Anthropic has released an interactive modeling tool that lets users plug in their own assumptions about artificial intelligence and watch what happens to the US economy.
The range of outcomes is wide. On one end: a gentle productivity nudge that barely registers. On the other: GDP growth exceeding seven times the current pace, paired with unemployment hitting 14%.
How the model works
Available at anthropic.com/institute/econ-scenarios, the tool lets users adjust several key variables: how capable AI becomes, how quickly businesses adopt it, whether it augments human workers or replaces them outright, and how fast displaced workers find new jobs.
The model draws on data from the company’s Anthropic Economic Index, or AEI, a research program that analyzes millions of anonymized conversations to measure how AI is actually being used across industries and geographies.
Among the AEI’s findings so far, directive and automated AI usage climbed from 27% to 39% in early measurement periods, with business and API users leading the charge. Task speedups averaged around 80%, and for complex work the gains were even more dramatic, ranging from 9 to 12 times faster completion.
Productivity analyses embedded in the model estimate that AI could boost US labor productivity growth by roughly 1% to 1.8% annually. For context, US labor productivity growth has averaged around 1.4% per year over the past two decades.
The extreme scenarios are where it gets uncomfortable
Anthropic co-founder Jack Clark has emphasized that the speed of AI diffusion and the policy response accompanying it will be the critical variables shaping real outcomes.
A public survey of nearly 11,000 individuals conducted alongside the AEI research revealed mixed expectations. People generally acknowledged AI’s potential to increase productivity, but opinions diverged sharply on whether it would create or destroy jobs on net.
Why this matters beyond the research lab
The geographic correlations surfaced by the AEI research add another layer. Regions with higher GDP per capita tend to show higher automation adoption, suggesting that AI’s economic effects won’t distribute evenly.