Anthropic Launches a 2030 U.S. Jobs Simulator

How Does a Nurse's To-Do List Add Up to $30 Trillion?

The model breaks every job in America into "tasks" and assigns AI one of four effects on each: no change, augmentation, replacement, or new task creation. It's a set of sliders, not a yes/no answer.

Rounding on patients, drawing blood, triaging, recording vitals, ordering supplies — that's one nurse's day, and the task list comes from the U.S. Department of Labor's O*NET occupational database. Each task isn't done by her alone; it repeats in every ward, every shift, across the country. Totaling it all up, you're tallying every occupation, every task, every execution across the entire U.S. economy.

AI intersects this task list in four ways: no impact at all, AI helps humans do it faster and better, AI handles it independently, or AI gives rise to entirely new tasks. The original piece offers a historical parallel: handwritten paper charting was still a job 30 years ago; almost nobody does it now. Remote patient monitoring didn't exist 30 years ago. The task list has never been static. The entire model's calculation rests on one ratio: what share of these four effects applies to each type of task.

To make it concrete: AI helps nurses draft discharge orders, monitor patients remotely, and plan shift workflows; it automatically handles charting of vitals and supply orders; it spawns new tasks like "audit whether AI triage was appropriate" and "review AI-proposed care plans." The model multiplies each of these four effects by how many times each task occurs daily nationwide, then aggregates the results into impacts on GDP, the labor market, and wages.

Breaking the Entire U.S. Economy into Billions of Tasks, Calculated One by One by AI

Anthropic's model skips the "job" as the unit of analysis and drills down to tasks: AI decides for each one whether to augment, replace, or create — and GDP, unemployment, and wages emerge from that granular level, not from top-down assumptions.

Conventional economics treats the job as the smallest unit: a person is either employed or unemployed, and wages follow the job. Anthropic instead breaks every occupation into O*NET task lists(the U.S. Department of Labor's standardized occupational task database) — rounding, drawing blood, triaging, recording vitals, ordering supplies — and treats each as an independent task. Stitch together the tasks of all occupations and you get the billions of task instances performed across the U.S. in a single year.

For each task, AI has four possible outcomes: no change, augmenting a human, partial or full automation, or spawning tasks that didn't exist before. The model maps "how much AI capability grows" onto the probabilities of these outcomes, converts "how fast firms adopt and workers re-skill" into time-based speeds, and runs the calculation across billions of tasks. GDP, unemployment, and wages come out as bottom-up sums.

This runs opposite to traditional macro models: those start with a given productivity growth rate(value created per hour of work) and work backward to employment. Anthropic lets employment surface from billions of small task-level decisions — the difference is that layer of granularity.

The Day GDP Doubles, White-Collar Wallets Are Thinnest

The more exaggerated the model's 2030 U.S. GDP projection, the more knowledge-worker wages actually fall.

Anthropic defines the "extreme scenario" as growth "faster than anything ever seen in economic history," with the economy's size doubling within a few years. As the GDP curve maxes out, unemployment also climbs to the model's highest point.

The losers on the wage front are knowledge workers — people who earn their pay through cognitive labor rather than manual or service work, such as programmers, analysts, administrators, and paralegals. Their job tasks are the ones AI can most readily replicate, which is why they're the first to feel the squeeze.

Anthropic writes: total wealth is growing, but the challenge is making sure everyone gets a share. The faster-growing scenarios increasingly resemble the prophecy of The Wealth of Nations — the pie gets bigger, and some people eat first.

Now look at the middle "substantial" scenario: by 2030, AI can independently handle half of knowledge work, but most of it hasn't been adopted yet, and the economy grows at roughly twice the normal rate. This is already the vibe of the "AI golden age" in most sci-fi narratives.

Common sense would say it's hardest on white-collar workers — but the model orders it differently: the modest scenario leaves white-collar wages nearly untouched, the substantial scenario drags them down, and only in the extreme scenario does growth's surplus partially offset the wage suppression.

Anthropic spelled this out in their blog: under "unprecedented growth," wages and employment prospects take a hit; under the modest and substantial scenarios, wages hold steady or rise. The counterintuitive bit lives right here: the stronger the GDP growth, the more compressed the relative share going to knowledge workers.

The reason hides in the task structure — the more capable AI becomes, the higher the share of knowledge tasks it can automate, and new tasks don't appear fast enough to keep pace with displacement.

Why does the extreme scenario squeeze from both sides? It assumes autonomy (the share of work AI completes without human oversight) and adoption rate are both maxed out. When these two sliders move together, white-collar workers' bargaining power per unit of labor gets compressed from both directions.

Push "autonomy" and "adoption rate" to the top and GDP takes off. Leave "re-skilling speed" low and workers can't move fast enough — unemployment climbs alongside GDP. In the wage distribution, knowledge workers become the first group to bleed.

Extreme growth and extreme wage compression are two faces of the same parameter set — provided the growth dividend isn't redistributed fast enough; otherwise, the wage compression in this scenario would be partially offset.

What Anthropic Wants Isn't Answers — It's a Table Full of Question Marks

The questions this model forces into the open are worth more than the conclusions it provides.

The extreme-scenario picture is deeply anomalous: total social wealth far exceeds any historical precedent, but a large slice of knowledge work disappears. Anthropic's long-form piece is explicit — in this kind of situation, the core question is no longer a choice between "growth or unemployment," but a structural problem of "everyone richer, but distribution sharply uneven."

All numbers in the model come from Anthropic's internal assumptions and extrapolations; no third party has independently replicated the scenario design to date. It's more prudent to read it as a reference framework than as a forecast. Anthropic itself stresses "we don't know how AI will reshape the economy" — handing the uncertainty straight to the reader is unusual among AI vendors.

An observable signal: whether the monthly current-usage data from the Economic Index moves in the same direction as the "adoption speed" slider in the scenario model is the most direct anchor for judging whether this reference framework holds up — provided the data stays public and methodology stays consistent. If an independent academic team replicates the methodology from Korinek et al., 2026 and publishes the result within the next year, the framework's reference value rises; otherwise, it remains just one vendor's word.

Drag the Sliders First, Then Decide What to Watch

This tool won't hand you a 2030 answer. It hands you sliders. Open the page, read the scenario overview, then start dragging — three numbers shift with every adjustment.

Anthropic's Scenarios for Our Economic Future page has five sliders: rate of AI capability growth, rate of industry adoption, share of tasks automated, rate of new task creation, and rate of worker re-skilling. Drag any of them and the right side updates in real time with 2030 projections for GDP, unemployment, and knowledge-worker wages.

Try pushing "share of tasks automated" and "capability growth" all the way right, then compare that to moving only "capability growth." The gap between your two 2030s is orders of magnitude. That's the fastest way to see the distance between the extreme and modest scenarios.

The gaps aren't minor. In the Modest scenario, AI's impact is like the internet's — barely visible in macro data. In the Substantial scenario, AI handles half of knowledge work by 2030, mostly autonomously, and economic growth doubles. The Extreme scenario requires "recursively self-improving AI systems" to hold. These three presets are the model's built-in anchors; whatever you drag into existence falls somewhere between them. Anthropic isn't picking a winner — it's laying out the possibilities and letting you weigh them.

What you can't do here is verify the model. The source rests on a technical report by Korinek et al., 2026, but the full report, third-party replication, and benchmarking methodology aren't linked. So what you can do right now isn't verification — it's observation. Two timestamps matter: when Anthropic ships an update, and when someone backtests the model against real economic data. Those are the markers for deciding whether to trust it.

One distinction shows up immediately: the same 2030 GDP number can sit next to wages and unemployment pointing in opposite directions. In the extreme scenario, society is richer, but knowledge-worker wages and job prospects deteriorate — the source frames the challenge as "how to make the gains broadly shared." When you're dragging, the row to watch isn't GDP. It's "knowledge-worker wages."