Do executives and employees agree on AI's job impact?

An NBER working paper (Yotzov, Barrero, Bloom, Davis et al., 2026-02) surveys "nearly 6,000 senior business executives" at US, UK, German, and Australian firms about AI adoption and its effects on jobs, productivity, and output. It finds that "69% of firms actively use AI," with "higher usage rates at younger and more productive firms," and that "more than two thirds of executives regularly use AI" — but their own usage "averages only 1.5 hours a week." Despite this reach, "nine-in-ten" executives report "no impact on employment or productivity" at their own firm over the past three years. Looking ahead, the same executives predict AI "will boost productivity at their firms by an average of 1.4%, raise output 0.8%, and cut employment 0.7%" over the next three years, while employees surveyed separately "anticipate that AI will raise employment 0.5%" at their firms over the same period — a divergence the authors describe as "an expectations gap between employers and employees."

The paper treats breadth of adoption, depth of reported effect, and forward prediction as three distinct, self-reported measurements, none of them drawn from output or payroll data. The 69% adoption figure and the 1.5-hour weekly-use figure describe exposure to AI tools. The "nine-in-ten report no impact" figure is the executives' own retrospective judgment of causal effect so far at their firm. The 1.4%/0.8%/-0.7% figures are the same executives' predictions about the next three years, not measurements — and it is this prediction, set against employees' competing prediction of a 0.5% employment gain, that the authors foreground as the headline result, ahead of the adoption or usage figures themselves.

This sits against Is generative AI displacing workers at economy-wide scale?, which draws on actual ADP payroll data through June 2026 and finds no broad displacement beyond a hiring shortfall for the youngest workers in exposed occupations — a measured result consistent so far with these executives' own retrospective report of no impact, not with their forward prediction of a 0.7% employment cut. It also echoes, at a different point in the employment relationship, Are recruiters and job seekers really adopting AI in hiring?, another case where two sides of the same employment relationship hold diverging, self-reported expectations about AI's effect on their own prospects.

The excerpt gives no survey instrument, no breakdown by firm size, sector, or country, and no detail on how the employee comparison sample was drawn or sized against the roughly 6,000 executives — so the "expectations gap" cannot be weighed against sampling or response differences between the two groups. Both the executives' past-impact report and their future prediction are self-reports of belief, not independent measures of productivity, output, or headcount; what the paper documents is a gap in what employers and employees expect, not yet a gap between what either side expects and what will actually happen.