Anthropic’s Economics Team has put an interactive model online. You set how capable AI gets by 2030, how fast it reaches actual companies, and whether it complements workers or replaces them — then you watch what that does to the US economy. Version 1.0 of the Economic Scenario Explorer.
Three base cases
- Modest: GDP sits 1.6 percent above the baseline path, at $34.1 trillion. Anthropic compares the effect to the internet. Unemployment and wages stay inside historical norms.
- Substantial: Up 8.3 percent, $36.3 trillion. AI handles half of all knowledge work, mostly on its own. Knowledge-worker wages stagnate while other workers gain. Unemployment around 5 percent.
- Extreme: Up 32.4 percent, $44.4 trillion. AI does nearly all knowledge work. Knowledge-worker wages drop by more than ten percent and unemployment climbs past what ordinary recessions produce.
The number that stuck with me is in the last line: the labor share falls from 60 to 45.2 percent. The economy gets much bigger, and a quarter less of what it produces reaches the people doing the work.
How the model works
Jobs are treated as bundles of tasks. Each task is something AI can augment, automate outright, leave alone, or expand. Out of that come labor-market transitions and wage effects, broken out by occupational group, through 2030. Anton Korinek and Chad Jones led the work; Szymon Sacher, Tess Cotter and Peter McCrory built the model. External reviewers included Daron Acemoglu and David Autor — neither of whom anyone would file under AI optimist.
There’s a Morning Consult survey attached: 10,980 people from the general population, 11,130 site visitors. The median in both groups lands on the substantial scenario. Roughly one in ten expects the extreme one.
Anthropic co-founder Jack Clark, on that extreme case: “If you end up with this level of GDP growth, you have moves available to you as a policymaker that are unimaginable today.” His own read is that capability arrives faster than expected and diffusion through the economy slower — which puts him between the usual camps.
What I make of it
A lab building a tool that lets you calculate how much damage its own product might do is unusual. It’s also useful. Anyone arguing about AI and jobs can now put their assumptions on the table instead of trading gut feelings.
But a model is a model. The sliders decide the outcome, and whoever sets them decides what comes out. That’s the point, though: you find out which assumption your optimism rests on. For me it was the augment-or-replace question. One click, and the friendly scenario turns into the ugly one.
Sources: Anthropic: Scenarios for our Economic Future · NPR: A new Anthropic model seeks to test how AI could impact the U.S. economy