The Gap Between Fear and Reality

The most powerful voices in artificial intelligence really want you to worry about your job — if you're a white-collar worker. Microsoft AI chief Mustafa Suleyman predicts AI will cause office jobs to crumble in 18 months. Anthropic CEO Dario Amodei thinks entry-level jobs in the space will be cut in half in a similar timeline. Even Federal Reserve Chair Jerome Powell has warned that AI is quietly impacting the labor market as job creation hovers near zero.

But behind closed doors, business leaders controlling company headcounts are telling a more nuanced story — and a new working paper from the National Bureau of Economic Research (NBER) quantifies exactly how big the gap really is.

What the Data Actually Shows

The NBER study, based on a survey of 750 chief financial officers from U.S. firms, found that less than half — just 44% — say they plan on some AI-related job cuts in 2026. When co-authors calculated what that amounts to across the broader economy, they found just 0.4% of total roles — about 502,000 jobs out of approximately 125 million — are expected to be lost this year due to AI. Roughly half of those losses will come from white-collar workers.

That 9x increase from last year's 55,000 AI-attributed layoffs is striking — but still, in the researchers' own words, "a rounding error against the overall workforce."

"It's not the doomsday job scenario that you might sometimes see in the headlines," said John Graham, co-author of the study and director of the Duke CFO Survey, conducted in partnership with the Federal Reserve Banks of Atlanta and Richmond.

The Productivity Paradox is Back

More troubling than the headline job numbers is a finding buried deeper in the report: there's a wide gap between the perceived and actual productivity gains from AI. Companies believe they're getting more productive, but the revenue data doesn't support it — yet.

Goldman Sachs senior economist Ronnie Walker noted earlier this month that "we still do not find a meaningful relationship between productivity and AI adoption at the economy-wide level." Workers have reported that AI is actually making them less productive in some cases, with time spent on certain job responsibilities increasing by up to 346%.

Economists have a name for this gap. The researchers invoke Solow's Paradox — coined by Nobel Laureate Robert Solow in 1987 — which describes the observation that transformative technology can appear ubiquitous while remaining absent from economic data.

"You can see the computer age everywhere but in the productivity statistics," Solow famously said. The AI era appears to be following the same script.

Graham put it bluntly:

"Companies have invested and they're realizing all these cool things they're either starting to do or they hope to do in the near future. But it's not really showing up yet in revenue."

The Real-World Layoffs Already Happening

The abstract numbers come to life with some notable examples already on the books. Jack Dorsey's Block cut about 40% of its workforce — more than 4,000 employees — citing AI. Australian-American firm Atlassian cut 10%. Meta is reportedly planning cuts of 20% as CEO Mark Zuckerberg builds an AI agent clone of himself to help run the company.

On top of that, the broader job market remains weak: U.S. employers posted 92,000 job losses last month, and the unemployment rate has ticked up to 4.4%.

A Silver Lining for Small Business

Not all the news is grim. The NBER study found that AI adoption could actually increase hiring among smaller firms. Companies with fewer than 500 employees have only just begun to invest in AI — and as adoption increases, small firms noted they plan to increase hiring in technical roles. Larger firms, meanwhile, plan to hold technical headcounts steady.

"If anything, small companies are hiring a bit on the technical side which will offset [losses] a little bit," Graham said.

The researchers are careful to note their study only captures the short-term picture. Whether the dire longer-term predictions materialize remains genuinely unknown.

"Who knows what's going to happen in 2028?" Graham said. "I'm not making a prediction that there will never be any jobs lost two, three and five years from now to AI."

The Takeaway

The honest answer is that AI's economic impact in 2026 looks much messier than the clean narratives from either optimists or doomsayers suggest. Job losses are real and accelerating, but they remain manageable at scale — for now. Productivity gains are promised but largely unproven. And the technology that was supposed to transform everything has, so far, mostly transformed the way people talk about work rather than work itself.

Solow's paradox proved temporary — the productivity gains from computers eventually showed up, spectacularly, in the late 1990s. The question for AI is whether that payoff is two years away, or two decades.