For many years, citing the research, acknowledging the uncertainty, and landing somewhere comforting was the typical procedure at any economics conference with a panel on AI and jobs. Indeed, some employment would be lost. Indeed, new ones would be produced. Indeed, this had already occurred with the printing press, industrial automation, and computing, and the labor market had reached equilibrium. Although the “augment, not replace” paradigm was not inherently naive, it did include an implicit pacing assumption that the change would occur slowly enough for people, institutions, and social policy to adjust. A growing percentage of those who created it have discreetly abandoned that assumption.
The most notable deviation occurred in May 2025 when Dario Amodei, CEO of Anthropic, one of the three or four businesses whose AI systems are actively driving the present revolution, sat down with Axios and expressed what most individuals in his position had been cautious not to say outright. He cautioned that within five years, AI might remove half of all entry-level white-collar occupations. As a direct consequence, unemployment might rise to 10–20%.
According to him, businesses would initially discreetly stop recruiting people before replacing them with AI as soon as it became economically feasible. He predicted that this would happen practically immediately. He continued by saying that the public wouldn’t comprehend its scope until it was already occurring. Amodei is not a pundit. The technology is being developed by him. This is the aspect of the statement that makes it more difficult to ignore than most warnings of this type.
Important Information
| Field | Details |
|---|---|
| Key Warning Figure | Dario Amodei — CEO, Anthropic; told Axios in May 2025 that AI could eliminate 50% of entry-level white-collar jobs within five years, pushing unemployment to 10–20% |
| Amodei’s Assessment | “We, as the producers of this technology, have a duty and an obligation to be honest about what is coming. Most lawmakers are unaware that this is about to happen.” — told CNN’s Anderson Cooper |
| Goldman Sachs Projection | Approximately 300 million full-time jobs globally could be affected by generative AI; Goldman’s base case projects transition over ~10 years with 6–7% U.S. workforce displacement — but warns front-loaded adoption would cause significantly larger near-term disruption |
| WEF Future of Jobs Report 2025 | 92 million roles displaced globally by 2030; 170 million new roles created — net gain of 78 million, but concentrated disruption in entry-level and routine white-collar work |
| McKinsey Estimate | Generative AI could automate 60–70% of current work activities before 2030, affecting the equivalent of 300 million full-time jobs worldwide |
| Yale Budget Lab (Sept 2025) | Used Anthropic’s actual AI usage data — found AI is suppressing hiring more than destroying existing jobs in the near term; entry-level positions being eliminated before workers fill them |
| Anthropic Researcher Warning | Sholto Douglas, Anthropic researcher, told AI podcaster Dwarkesh Patel: “There is almost guaranteed to be a drop in white-collar workers at some point in the next five years, even if current AI progress stalls” |
| The Transition Problem | Goldman Sachs 2026 senior economist Ronnie Walker: still no meaningful relationship found between AI adoption and productivity at the economy-wide level; some workers report AI making them less productive |
| Entry-Level Career Ladder Risk | Companies slowing hiring in anticipation of AI — class of 2025 hit a significantly different hiring market than prior years; junior roles disappearing before new graduates can enter them |
The calculations had already been performed at a different scale by Goldman Sachs. According to their research, which was updated through 2025, generative AI may have an impact on the equivalent of 300 million full-time employment globally. It wouldn’t necessarily abolish these occupations, but it might drastically change them to the point that employees would need to perform radically different tasks in order to stay employed.
According to their base case, this would occur over about 10 years, displacing six to seven percent of the U.S. workforce during the transition period—an increase in unemployment that the company deemed reasonable. However, their own economists recognized the crucial warning: the short-term effects on unemployment are far greater if adoption is front-loaded as opposed to dispersed over ten years. In other words, the optimistic option is the slow-transition scenario. Amodei is explaining the fast-transition situation.
The entry-level pipeline is the aspect of this that labor economists are taking the most seriously, not the total number of jobs lost. According to a September 2025 Yale Budget Lab research using Anthropic’s own AI usage data, AI is currently suppressing hiring more than it is eliminating current jobs. Businesses aren’t terminating a lot of employees just yet; instead, they aren’t filling the junior positions that they would have created in the past.
A generation of workers who cannot find entry-level jobs, cannot acquire the institutional knowledge and skills those positions have always provided, and arrive in their late twenties without the career capital that the professional ladder was intended to provide is the downstream effect of that, which is less obvious than mass layoffs but potentially just as harmful. The ladder is still breaking even if it is breaking more softly than loudly.
An additional degree of uncertainty is created by the productivity conundrum that has surfaced surrounding the use of AI. The “productivity paradox” that emerged in the early decades of personal computing, when the machines were obviously being purchased and deployed but their impact on output took years to appear in data, is echoed by Goldman’s senior economist Ronnie Walker’s observation in early 2026 that there is still no meaningful relationship between AI adoption and productivity at the economy-wide level.
According to some employees, AI is actually making them spend more time on some jobs rather than less, adding coordination and quality-checking chores to formerly direct operations. This may just indicate that the displacement is occurring before the productivity improvements have completely manifested, which would be the worst possible sequencing for workers, but it does not disprove the displacement argument.
There is a sense that the public debate is finally catching up to what a smaller group of people have been saying in private for longer than the headlines suggest. This is evident when one watches the conversation change in real time, from careful economic modeling about net job creation to CEOs of AI companies using phrases like “stop sugar-coating it” in national television interviews. The most pressing question at this time is not so much the quantity as it is whether the gap between private awareness and public readiness closes quickly enough to have any real impact.
