AI adds 8% to GDP and flatlines knowledge-worker wages. Anthropic's economists are saying this out loud.

Everyone in knowledge work has been doing the same math lately. You've seen the headlines. You have colleagues whose jobs have changed, or disappeared. And you're wondering, honestly, where your work fits in what comes next.

Anthropic's economists put some numbers on that question. In September, they published the Econ Scenario Explorer, an interactive model that maps three futures for the US economy through 2030. The underlying report is called Economic Scenarios for Transformative AI, authored by Anton Korinek, Chad Jones, Szymon Sacher, Tess Cotter, and Peter McCrory. Four prominent economists — Daron Acemoglu, David Autor, Ben Jones, and Pascual Restrepo — reviewed it. None of them endorsed the conclusions. That detail matters and I'll come back to it.

Here's what the model actually found.

Three futures, one uncomfortable pattern

The model runs three scenarios:

Modest: AI affects the economy the way the internet did. Real gains, gradual, within historical norms. AI can do some tasks but humans still drive most output.

Substantial: AI can do half of knowledge work by 2030. Most of that work is still done by people, but the balance shifts meaningfully. This is roughly where the public survey respondents landed.

Extreme: AI outperforms humans on most knowledge tasks. This likely requires recursively self-improving AI and fast adoption. GDP grows at about 15% per year — which would double the economy in about four and a half years.

The pattern across all three scenarios is the same: GDP grows. Capital captures a larger share of that output — meaning investors, owners, and companies get more, relative to workers. And knowledge-worker wages (coders, writers, analysts, call center agents) are flat or down even as the broader economy improves.

Average wages rise in the Modest and Substantial scenarios. That's not a contradiction. It means lower-exposure workers — electricians, nurses, tradespeople — benefit as demand for their labor increases while AI drives demand for knowledge work down. If you are in knowledge work, you're in the flat column.

Who says Substantial is most likely

Anthropic surveyed 10,980 members of the general public in August 2026 alongside the model release, using Morning Consult. The typical response pointed to something close to Substantial: about 10% above baseline GDP by 2030, unemployment near 5%. About 10% of respondents held views consistent with the Extreme scenario.

The page is careful about what this means. It explicitly excludes policy responses, business cycles, the demand effects from the data center buildout, financial disruptions, and catastrophic risks. The reviewers flagged that the model can't track individual workers and may underestimate AI's effect on technological progress. The authors call it "a tool for thinking through scenarios, not a forecast."

That's honest. It's also how you should read the numbers.

Source spread

• Anthropic — Econ Scenario Explorer [hype]: The model itself, published by an AI company that has obvious interest in characterizing AI's economic impact as large but manageable. The limitations section is unusually frank for a corporate publication. The reviewer names (Acemoglu, Autor) carry weight; their non-endorsement is worth noting.

What's real:

• GDP grows in every scenario. That much is probably right — technological change historically expands economic output.
• The distribution finding is robust to scenario choice. Even in Modest, capital takes a larger share. That's not a fluke; it's a structural prediction about where economic gains flow.
• The public pointing to Substantial is a revealed expectation worth taking seriously. These aren't researchers; they're the people the transition will affect.
• The limitations section on the model's own page is unusually honest. The authors say what it can't do.

What deserves a side-eye:

• This was published by Anthropic. They're not neutral. The framing — "the main challenge is not achieving economic growth, but making sure benefits are broadly shared" — is true and also convenient if you want to shift conversation from "should this technology exist" to "how do we distribute proceeds."
• The Extreme scenario requires recursively self-improving AI and fast adoption. Neither is guaranteed. The model presents it without assigning probabilities to any scenario.
• The gap between "average wages rise" and "your wages rise" is doing a lot of work here. If you're in knowledge work, the Substantial scenario is more Modest than it sounds.

What to do about it

A research model from September isn't a reason to panic. But it's not nothing either.

• If you're in knowledge work, this is the scenario worth planning for. Not "AI will take my job" — the Substantial scenario keeps humans doing most tasks. But wage growth in knowledge work is likely to lag other sectors. That's a career-positioning consideration, not a reason to spiral.
• The skills that remain valuable are the ones AI can augment but not fully replace: judgment in context, physical work, work that requires legal accountability, relationships with clients or patients who want to talk to a human. Electricians and nurses did better than analysts and coders in every scenario. Worth noting.
• This is a reason to care about tax and transfer policy. The distribution problem named in this model is solvable — it's been solved before, partially, in other technology transitions. The mechanism is policy, not individual career pivoting alone. Following what Congress is or isn't doing on this is not just politics; it's material to your financial future.
• Use the interactive model. The Anthropic Econ Scenario Explorer lets you adjust the assumptions and see how the numbers change. It's a useful tool for calibrating your own view — not a forecast, but a structured way to think about the range.
• Don't treat this as a forecast. The authors say clearly it's for thinking, not predicting. The actual trajectory depends on policy, adoption rate, and which version of AI development actually arrives. All three are genuinely uncertain.

Further reading

• Anthropic Econ Scenario Explorer — the interactive model and report; the limitations section is worth reading in full