Anthropic's Economics team has released an interactive scenario explorer that estimates how different paths for artificial intelligence could affect the US economy by 2030. The tool is not a forecast. It lets users enter assumptions about AI capability and adoption, then follows those assumptions through a model of tasks, productivity, output, wages and unemployment.
The framework represents occupations as bundles of tasks using the US Department of Labor's O*NET taxonomy. AI may help a person perform a task, automate it, leave it unchanged or create new work. That distinction matters because an occupation can remain while its composition changes. A nurse, for example, might use AI for documentation or supply ordering while taking on additional review duties and spending more time with patients.
Anthropic presents three broad scenarios. A modest path produces gains within the historical range of major technologies and arrives gradually. Under a substantial-change path, AI is capable of performing half of knowledge work by 2030, much of it autonomously, but adoption remains incomplete. The model has economic growth running at roughly twice its normal pace; knowledge-worker wages do not increase, while other workers see gains.
The extreme scenario assumes AI outperforms people on most knowledge tasks, carries out nearly all of them autonomously and creates few replacement tasks for humans. Under those inputs, annual gross-domestic-product growth reaches 15%, which would double the economy in about four and a half years. The same scenario produces unemployment beyond ordinary recession levels and particularly difficult prospects for knowledge workers. These are modeled results contingent on exceptional assumptions, not claims that such outcomes will occur.
The company also surveyed more than 10,000 Americans in August about expected AI capabilities, adoption and their ability to find new work. Anthropic says the typical responses align most closely with its substantial-change scenario: GDP in 2030 about 10% above a no-AI baseline and unemployment near 5%. Roughly one in ten respondents supplied assumptions resembling the extreme case.
By exposing inputs, the explorer makes uncertainty visible rather than compressing it into a headline number. Outcomes depend not only on what models can do, but how quickly companies deploy them, whether technology augments or replaces people, and how much new work emerges. The exercise also separates aggregate wealth from distribution. A much larger economy can coexist with weaker wages or fewer jobs for particular groups, leaving policy makers to decide how gains are shared.
The useful result is a structured range, not a prediction. Readers can compare their expectations with survey responses and see which assumptions drive economic consequences—while recognizing that the model cannot settle how quickly technical progress or institutional adoption will unfold.
