The anxiety isn't abstract anymore. Employee concerns about AI-driven job loss have jumped from 28% in 2024 to 40% in 2026, according to consultancy Mercer. Entry-level hiring in roles exposed to AI automation has dropped 13% since large language models began proliferating in earnest. And CNBC has spent the past two months running a near-continuous thread of coverage on the labour question, each piece a little more urgent than the last. The question is no longer whether AI is changing the labour market — it visibly is — but how much, how fast, and for whom.
The most alarming forecast came from ServiceNow CEO Bill McDermott, who told CNBC this month that unemployment for new college graduates "could easily go into the mid-30s in the next couple of years" as AI agents displace the entry-level white-collar work that has traditionally served as the first rung on the career ladder. That's a stark number, and McDermott's framing points to a structural issue that goes beyond individual job losses: it's not just that AI is taking specific roles, it's that it may be collapsing the pipeline through which people build skills and move up.
Jack Dorsey's decision to cut a large portion of his company's workforce was the case study economists were asked to weigh in on most heavily. The consensus was cautious: Dorsey's cuts reflect genuine AI capability gains, but they also reflect a CEO with strong convictions about lean organisations who may have acted faster than the technology strictly required. That distinction matters. Some of what's being attributed to AI displacement right now is actually managerial choice dressed in AI clothing — companies using the moment as cover for restructuring they wanted to do anyway.
Anthropic's own research on labour market impacts — published quietly on its website — takes a measured tone. The company finds that AI is already having measurable effects on labour demand in certain knowledge-work categories, particularly roles involving routine research, data synthesis, and content production. But the research also notes significant variation by sector and role type, and cautions against linear extrapolation. The history of technology and labour is full of predicted mass displacement events that arrived more slowly, more unevenly, and with more job creation in adjacent roles than the predictions allowed for.
The counternarrative getting the most traction is skilled trades. Plumbers, electricians, HVAC technicians, construction workers — roles that require physical presence, dexterous problem-solving, and contextual judgement in variable environments — are holding up well, and in many markets are commanding higher wages than equivalent white-collar roles. The phrase "AI-proof" is probably too strong; robots are getting better at physical tasks too. But the near-term picture does suggest that the labour market disruption is highly uneven, concentrated in certain kinds of office work rather than evenly distributed across the economy.
Goldman Sachs estimates 6–7% of US workers could lose their jobs to AI adoption. That's millions of people — genuinely significant — but also substantially less than the more alarming forecasts circulating. The honest position is that nobody knows the pace or final scale of the disruption, and that the people making the loudest predictions in either direction tend to have an interest in a particular answer.