The artificial intelligence doom narrative had a clear victim: the entry-level knowledge worker. Fresh graduates, we were told, would be the first casualties of the great automation wave—their research skills, writing abilities, and analytical functions rendered obsolete by large language models before they could even update their LinkedIn profiles. Two years into this supposed massacre, the data tells a different story.
Unemployment rates for recent college graduates have remained within historical norms throughout 2026, hovering around 4.2% according to the latest Bureau of Labor Statistics figures. This is roughly where they sat in 2019, before anyone outside a machine learning lab had heard of GPT. The labor market, it turns out, is considerably more complex than a technology demonstration.
The prediction gap
The disconnect between forecast and reality reveals something important about how we process technological change. When ChatGPT emerged in late 2022, consultancies and think tanks competed to produce the most alarming displacement projections. Goldman Sachs suggested 300 million jobs could be affected globally. The World Economic Forum warned of wholesale labor market restructuring. University career counselors reportedly saw spikes in student anxiety about obsolescence before graduation.
What these projections missed was the friction inherent in actual economic transformation. Employers did not, as predicted, immediately replace junior analysts with AI subscriptions. Instead, they experimented cautiously, discovered limitations, encountered compliance concerns, and ultimately continued hiring humans—often to manage the AI tools themselves. The technology proved genuinely useful but far from the autonomous replacement engine its boosters promised.
Where the jobs actually went
The composition of graduate employment has shifted, even if the overall numbers remain stable. Roles explicitly involving AI oversight, prompt engineering, and human-machine collaboration have proliferated. Traditional entry-level positions in legal research, financial analysis, and content production have contracted modestly, but not at the catastrophic rates anticipated. Many firms discovered that the cost of training AI systems to match institutional knowledge exceeded the cost of simply hiring graduates who could learn it.
Meanwhile, sectors less susceptible to automation—healthcare, education, skilled trades—continue absorbing graduates who might previously have pursued knowledge-work careers. The labor market's adaptive capacity, so often underestimated by technologists, has once again proved resilient.
Our take
The AI employment apocalypse joins a long list of technological disruptions that failed to materialize on schedule: the paperless office, the death of retail, the end of human-driven vehicles. This is not to suggest AI will have no labor market impact—it clearly already has, and will continue to. But the gap between Silicon Valley's timeline and economic reality should give pause to anyone making confident predictions about workforce transformation. Technology changes everything, eventually. The eventually matters more than the everything.
