Anthropic’s economics team says AI could push the U.S. economy onto very different paths by 2030, but in every scenario it tested, the same question keeps surfacing: how much work gets replaced, and how quickly. In the most extreme case in its research project, Our Economic Future Scenarios, overall unemployment reaches 11.9%, while GDP ends up 32.4% higher than in a world without AI.
Jobs are modeled as bundles of tasks
Instead of treating occupations as the unit of analysis, the technical report breaks each job into a set of tasks. Those tasks are then sorted by whether AI is likely to enhance them, automate them, or leave them mostly unchanged.
The report uses nurses as an example. Drawing blood, ward rounds, triage, recording vital signs, and calling for supplies all sit on the task list. AI cannot bathe a patient, the report notes. But it can help draft discharge instructions, support remote monitoring, and handle scheduling, which fall into the “augmentation” category. Repetitive duties such as recording vital signs or calling for supplies can be directly automated.
Anthropic also argues that AI may create new tasks of its own, such as checking whether AI-generated triage decisions are accurate. Under that framework, whether a job disappears depends less on its title and more on what happens to each task inside it.
Three scenarios, three growth paths
After applying that task-based logic across the U.S. economy, Anthropic modeled three scenarios: mild, substantial, and extreme.
- In the mild scenario, 2030 GDP is just 1.6% above a no-AI world, with annual growth at 2.4%. The report says the effect is roughly on the scale of the internet.
- In the substantial scenario, GDP is 8.3% higher and annual growth doubles to 5.4%, with an impact larger than the internet and railroads.
- In the extreme scenario, the assumptions are much stronger: self-improving AI systems appear, and companies adopt them at an unusually fast pace. Under those conditions, GDP jumps 32.4% above the no-AI baseline and annual growth reaches 15.4%.
Wages, unemployment, and income shares move unevenly
The study says the distribution of gains and losses is highly uneven across the three paths.
- In the mild case, there are almost no losers. Wages for knowledge workers rise 0.4%, wages for other workers increase 1.1%, and labor’s share of income slips from 60% to 59.4%.
- In the substantial case, cracks begin to show. Wages for knowledge workers fall 0.3%, while wages for other workers climb 5.9%. Labor’s share drops to 56.1%, and capital’s share rises to 43.9%.
- In the extreme case, the gap widens sharply. Wages for knowledge workers fall 11.5%, while wages for other workers surge 33.6%. Labor’s share falls to 45.2%, and capital’s share rises to 54.8%.
Anthropic says that in the extreme scenario, capital income is 81.4% higher than in a world without AI, while labor income is only 0.5% higher. The economy is larger, but the report’s numbers show that the added output is distributed far more heavily toward capital than labor.
Public expectations cluster near the substantial case
In August, Anthropic commissioned Morning Consult to survey more than 10,000 Americans. The general public sample included 10,980 people, and the website visitor sample included 10,559.
Using the median respondent’s assumptions about AI, the implied 2030 outcome is a GDP level 8.6% above the no-AI path, with an interquartile range of 3% to 19%. Overall unemployment comes in at 4.6%, employment for knowledge workers falls 4.2%, and wages for knowledge workers rise just 0.6%. That set of results lands close to the report’s “substantial” scenario.
At the same time, Anthropic says about 10% of respondents gave answers that already line up with the most extreme version of the future.
People disagree on retraining and time savings
The survey also found a gap between how quickly people think job switching happens and how long they expect it would take in practice. Many assume changing jobs takes only a few weeks, but the median respondent expected it would take about eight months to find work again.
Views on AI’s ability to save time were also split. About 30% of respondents said AI would save no time at all, while 49% said it could save at least half of the time. Anthropic says that divide shows how far apart public expectations still are when it comes to AI capability.
The bigger question is who captures the gains
The report’s main focus is not only whether AI will replace jobs, but who ends up keeping the gains from faster growth. Across all three scenarios, stronger AI leads to faster economic expansion and a larger income share for capital, while labor’s share declines.
Even if average wages still rise in nominal terms under the extreme case, the outlook for knowledge workers worsens. The report says knowledge-intensive roles such as engineers and customer service staff may need to move toward occupations that are harder for AI to affect, such as electricians or nurses, though switching fields is rarely easy.
Anthropic says it is linking the model to research it funds and to policy proposals, while stressing that the economic gains from AI should be shared more broadly.