Anthropic models how AI could reshape the US economy by 2030

The economy, broken into tasks

The model treats jobs as bundles of tasks. AI might help someone do a task faster, automate it, leave it unchanged or create new work. The task lists draw on the US Department of Labor’s O*NET classification.

A nurse, for example, may check patients’ vital signs, prioritize admissions and order supplies. AI could help with those tasks, while also creating new ones, such as checking whether an AI system has prioritized patients appropriately. As workers use AI to handle more tasks, the model assumes, productivity can rise.

The scale matters: add up the tasks performed by people, machines and software, and the result is the economy. The US economy produced more than $30 trillion in value over the last year, according to the article. AI’s effect depends not just on what it can do, but on how quickly companies and workers adopt it and whether it assists people or replaces their work.

Anthropic’s three scenarios differ in the pace and reach of that change:

The extreme case would likely require AI systems capable of repeatedly improving themselves. In that scenario, knowledge-work jobs shrink sharply and unemployment rises above levels typical of recessions. The model’s more substantial scenario is less dramatic: knowledge workers’ wages do not rise, while other workers receive increases.

Growth is not the same as shared gains

In August, Anthropic surveyed more than 10,000 Americans about AI’s capabilities, adoption and the difficulty of finding another job after a career change. Typical answers aligned with the substantial scenario: by 2030, GDP is 10% higher than it would have been without AI, while overall unemployment reaches about 5%. Around 10% of respondents held views close to the extreme scenario.

The model tracks more than GDP. In its substantial and extreme cases, labor’s share of the economy falls and capital’s share rises. Average wages still increase, largely because workers outside knowledge professions earn more; wages for knowledge workers may stagnate or decline, alongside rising unemployment.

Today, the article says, workers receive about 60 cents of every dollar produced, with capital owners receiving 40 cents. In the model’s three scenarios, capital’s share rises by 0.6, 3.9 and 14.8 percentage points:

People displaced from knowledge-work jobs may need to move into occupations less exposed to AI, such as electrician or nurse. That transition can take time, and the model projects that in the extreme case many affected workers could remain out of work for a long time. Across scenarios, jobs in AI-exposed professions decline from 2026 to 2030, while jobs in professions not affected by AI increase.

I think the most important distinction here is between a larger economy and a better outcome for workers. Anthropic’s scenarios suggest that productivity gains can coexist with stagnant pay for the people whose work is most exposed, while the owners of capital capture a larger share of the growth. The headline GDP number alone would miss that distributional shift.

What the model leaves out

Anthropic describes the interactive model as a simplified tool, not a forecast. Its first version, Version 1.0, was released in September 2026. It does not include policy responses, business cycles, shocks to aggregate demand or financial markets, or catastrophic risks. It also has no scenario in which humanity creates super-capable robots.

External reviewers flagged other limits: the model does not follow individual workers, and it assumes AI-exposed occupations may shrink even though they could also grow. Reviewers also questioned whether the extreme case is better treated as a thought experiment than a prediction, and whether the moderate case understates changes already underway. The model does not account for the effect of data-center construction on aggregate demand, and some reviewers thought it might underestimate AI’s ability to accelerate technological progress.

Anthropic says it will continue developing the model as data and economic research evolve. What I’d want to know is how the results change when those missing forces are included. For now, the model’s strongest message is not that one scenario will happen, but that even a fast-growing economy can leave the workers most affected by AI with less bargaining power and a smaller share of the gains.