Companies spent the last three years firing workers and telling shareholders that artificial intelligence would pick up the slack. The pitch was clean: automate routine tasks, shrink payroll, and let the productivity numbers speak for themselves. But two large-scale studies now suggest the numbers have almost nothing to say. The jobs vanished. The efficiency gains did not follow.
A survey of roughly 6,000 senior executives across the United States, United Kingdom, Germany, and Australia, published as NBER Working Paper 34836, asked the same battery of questions between November 2025 and January 2026. More than 90% of respondents said AI had produced no discernible change in their company’s employment levels over the prior three years. Eighty-nine percent said it had done nothing measurable for labor productivity.
A digest of that research, released by the National Bureau of Economic Research in May 2026, put the finding in plain terms: among the CEOs, CFOs, and senior finance managers polled, the overwhelming majority reported that AI tools had failed to move the metrics shareholders care about most, specifically output per worker and headcount efficiency.
A second, independent NBER study, Working Paper 34984, surveyed approximately 750 corporate executives and reached a similar conclusion. It found limited near-term employment declines attributable to AI at the aggregate level, though effects varied by firm size and sector. When two separate surveys of different executive populations land in the same place, the signal is hard to dismiss.
What the 80% figure actually tells us
Running alongside the NBER data is a widely cited Gartner finding: 80% of companies that deployed AI reduced their workforces. That number has circulated through business media for months. Gartner has not published the full survey methodology, sample composition, or question framing in a format available for independent review, so the statistic works better as a directional signal than a precise measurement. It points to a broad pattern of AI-linked restructuring, but outside analysts cannot verify how representative the respondents were or how “workforce reduction” was defined.
Still, the overlap between the Gartner figure and the NBER research carries a sharp implication. If something close to 80% of AI-adopting firms did cut staff, and roughly 90% of executives say those same tools have not improved productivity, then a large number of companies eliminated jobs for gains that never showed up. That is not a rounding error. It is a strategic misfire at scale.
Why the productivity payoff has stalled
Several explanations compete for attention, and none of them rule the others out.
The integration lag. AI tools may need years of workflow redesign, data cleanup, and employee retraining before they register in productivity metrics. The NBER surveys capture a three-year lookback window. For technologies that demand deep organizational change, three years may not be long enough. Some economists have compared the current moment to the decades-long lag between the initial adoption of electricity in factories and the productivity gains that eventually followed, though the analogy is imperfect and the parallel remains debated.
The capability gap. Generative AI can draft emails and summarize documents, but many firms are deploying it in environments with messy data, unclear processes, and employees who received little or no training. In those conditions, the technology can create new bottlenecks as fast as it clears old ones.
The concentration problem. The aggregate “no impact” finding may be hiding a sharp split. A small number of large technology firms with dedicated AI engineering teams could be capturing real gains while the majority of adopters see nothing. NBER Working Paper 34984 flagged exactly this kind of variation by firm size and industry. If the benefits are pooling at the top, AI is not failing broadly so much as widening the gap between dominant firms and everyone else.
The measurement blind spot. Executives typically evaluate productivity through high-level indicators like revenue per employee or output per hour. AI may be reshaping what individual workers do, eliminating some tasks while generating others, without yet registering in those headline numbers. The effects could be real but too diffuse to surface in the metrics C-suite leaders watch most closely.
The human cost is already locked in
None of these explanations change much for the people who already lost their jobs. Whether the productivity gains are delayed, illusory, or concentrated in a handful of firms, the layoffs happened. Workers in customer service, content production, back-office operations, and mid-level administrative roles have absorbed the heaviest impact of AI-driven restructuring. Concrete data on reemployment rates and wage outcomes for these displaced workers remains scarce, which itself underscores how little attention has been paid to the human side of AI-linked job cuts. For them, the question of whether AI will eventually justify the reductions is beside the point. The paychecks stopped months or years ago.
The reputational cost for companies is compounding, too. Firms that announced AI-driven layoffs positioned themselves as forward-thinking. If the promised efficiencies keep falling flat, those same announcements start to look like conventional cost-cutting wrapped in technological language. Investors, regulators, and the employees who remain will all be scanning the next round of earnings reports for proof the strategy worked. As of mid-2026, that proof has not arrived.
What the research means for the next round of AI decisions
For executives weighing whether to restructure around AI, the data from the NBER studies points in one direction: cutting headcount in anticipation of AI-driven efficiency is a bet that, according to the best available evidence, has not paid off for the vast majority of firms. Companies considering similar moves face a compounding risk. They absorb the human and reputational damage of layoffs upfront, then face financial exposure when the promised efficiencies fail to arrive on schedule.
The research does not prove AI will never deliver broad productivity gains. It proves that across thousands of firms and four countries, over a meaningful stretch of time, it has not done so yet. Until stronger evidence surfaces, the more defensible assumption is that AI remains an incremental tool whose value depends on careful, sustained integration, not a shortcut to leaner operations. Companies that treated it as a reason to cut workers may eventually discover they traded institutional knowledge and workforce stability for a technology that, through the first half of 2026, has mostly generated expectations it cannot meet.
Warren Cohen is a finance writer based in Phoenix, Arizona, covering personal finance topics including credit, banking, and beginner investing. He earned his degree in business administration from Arizona State University and began his career working in consumer finance, where he gained direct experience with lending and credit systems. He now writes for personal finance websites and fintech platforms, focusing on clear, practical content that helps readers make informed financial decisions.
