News · 2026-09-09

Anthropic's economists sketch three AI futures and say workers lose share in all the big ones

Anthropic published a set of three economic scenarios for AI's impact through 2030, and the finding that matters is not the growth figure but the distribution. Across every transformative case the report models, total output rises — and the share of it going to workers rather than to capital falls. In the middle scenario, which Anthropic says is closest to what surveyed experts expect, the economy grows at roughly twice its normal rate while knowledge-worker wages stay essentially flat.

Key facts

• The central case: AI performing half of all knowledge work by 2030, most of it autonomously.
• When: published September 2026 by the Anthropic Economic Index team.
• Who: Anthropic's economics group, with work credited to Korinek and colleagues.
• Primary source: Anthropic's economic scenarios.

The three cases are laid out plainly. In the modest scenario, AI turns out to be about as economically significant as the internet — the report describes "real economic gains, but they're within the historical norm for new technologies." In the substantial scenario, AI is "capable of doing half of all knowledge work by 2030, the majority of it autonomously," and the economy grows at roughly double its usual rate. In the extreme scenario, AI is "more productive than humans at the vast majority of knowledge-work tasks," with annual growth around 15% — fast enough to double the economy every four and a half years.

Anthropic does not pick a winner outright, but it does report where expert opinion clusters: "the typical respondent's answers imply outcomes close to the 'substantial change' scenario."

The distributional finding is what distinguishes this from the usual genre. Growth is positive in all three cases. Wages are not. In the substantial scenario, knowledge-worker wages are "essentially flat" even as output surges. In the extreme scenario they "fall by more than 10% by 2030." The report's own compression of this is the line worth remembering: "the pie will grow, but a larger share might go to capital."

The mechanism is not mysterious. Wages track the scarcity of what a worker supplies. If a machine can do a large fraction of knowledge work autonomously, then the thing knowledge workers were being paid for stops being scarce, while the thing that is scarce — the capital that buys and runs the machines — captures the returns. A useful analogy is what mechanisation did to agricultural labour: national food output rose enormously, food got cheaper for everyone, and the share of that value flowing to farm workers collapsed, because the tractor owner rather than the field hand was holding the scarce asset.

Why it matters is partly the content and largely the author. A company whose product is the thing being modelled has published a document saying that in the futures its own product makes likely, the people who currently do knowledge work do not capture the gains. Labs are not usually in the business of publishing that. It is the sort of analysis that normally arrives from a think tank or a central bank, and it is more useful coming from Anthropic precisely because it cuts against the company's commercial interest in a comfortable story.

It also connects to Anthropic's own measurements. Ground Truth has covered the company's reporting on agent workdays logged per human day, which is the kind of operational figure that turns a scenario into an observation.

The honest caveats are worth stating clearly, and the first is structural. These are scenarios, not forecasts — the report offers three futures and declines to attach probabilities to them, which means it cannot be wrong in the way a prediction can be wrong. That is intellectually respectable and also convenient. The 2030 horizon is close enough for the substantial scenario to require a very fast transition from where deployment actually stands today, and economy-wide productivity statistics have so far been stubbornly unmoved by AI adoption. The survey underpinning "what experts expect" draws on a population with obvious selection effects. And an AI company's economics team modelling AI's economic impact has an incentive structure that readers should keep in view in both directions — toward overstating transformation, and toward appearing sober about it.

What survives all that is the conditional, which is the part that does not depend on getting the timing right: if AI does automate a large share of knowledge work, the gains accrue to whoever owns the systems. That is a claim about how markets distribute returns, not about when the technology arrives — and it is the one policymakers can act on before knowing which scenario turned out right.

Primary source, verified: read the paper →

Ground Truth. (2026, September 9). Anthropic's economists sketch three AI futures and say workers lose share in all the big ones. Ground Truth. https://groundtruth.day/news/anthropics-economists-say-the-pie-grows-and-capital-takes-more-of-it.html

@misc{groundtruth:anthropics-economists-say-the-pie-grows-and-capital-takes-more-of-it,
  title  = {Anthropic's economists sketch three AI futures and say workers lose share in all the big ones},
  author = {{Ground Truth}},
  year   = {2026},
  month  = {sep},
  url    = {https://groundtruth.day/news/anthropics-economists-say-the-pie-grows-and-capital-takes-more-of-it.html}
}

Topics: economics · policy · anthropic · labor · forecasting