In the most extreme of Anthropic's three scenarios, the US economy grows at 15.4% a year by 2030 — a rate no large economy has ever sustained. In that same column, wages for cognitive workers are 11.5% lower than they would have been without AI, and nearly one in five of them is unemployed.
Those two numbers sit side by side in a model an AI company published about its own product. That combination is the story.
Anthropic’s Economics team is sharing a new model of how AI might affect economic growth, jobs, wages, and more by 2030.
— Anthropic (@AnthropicAI) September 9, 2026
Explore the scenarios, tell us what you think will happen, and see how your answers compare to more than 10,000 Americans. https://t.co/AvQlEZNxR0
What Anthropic actually built
The model comes from Anthropic's Economics team: a technical paper, Economic Scenarios for Transformative AI (Korinek et al., Anthropic Institute Working Paper 2026-02), plus an interactive explorer where you can turn the dials yourself.
The framework is simple enough to explain over coffee. Every job is a bundle of tasks. A nurse checks on patients, draws blood, charts vitals, orders supplies. AI can do four things to any task in that bundle: nothing, help a human do it faster, do it entirely, or create a brand-new task that didn't exist before. Add up every instance of every task performed across the US and you get an economy that produced over $30 trillion of value last year.
The model converts a handful of parameters — what share of tasks AI touches, how widely it's actually deployed, how much faster it makes them, how often it replaces the human rather than assisting them, and whether new work appears to replace what's automated — into paths for GDP, wages, the labour share and unemployment out to 2030.
One framing matters before any number does: these are not forecasts. The authors attach no probabilities. The point is to make an argument about AI's economic impact into an argument about specific dials rather than vibes.
Three futures, one table
| In 2030 | Modest | Substantial | Extreme |
|---|---|---|---|
| GDP vs no-AI path | +1.6% | +8.3% | +32.4% |
| GDP growth (2% without AI) | 2.4% | 5.4% | 15.4% |
| Cognitive wages vs no-AI | +0.4% | −0.3% | −11.5% |
| Wages, all other jobs | +1.1% | +5.9% | +33.6% |
| Labour share of income | 59.4% | 56.1% | 45.2% |
| Unemployment, all workers | 3.9% | 4.6% | 11.9% |
Modest is AI as a normal technology — roughly the internet. Only 4% of the economy's tasks are touched by 2030 and the macro data barely notices.
Substantial is AI as a pivotal technology: around 12% of all tasks affected, productivity on them up 57%, three-quarters of that work automated rather than assisted. Growth hits 5.4% a year — for scale, the fastest year of the entire dot-com boom was 4.7% in 1999.
Extreme assumes AI handles roughly half of today's cognitive tasks, more than doubles productivity where it lands, automates 90% of what it touches, and creates no new cognitive work to replace what it takes. Sustained, that growth doubles per-capita income every five years; the historical rate is every thirty-five.
Read that table left to right. Wages for people outside cognitive work climb steadily. Wages inside it go the other way. The gap between those two rows is the whole argument.
Everything grows except the knowledge worker
In the extreme scenario capital income is 81% above its no-AI path while the cognitive wage bill is 31% below it. (Chart: BougainWell · Data: Anthropic Institute WP 2026-02)
Here is the finding that deserves more attention than the growth rate. In the two serious scenarios, the pie gets much bigger — and labour's slice gets smaller.
The labour share is the fraction of every dollar the economy produces that ends up in someone's paycheque rather than as a return on capital. Today it's about 60 cents. In the substantial scenario it falls to 56 by 2030 — a four-point move in four years, very nearly the entire decline in the US labour share across the four decades after 1980, compressed into a single presidential term. In the extreme scenario it falls to 45.
The mechanism isn't mysterious. If AI makes capital useful for work that previously needed a person, demand for capital rises and more of the output flows to whoever owns it, while displaced cognitive workers spend time finding something else — which pushes down the wage they can command on the way.
There's one number here that almost nobody quotes, and it reframes everything. I'll come back to it.
The public is neither doomer nor denier
The median American's answers imply GDP 8.6% above the no-AI path — almost exactly the substantial scenario. (Chart: BougainWell · Data: Anthropic Institute WP 2026-02)
Anthropic surveyed 10,980 US adults through Morning Consult between 11 and 23 August 2026, asking five questions: what AI can do, how much people use it, how much it does alone, how much faster it makes them, and how long it takes to find a new job. Each person's answers were then run through the model.
The median American lands at GDP 8.6% above the no-AI path — essentially the substantial scenario — with unemployment around 4.6%. Not a boom, not a catastrophe.
The interesting part is why. The median respondent expects a genuinely capable AI, one handling six of eight tasks by 2030. What holds their number down isn't scepticism about the technology but about deployment: they expect it used on only two-fifths of the work it could do, and assisting rather than replacing about half the time.
That's a sophisticated intuition, and the model agrees — capability and diffusion are separate dials, and the second is where most of the outcome lives. People are not converging, though: the middle half of respondents span +3% to +19% GDP, and roughly 30% think AI saves no time at all on a task it's suited to while 49% think it at least halves the time.
The number nobody quotes
Here's the open loop, paid off. In the extreme scenario, total labour income is 0.5% above its no-AI path — flat, while GDP is a third larger.
That single figure says the entire gain from the most transformative economic event ever modelled accrues to capital, whose income is 81% above its no-AI path. Workers as a class end up neither better nor worse off in aggregate; they are simply not participants in the growth.
The paper works out the fix: a transfer of about 9% of GDP — roughly Social Security and Medicare combined — would hold cognitive workers at their no-AI income and still leave everyone else more than 20% ahead. The arithmetic works. The authors are candid that transfers on that scale have no precedent, and that past displacements mostly went uncompensated.
Worth noticing, too: a lab published a model in which heavy adoption of its own product cuts the knowledge-work wage bill by 31%, then built an explorer so the public could reach that conclusion themselves. Publishing your worst case early is also how you shape a debate you know is coming, and the framing that emerges — growth is not the problem, distribution is — is a convenient one for a lab. Both can be true; the parameters are public, so anyone who disagrees can now argue about a dial instead of a mood.
What to Take Away
- Capability and diffusion are different dials, and diffusion decides the outcome. The median American expects a very capable AI and still lands on a moderate economy, purely because they expect limited deployment. Ask which of the two someone is actually forecasting.
- Watch the labour share, not the growth rate. GDP growth can look spectacular while the wage bill doesn't move.
- A number worth remembering: +0.5%. Total labour income in the extreme scenario, against GDP a third higher. Growth and shared prosperity are not the same event.
- The pain is concentrated, the gains are diffuse. Wages outside cognitive work rise 33.6% while cognitive wages fall 11.5%. "AI and jobs" is the wrong unit of analysis; occupation is the right one.
- Scenarios are not predictions. No probabilities are attached. The value is making disagreements precise, not telling you what happens.
Charts: BougainWell, built from Anthropic's published technical report and scenario explorer. This article is for general information only and is not investment advice.
Sources
All analysis and opinions in this article are BougainWell's own.
