Anthropic's own model keeps unemployment inside the historical range through a doubled-growth AI scenario

The result comes out of the way the model is built. Every job in it is a bundle of tasks, and AI can help a person do a task better or faster, automate it, leave it alone, or create a new one ^[5]. The task list comes from the Labor Department's O*NET taxonomy ^[6]. Anthropic's worked example is a nurse: AI drafts her discharge instructions, monitors patients remotely and may chart vitals or order the ward's supplies, while it cannot bathe a patient ^[7]. The ending of that example is the part a cost case has to reckon with, because the nurse oversees and accomplishes more and productivity rises ^[8]. Summed across the country, every instance of every task added up to more than $30 trillion of value over the past year in the model's accounting ^[11], and that aggregate is what the scenarios move.

The benign band is wide. It runs from business as usual up to an economy in which AI raises growth to roughly twice the normal rate, and across that entire range unemployment stays within its historical span while wages are flat or rising depending on the industry ^[3]. Adverse effects on knowledge workers' wages and job prospects show up only where growth is faster than anything in economic history, and those same scenarios leave society far wealthier, which converts the problem into one of distribution ^[4]. That is the tradeoff sitting under any plan that books labour cost out of a deployment. The conditions that deliver the saving are the conditions Anthropic describes as without precedent.

Aggregate unemployment inside a historical range is entirely compatible with one function shedding twenty roles, since a historical range accommodates a great deal of churn as displaced work is re-absorbed elsewhere. The harder number for a cost case is the wage one. Flat or rising wages by industry ^[3] mean the price of an hour of labour does not fall, so the saving has to come from buying fewer hours in an economy where the same model shows employment holding up.

What the explorer will not do is tell a buyer which scenario they are in. It takes the reader's own expectations for how capable AI becomes and how extensively it gets used, then shows the 2030 economy those inputs imply ^[9]. Anthropic's own opening position is that it does not yet know how AI will reshape the economy ^[10], and its Economic Index measures current usage rather than the future the explorer explores ^[12]. The report the explorer rests on is dated 2026 ^[2], putting the endpoint four years out, or about four annual planning cycles ^[1]. The capability judgment stays with whoever writes the plan.

Two lines follow for the deck this quarter, and each one demands its own evidence. The throughput claim can cite a vendor's own mechanism, in which the worker's scope expands and output per hour rises ^[8]. The headcount claim is a separate assertion, and the same vendor's model places the conditions for it outside anything in the historical record ^[4]. Keeping them apart is what allows next year's review to test each against what it actually promised.