Anthropic Modeled the 2030 Economy Three Ways. The Richer the Country Gets, the Worse It Gets for Knowledge Workers

Anthropic published a scenario explorer and technical report modeling how transformative AI could reshape the US economy by 2030, projecting GDP gains of 1.6% to 32.4% across modest, substantial, and extreme scenarios. The research finds that the fastest-growing scenarios impose the steepest costs on knowledge workers, with the extreme case showing cognitive-worker wages falling more than 10% and unemployment near one in five. The model excludes physical robotics and focuses on cognitive work, mapping task-level automation and augmentation onto GDP, wages, and labor's share of income.

Anthropic just published a model of what AI could do to the US economy by 2030, and the headline is uncomfortable: in the scenarios where the economy grows the fastest, knowledge workers are the ones who pay for it. The team behind the Anthropic Economic Index, the same research program that tracks how people actually use Claude at work, built a scenario explorer based on a technical report called Economic Scenarios for Transformative AI https://www.anthropic.com/institute/econ-scenarios by Anton Korinek, Charles I. Jones, Szymon Sacher, Tess Cotter, and Peter McCrory. It went live on September 10, 2026. It does not predict the future. It maps three futures, and lets you plug in your own assumptions about AI capability and adoption to see which economy your beliefs imply. Full disclosure: this article is a researched breakdown of the published model and report, not a prediction of my own. Every number below comes from the scenario explorer, the technical report, or coverage of both. If you work in software, analytics, writing, or any other desk job that produces output from a keyboard, the findings deserve your attention. The model starts from a simple idea: a job is a bundle of tasks. A nurse checks on patients, draws blood, triages arrivals, charts vitals, and orders supplies. AI can augment a task, automate it, ignore it, or create new tasks nobody did before. Task bundles have always shifted. Hardly anyone hand-writes paper charts anymore, and thirty years ago nobody monitored patients remotely. Add up every task performed in the US, by humans and machines, and you get an economy that produced over $30 trillion in the past year. The model maps different assumptions about AI capability and adoption onto that task structure and projects GDP, wages, unemployment, and how income splits between workers and capital, out to 2030. One limit worth knowing upfront: the model excludes hyper-capable physical robots. This is a model about cognitive work. Modest. AI's impact is roughly what the internet's was. Real gains, but within the historical norm for new technologies, arriving gradually. GDP in 2030 comes in about 1.6% higher than it would without AI, at $34.1 trillion. Unemployment and wages barely move. If you squint at macroeconomic data, you cannot see AI in it. Substantial. AI can do half of all knowledge work by 2030, mostly autonomously, but adoption is partial: most knowledge tasks still get done without it. GDP runs at roughly twice the normal growth rate, landing 8.3% higher at $36.3 trillion. Wages for knowledge workers stay essentially flat while other workers see gains. Extreme. AI beats humans at the vast majority of knowledge-work tasks, does nearly all of them autonomously, and creates essentially no new knowledge tasks for people. This scenario likely requires recursively self-improving AI adopted quickly. Annual GDP growth reaches 15%, doubling the economy every 4.5 years. GDP lands 32.4% higher, at $44.4 trillion. And knowledge workers face wages falling more than 10% and unemployment beyond typical recession levels, close to one in five cognitive workers out of a job. Read those three together and the pattern jumps out: the scenarios where society gets richest are the ones where knowledge workers individually get hurt most. There is no scenario in this model where AI shrinks the economy. The smallest case adds $600 billion or so by 2030. The largest adds over $10 trillion. The question the model keeps returning to is who receives that growth. Job markets always churn. People lose jobs and find new ones, and in normal times the process is painful but works at the macro level. In the substantial and extreme scenarios, the model shows something heavier: knowledge workers facing so much automation that many have to switch occupations entirely. The report's own example is coders and call center agents moving toward electrician and nurse, occupations that are less exposed to AI. Switching occupations is slow. It often means retraining, relocation, and months between jobs. The more switching a scenario requires, the more people are stuck in between. In the extreme scenario, unemployment in knowledge work rises sharply while unemployment in other occupations actually falls. The total number lands above typical recessionary levels. Here is the detail worth sitting with: in the substantial scenario, which most survey respondents consider the realistic one, the overall unemployment rate only rises to around 5%. Historically unremarkable. But that average hides the split. The pain concentrates in one segment of the workforce while the rest of the economy hums along. Across all three scenarios, average wages rise. The distribution is the problem. The mechanism runs both directions. Less demand for human knowledge work pushes knowledge-worker wages down. Meanwhile, AI-driven productivity elsewhere raises demand for manual work: cheaper design and permitting means more construction projects, which pushes construction wages up. So the substantial scenario leaves knowledge-worker wages flat while other workers gain. The extreme scenario cuts knowledge-worker pay by more than 10% by 2030, even as the country as a whole becomes wealthier than it has ever been. Today, of every dollar the US economy produces, roughly 60 cents goes to workers and 40 cents to capital. The model projects how that split shifts: That last number is the most striking in the whole report. In the extreme scenario, workers collectively receive barely more total labor income in 2030 than they would without AI, despite the economy being a third larger. The gains flow to whoever owns the technology. Anthropic states the conclusion plainly: in that world, the challenge is not achieving growth. It is making sure the benefits are broadly shared. Anthropic surveyed US adults alongside the model, and expectations vary widely. The typical respondent's assumptions produce an economy close to the substantial scenario: GDP about 10% higher by 2030 and overall unemployment around 5%. Roughly 10% of respondents hold views in line with the extreme scenario. The researchers themselves are more divided on timing. Anton Korinek, who leads Anthropic's transformative-AI economics work, has emphasized that impact depends on adoption, because a capable AI that nobody uses has no economic effect. Daron Acemoglu, one of the economists who reviewed the report, told NPR he expects the technology to keep improving fast but to spread through the economy more slowly than many predict. Even so, if growth ever did reach the extreme range, he noted the extra tax revenue would give policymakers options that are unimaginable today. It is worth noting who reviewed the model: Daron Acemoglu, David Autor, Pascual Restrepo, David Romer, and other leading labor economists. This is not a vendor marketing deck. It is a serious attempt to build a shared framework, published with its assumptions open for attack. The model is not destiny, and Anthropic says so explicitly: the 2030 economy depends on what AI can do, how companies adopt it, and how the gains are distributed. But three practical takeaways survive even under the modest scenario. The bundle shifts before the job dies. In every scenario, tasks leave and arrive. The workers who fare best across all three futures are the ones whose task bundles include things AI cannot do, and who can move new tasks into their bundle faster than automation removes old ones. Occupation switching is the risk, not unemployment. The model's sharpest costs come from transition: time between jobs, retraining, moving to less-exposed occupations. The realistic hedge is not panic. It is shortening your own switching distance, building skills adjacent to both your current occupation and the less-exposed ones the model expects demand to flow into. Watch the labor share, not just the headlines. GDP records will look great in most of these futures. The number that decides how the decade feels for desk workers is what fraction of each dollar flows to wages versus capital. That is the number to track as this plays out. The full model is interactive, so you can plug in your own assumptions about AI capability and adoption and see the economy they imply, alongside everyone else's predictions. If the survey distribution is any guide, most of us already believe something close to the scenario where our own wages stagnate. The model just makes the arithmetic explicit. Sources: