Knowledge-work unemployment reaches 18 percent in Anthropic's extreme 2030 scenario

The person writing next year's headcount plan gets two choices in the explorer: how capable AI becomes, and how fast it gets adopted ^[6]. Underneath, every job is a bundle of tasks, and AI can leave a task alone, augment it, automate it, or add new ones; the model scales those effects across the economy ^[7].

Two of the three scenarios leave that plan alone. In the modest case, AI has roughly the impact the internet did and the effect is hard to see in the macroeconomic data ^[8]. In the substantial case, AI does half of all knowledge work by 2030, most of it autonomously, and knowledge-worker wages stay roughly flat while other workers see gains ^[9]. That is the case the write-up in The Next Web says echoes findings that AI has been hitting paychecks rather than payrolls ^[10].

Back out the baseline and the three numbers become comparable. If the modest scenario's $34.1 trillion is 1.6 percent above the no-AI path, that path is about $33.6 trillion in 2030 ^[24]. Substantial sits roughly 8 percent above it, extreme roughly 32 percent ^[25][26]. The typical respondent in Anthropic's August survey expected GDP about 10 percent higher by 2030 with unemployment near 5 percent ^[19], slightly more growth than the substantial scenario produces.

The extreme case is the one that changes staffing, and Anthropic says that path would likely require recursively self-improving systems, adopted quickly ^[14]. Inside it, the 2030 wage outcomes for knowledge workers and everyone else end up more than 40 percentage points apart ^[29]. Anthropic puts annual growth at 15 percent and says that doubles the economy every four and a half years ^[15]. Compounded at 15 percent, doubling takes a bit over five years, and four and a half years of it gets to about 1.9 times ^[27].

The omissions matter for anyone quoting the number in a meeting. The model leaves out possible catastrophic risks and any scenario in which humanity builds hyper-capable robots ^[18]. Gizmodo noted that Anthropic released it a day after the company's own alignment lead, Evan Hubinger, said there was a greater than 10 percent chance AI could kill all humans within a decade, and the economic model does not attempt to price that ^[21].

Anthropic co-founder Jack Clark told NPR the technology would keep improving fast but would spread through the economy more slowly than many expect, and that if growth did reach the extreme range, the extra tax revenue would give policymakers options that are unimaginable today ^[22].

A role-by-role sort does more work here than a scenario-by-scenario one. For most roles the plan is identical under all three cases, and the model adds nothing to it. For the roles that only move in the extreme case, what a planner needs is a written capability trigger with a date to check it, owned by whoever reads model releases. Anthropic's authors say they are not forecasting, and note that AI experts in general expect the technology to spread faster than economists do ^[23].