A new survey from Anthropic highlights a stark divergence in how AI developers and the general public view the near future of automation. While both groups largely agree on current AI capabilities, Anthropic staff anticipate a much steeper trajectory for progress, adoption, and productivity gains over the next four years. The data suggests that proximity to the technology fundamentally alters expectations, with builders expecting AI to handle the majority of their own workload by 2030.

The Capability Gap

The survey, conducted in August 2026, compared responses from a nationally representative sample of US adults with those of Anthropic employees. The most striking difference lies in the expected timeline for breakthrough capabilities. While 39% of US adults believe AI will be able to make a Nobel-level discovery by 2030, 91% of Anthropic staff hold this view. Furthermore, when asked if AI will ever be able to perform certain complex tasks, between 19% and 40% of the general public said 'never,' compared to at most 2% of staff. This suggests that those closest to the development process see no hard ceilings on capability, whereas the public perceives significant limits.

Productivity and Adoption Discrepancies

Expectations for time savings also vary wildly. Among US adults who use AI at work, only 26% report that a task takes a quarter of the time or less when using AI. In contrast, 59% of Anthropic staff report this level of efficiency gain. This aligns with internal data showing that engineers at Anthropic tripled their lines of code added in the month after adopting coding agents. Regarding adoption, the general public expects AI to handle 42% of applicable work tasks by 2030, while staff expect that number to reach 65%. The public cites customer preference for human interaction and data privacy as the biggest obstacles, whereas staff seem to view these as solvable engineering or policy challenges rather than fundamental barriers.

The Self-Automation Paradox

Perhaps the most critical finding for developers and infrastructure builders is the expectation of self-automation. Anthropic staff expect AI to be able to perform 75% of their current work hours by 2030, compared to just 28% for US workers overall. Despite this high level of expected automation, 95% of Anthropic staff believe their company’s headcount will grow in the next 12 months, with the median expectation being a 66% increase. This indicates that in frontier AI labs, automating tasks and shedding jobs are viewed as distinct outcomes. However, this optimism comes with a caveat: staff are slightly more pessimistic than the general public about how quickly displaced workers can find new roles, estimating a 12-month transition period for those needing to switch occupations.

Key Takeaways

  • Anthropic staff expect AI to automate 75% of their work by 2030, versus 28% for the general public.
  • 91% of staff believe AI will achieve Nobel-level discoveries by 2030, compared to 39% of US adults.
  • Staff report significantly higher productivity gains, with 59% saying tasks take 25% of the time or less.
  • Despite high self-automation expectations, 95% of Anthropic staff expect company headcount to grow.
  • The general public sees customer preference and privacy as top adoption barriers; staff see these as minor hurdles.

The Bottom Line

If you are building AI infrastructure, stop listening to the general public's hesitation and start optimizing for the builders' speed. The gap between 'never' and 'soon' is where the market opportunity lies, but it also signals a coming labor crisis that the tech industry is underestimating.