Anthropic Modeled Its Own Industry's Effect on Knowledge Workers. Even the Best Case Barely Moves Their Wages.

Anthropic, the company that builds the Claude models, published an economic model of its own industry's effect on jobs and wages by 2030. The report is called "Scenarios for our Economic Future," built on a technical paper, Economic Scenarios for Transformative AI (Korinek et al., 2026). It lays out three paths, and none of them are simple good news for the people doing knowledge work today.

What happens in the modest scenario?

In the modest scenario, AI's impact on the economy looks something like the internet's did: useful, gradual, spread out over years. US GDP in 2030 reaches $34.1 trillion, 1.6% above what it would be without AI. Knowledge workers do see a gain here, though a thin one: the cognitive wage ends up 0.4% above where it would have been with no AI at all, against 0.7% for the average wage. Labor's share of total output edges down from 60.0% today to 59.4%.

What happens in the substantial scenario?

Anthropic writes that AI is capable of doing half of all knowledge work by 2030, the majority of it without a person driving each task, "but it's not adopted for all of that work: most knowledge work tasks are still done without AI." What AI touches is far smaller. The technical paper puts that at 12% of the tasks in the economy by 2030, roughly 20% of the tasks cognitive workers were doing in 2025, and says those tasks "have been impacted by AI." In the paper's split, three quarters of those affected tasks move to the machine and one quarter stay with a worker who now has AI help. That works out to about 9% of the economy's tasks done by machine and about 3% done by a person with help. Capability and uptake are two different numbers, and only one of them is half.

On those assumptions, GDP reaches $36.3 trillion, 8.3% above baseline. Knowledge-worker wages come in 0.3% below their no-AI path, which is what Anthropic means by essentially flat, while wages elsewhere in the economy run 5.9% higher. Labor's share of output slips to 56.1%, as capital's share picks up 3.9 points from today's 40.0%.

What happens in the extreme scenario?

In the extreme scenario, AI outperforms humans at most knowledge-work tasks and keeps improving on its own. GDP reaches $44.4 trillion, 32.4% above baseline. Knowledge-worker unemployment climbs past anything seen in a past US recession. Wages for the workers whose tasks AI took over fall by more than 10%. Labor's share of output drops to 45.2%, a 14.8-point shift toward capital.

Anthropic's own extreme-scenario finding states the real risk directly: "The main challenge is not achieving economic growth, but making sure the benefits are broadly shared and the costs aren't unequally dispersed."

Which single variable decides the outcome?

Read the three scenarios side by side and a single variable decides which side of that split a knowledge worker lands on. Every scenario assumes AI touches their work. The difference is how much of the task AI takes over outright, versus how much runs alongside a person who still makes the call. The modest scenario, where the cognitive wage edges up, leans toward a person keeping that call. The extreme scenario, where wages fall by double digits, is the one where AI runs the task "autonomously," Anthropic's own word.

Augmentation, in this sense, means AI does the gathering, sorting, and first-draft work while a person still owns the judgment at each decision point. Replacement means the task runs end to end with no one making that call.

Where does Foragentis sit on that fork?

That fork is more than an abstraction to us. It is the design choice Foragentis makes explicit in every ForIntel report. Knowledge work, exactly the category Anthropic is modeling, is what we build: competitive analysis, market-entry dossiers, industry research. A ForIntel report never lets AI make the final call alone. Every claim resolves to a checkable source. A human sits at the decision points that determine what is published. That is the augmentation side of Anthropic's own fork, applied to the category of work its report says is most at risk.

We have written before about what happens when the human call drops out of research work: all four of the Big Four published AI-hallucinated reports in under a year, and each one traced back to a claim no person had checked against a real source.

Why read the scenarios yourself?

The company building the models just published the numbers behind what happens to knowledge workers when that choice goes the other way. The scenarios are worth reading directly, as a gut check on how your own team is using AI in research or analysis work, and on whether a person still makes the call: foragentis.com.

FAQ

What does Anthropic's "Scenarios for our Economic Future" report say?

Anthropic, the company that builds the Claude models, published an economic model of its own industry's effect on jobs and wages by 2030. Built on the technical paper "Economic Scenarios for Transformative AI" (Korinek et al., 2026), it lays out three paths, modest, substantial, and extreme, and none of them is simple good news for the people doing knowledge work today.

What happens to knowledge-worker wages in each Anthropic scenario?

In the modest scenario, the cognitive wage ends 2030 just 0.4% above where it would have been without AI, and labor's share of total output edges down from 60.0% today to 59.4%. In the substantial scenario, where AI is capable of doing half of all knowledge work without being adopted for all of it, knowledge-worker wages land 0.3% below their no-AI path, which Anthropic describes as essentially flat, and labor's share slips to 56.1%. In the extreme scenario, wages for the workers whose tasks AI took over fall by more than 10%, knowledge-worker unemployment climbs past anything seen in a past US recession, and labor's share of output drops to 45.2%.

How much does US GDP grow in Anthropic's three scenarios?

By 2030, US GDP reaches $34.1 trillion in the modest scenario (1.6% above what it would be without AI), $36.3 trillion in the substantial scenario (8.3% above baseline), and $44.4 trillion in the extreme scenario (32.4% above baseline). Anthropic's own extreme-scenario finding states the risk directly: "The main challenge is not achieving economic growth, but making sure the benefits are broadly shared and the costs aren't unequally dispersed."

What single variable separates the good and bad outcomes for knowledge workers?

Every scenario assumes AI touches knowledge work. The difference is how much of each task AI takes over outright, versus how much runs alongside a person who still makes the call. The modest scenario, where the cognitive wage edges up, leans toward a person keeping that call. The extreme scenario, where wages fall by double digits, is the one where AI runs the task "autonomously," in Anthropic's own word.

How does ForIntel apply this fork in practice?

ForIntel produces exactly the category of work Anthropic is modeling: competitive analysis, market-entry dossiers, industry research. A ForIntel report never lets AI make the final call alone. Every claim resolves to a checkable source, and a human sits at the decision points that determine what is published. That is the augmentation side of Anthropic's own fork, applied to the work its report says is most at risk.

ForIntel is a research-first business-intelligence product from Foragentis. Every quantitative claim in a ForIntel brief traces to source data and carries the confidence the evidence actually supports. Before publication, a second, independent read of the same data tries to break the finding without seeing how it was reached, and whatever it comes back with is printed in the brief next to the finding. Where two measures of the same thing disagree, the brief reports the disagreement instead of keeping the tidier number. To see a sample brief or commission a read, reach the ForIntel desk at forintel@foragentis.com.