# Anthropic staff expect a faster AI future than US workers, study finds > An Anthropic study finds large differences between staff and US adults in expectations for AI by 2030. Its GDP figures are scenarios derived from survey answers, not company or economic forecasts. By BIG CHANGE Editorial Published: 2026-10-18T10:52:34.846Z Updated: 2026-10-18T10:52:34.846Z Canonical: https://bigchange.ai/blog/anthropic-ai-expectations-gap-us-workers ![Conceptual charcoal illustration of two separate trays holding blank survey forms, divided by an orange line.](https://bigchange.ai/api/media/file/anthropic-two-surveys-concept-v1.png) AI-generated conceptual illustration by BIG CHANGE; no actual Anthropic respondents, survey materials, results or GDP forecast are depicted. Anthropic staff and US adults see a similar range of things AI can do today. Their expectations for 2030 diverge sharply, according to [a research report published October 16](https://www.anthropic.com/research/ai-expectations-gap). The gap matters for decisions about work and policy, but the study measures beliefs, not the pace AI will actually achieve. The report asked when AI would perform eight tasks as well as a skilled professional. By 2030, 91% of surveyed Anthropic staff expected AI to be able to make a Nobel-level discovery, compared with 39% of surveyed US adults. On their own jobs, staff expected AI to be able to perform an average of 75% of their current work hours by 2030. The average among employed US respondents was 28%. Those figures describe expected capability, not work already automated or jobs certain to disappear. ## The big change - **What changed:** Anthropic published a comparison of public and staff expectations across AI capability, workplace use and economic impact. - **How it works:** The researchers surveyed the two groups, then put complete answers to five economic questions through a previously published model to produce 2030 scenarios. - **Why it matters:** Workers, managers and policymakers can see where expectations differ, while keeping survey beliefs separate from measured adoption and economic forecasts. ## A comparison with two different samples Morning Consult surveyed 10,980 unique US adults through an online opt-in panel from August 11 to 23, 2026. The responses were weighted on demographics to reflect the US adult population. Online recruitment excludes people without internet access, and weighting cannot correct every difference between volunteers and the population. Questions about the future were split between 2027 and 2030; the report mainly presents the 2030 answers. Anthropic's separate, unweighted internal poll ran from August 9 to 12 and received 1,334 responses across the company. Its employees are neither a representative sample of US workers nor a statement of Anthropic's corporate plans. The [technical appendix](https://www-cdn.anthropic.com/1bc1a3067f9c9883d6d556b38f26fd4f1e1b640f/The%20AI%20Expectations%20Gap%20-%20Appendix.pdf) gives the sample and weighting details. The groups also differ in how they see adoption. Of tasks they expect AI to be capable of doing, US adults expected it to be used for 42% by 2030, against 65% among staff. Among US adults who use AI at work, 26% said a task suited to today's AI takes a quarter of the time or less with AI, counting checks and fixes. That share was 59% among staff. These are respondents' assessments of suitable tasks, not measured economy-wide productivity gains. ## Where the GDP figures come from The researchers took complete answers from 3,259 US adults and 1,093 staff on five inputs: capability, adoption, the extent of automation, time savings and how long an AI-displaced worker might take to find work in another occupation. They mapped each person's answers into their [economic scenarios model](https://www.anthropic.com/institute/econ-scenarios), holding its other parameters at the values of the model's “Substantial” scenario. On that basis, the median US adult's answers implied annual GDP growth of 5.4% in 2030; the median staff member's implied 17.1%. The distributions overlapped widely. These are conditional model results built from expectations, not observed growth, predictions that either group will prove right or an Anthropic company forecast. The model's assumptions also matter: changing inputs outside the five survey-derived dimensions could change the result. The practical question is where capability turns into use. US respondents most often selected customer preference for a person as a prospective obstacle to workplace adoption, followed by privacy and security, reliability, and law or liability. Each could choose up to three; staff were not asked that obstacle question. For an employer weighing AI deployment, those answers point to concerns worth examining, while the survey cannot establish which barriers will bind in 2030. This article continues our [coverage of Anthropic's October 16 workforce research](https://bigchange.ai/blog/anthropic-junior-technical-hiring-workforce-study). That report examined hiring and work inside the company. This one compares expectations across a public survey and an internal poll. ## Sources & further reading - [The AI Expectations Gap, Anthropic Institute](https://www.anthropic.com/research/ai-expectations-gap): October 16, 2026 report with the comparisons, question descriptions and model-implied GDP results. Its footnotes describe the two samples and their limits. - [Appendix to The AI Expectations Gap](https://www-cdn.anthropic.com/1bc1a3067f9c9883d6d556b38f26fd4f1e1b640f/The%20AI%20Expectations%20Gap%20-%20Appendix.pdf): Sample construction, weighting and response details. - [Economic Scenarios for Transformative AI](https://www-cdn.anthropic.com/files/4zrzovbb/website/cf58f84d46a4a76bf5a5b039ac695fba6b80041c.pdf): Technical model and Appendix B mapping survey answers into five parameters; the remaining parameters are fixed for this exercise. - [Scenarios for our Economic Future](https://www.anthropic.com/institute/econ-scenarios): Anthropic's public explanation and interactive explorer of the scenario model, separate from the research samples. ## Sources - [The AI Expectations Gap](https://www.anthropic.com/research/ai-expectations-gap) — Primary comparison and Figure 7; its footnotes define samples and clarify that respondent-derived GDP numbers are model-implied scenarios. - [Appendix to The AI Expectations Gap](https://www-cdn.anthropic.com/1bc1a3067f9c9883d6d556b38f26fd4f1e1b640f/The%20AI%20Expectations%20Gap%20-%20Appendix.pdf) — Sample construction, demographic weighting, opt-in panel limits and unweighted internal staff poll. - [Economic Scenarios for Transformative AI](https://www-cdn.anthropic.com/files/4zrzovbb/website/cf58f84d46a4a76bf5a5b039ac695fba6b80041c.pdf) — Technical model, page 29 and Appendix B; five responses mapped into parameters while all others use the Substantial scenario values. - [Scenarios for our Economic Future](https://www.anthropic.com/institute/econ-scenarios) — Public scenario explainer. Its self-selected visitor quiz is separate from the August research samples.