widespread adoption of AI in the workplace. When asked to identify the biggest obstacles, US adults prioritized non-technological constraints, such as people's preferences and regulatory rules, over purely technical limitations.

Anthropic staff, accustomed to building and deploying frontier models, may perceive technological barriers as more surmountable. Their higher expectations for both capability and adoption suggest they view the path to integration as primarily one of refinement and scaling, rather than one blocked by societal or regulatory friction.

This distinction highlights a key challenge for policymakers and businesses: bridging the gap between technical feasibility and social acceptance. As the data shows, the technology may be advancing faster than public trust and institutional readiness, creating a significant expectations gap that could slow real-world economic impact.