The short version: Anthropic published an interactive economic model on September 10, 2026 projecting the US economy to 2030 under three AI scenarios. Multiply each scenario’s GDP by its own labor share and total labor income lands at $20.26 trillion, $20.36 trillion and $20.07 trillion. Between the mildest future and the wildest one sits $10.3 trillion of additional output. What workers collectively earn does not move.
The Econ Scenario Explorer puts both halves of that sum on screen. It shows GDP for each path, and it shows labor’s share of income for each path. It never multiplies them together. Doing it yourself takes about thirty seconds and changes what the tool is telling you.
What is in the Econ Scenario Explorer?
Anthropic’s Economics team built the explorer on the Department of Labor’s O*NET taxonomy, which lists the tasks making up each occupation. Every job is a bundle of tasks, and each task can go one of four ways: AI leaves it alone, AI makes a person faster at it, AI takes it over outright, or AI creates a task that did not exist before. The technical companion is Anthropic Institute Working Paper No. 2026-02, Economic Scenarios for Transformative AI.
Three scenarios come preloaded. In the modest path, AI lands with roughly the force of the internet and 2030 GDP is 1.6 percent above where it would have been, about $34.1 trillion at 2025 prices. The substantial path puts GDP at $36.3 trillion. The extreme path, which Anthropic says would likely require recursively self-improving systems, reaches $44.4 trillion, 32.4 percent above baseline. Labor’s share of income falls in all three: 59.4 percent, then 56.1, then 45.2, with capital taking the rest. Those figures were read off the explorer on September 10, 2026.
Why does the biggest economy pay labor the least?
Here is the arithmetic the tool leaves to you. Labor income is GDP multiplied by labor share:
- Modest: $34.1 trillion times 59.4 percent equals $20.26 trillion
- Substantial: $36.3 trillion times 56.1 percent equals $20.36 trillion
- Extreme: $44.4 trillion times 45.2 percent equals $20.07 trillion
Three wildly different futures, and total labor income varies by less than one and a half percent. Run the same multiplication on capital and the picture inverts: $13.84 trillion, then $15.94 trillion, then $24.33 trillion. Capital income grows by $10.49 trillion between the mildest and wildest scenarios while GDP grows by $10.30 trillion. Capital collects slightly more than the entire difference, because labor’s dollar total actually dips.
This is not a hidden agenda in the model. It is what happens when a share shrinks faster than a pie grows, and it is arguably the most important thing the explorer has to say. It is also the number almost none of the coverage led with, because the headline that writes itself is the unemployment one: in the extreme scenario, 17.9 percent of cognitive workers are out of work and economy-wide unemployment hits 11.9 percent.
Who does better in the scenario everyone calls the disaster?
Look closer at that flat total, because flat is doing a lot of work here. The same $20 trillion is spread across a workforce carrying 11.9 percent unemployment, so the people still working do better per head than the aggregate suggests. The model is specific about who they are.
Anthropic splits the workforce into cognitive occupations, meaning management, professional, sales and office work, and everyone else: nurses, electricians, technicians, construction workers. In the extreme scenario, knowledge-worker pay falls more than 10 percent below the no-AI path. Wages for workers outside knowledge work rise above it, as displaced cognitive workers move into less exposed occupations and bid up the competition for hands-on jobs.
Read plainly, Anthropic’s most alarming scenario is one where your field technician gets more expensive and harder to keep, while the quoting, scheduling and paperwork around that technician gets very cheap. That is close to the opposite of the story most people took from this release, and if you run a service business it is the part worth planning against.
What does this mean if you own a small business?
Take a five-person shop, an owner plus four. That owner sits on both sides of the model’s line. The hours they personally work are labor. The business they own is capital. Across the three scenarios, labor income is flat while capital income rises by about 76 percent.
So the part of an owner’s income that grows is the part attached to the business as an asset, not the part attached to their own hours. That points somewhere more interesting than cutting payroll, which is not where a $10 trillion reallocation gets captured anyway. What compounds is ownership: the customer relationships, the booking history, the pricing data, the process that runs without the owner standing over it. If your business gets faster at quoting because the paperwork got cheap, the value shows up in the asset, and you own the asset.
The mirror image is the risk. If your people are freed from the absorbed work and you use that only to run the same volume with fewer hands, you have converted a growing asset into a one-time saving, in a scenario where the model says skilled hands-on labor is getting scarcer and dearer. Capacity constraints move the moment you fix them, and the binding one in 2030 may well be people, not admin.
Which number in the explorer is an assumption?
Most of them, and the tool is honest about it. The explorer asks you to set five things: what AI will be capable of, how widely it gets adopted, how autonomously it runs, how large the productivity multiplier is, and how long a displaced worker takes to find new work. The autonomy setting, running from “almost none” to “almost all of them”, governs the split between AI taking a task over and AI making a person better at it.
That split is an input, not something the model discovers. Whether AI automates or augments is a dial the user turns before the scenario runs, which means every headline drawn from this tool sits downstream of an assumption somebody chose. Anthropic states its limits plainly, calling the model “a stark simplification of a complex reality” and omitting policy responses, business cycles, financial shocks and robotics. The disclaimer says outcomes “may differ materially”.
That dial is not only a modeling choice at national scale. It is a procurement choice at yours, made every time you decide whether a tool runs on its own or hands work back to a person. We have written before about which jobs are safe to leave unsupervised, and the test has not changed: it depends on whether a wrong step gets caught for free. Near-term economics move faster than any 2030 projection anyway, and the 200 to 1 price gap between flagship and cheap models is a live decision this quarter. The scenarios are a way to think, not a plan. The multiplication is the part worth carrying: growth and your share of it are two different numbers, and only one shows up on the dashboard.
Frequently Asked Questions
What is Anthropic’s Econ Scenario Explorer?
It is an interactive model published by Anthropic’s Economics team in September 2026 that projects US growth, wages, jobs and unemployment through 2030 under different AI assumptions. It is built on the Department of Labor’s O*NET task taxonomy and comes with a technical companion, Anthropic Institute Working Paper No. 2026-02. Users set five parameters covering AI capability, adoption, autonomy, productivity gains and how long displaced workers take to find new work, and the tool returns a scenario. Three presets, labelled modest, substantial and extreme, come loaded by default.
Does the model predict that AI will replace small business staff?
No, and the distinction matters. Whether AI automates a task or makes a person better at it is set by the user through the autonomy parameter before the model runs, so it is an assumption feeding the scenario rather than a finding coming out of it. The model does project displacement concentrated in cognitive occupations, meaning management, professional, sales and office work, under its more aggressive settings. It projects the opposite for hands-on occupations, where wages rise above the no-AI path in the extreme scenario as displaced workers compete for less exposed jobs.
Which scenario does the American public expect?
Anthropic surveyed a representative sample of 10,980 US adults through Morning Consult between August 11 and August 23, 2026, asking them the same five questions the explorer asks. The median respondent’s answers landed close to the substantial scenario, implying GDP roughly 10 percent above the no-AI path by 2030 and overall unemployment near 5 percent. That is slightly more optimistic on growth than Anthropic’s own substantial scenario, which puts GDP 8.3 percent above baseline. The typical American expects real economic gains alongside real disruption to AI-exposed work.
What should a small business owner do with these numbers?
Treat them as a way to frame decisions rather than a forecast to plan against, since Anthropic itself says outcomes may differ materially. The durable takeaway is that income growth in every scenario accrues to capital rather than to hours worked, and for an owner-operator the business itself is the capital. That argues for putting recovered time into things that stay with the business, such as customer relationships, pricing data and processes that run without you, rather than into a one-time payroll saving. It also argues for taking retention of skilled hands-on staff seriously, because the model says those wages go up.
One thing we keep turning over: if the growth really does land on the asset rather than the hours, does that change what you would want your business to be worth in five years, or only how you get there? Tell us how you are thinking about it.
