A new peer-reviewed paper from researchers at the University of Pennsylvania and Boston University is making waves in business circles — and for good reason. The study, titled “The AI Layoff Trap,” argues that companies racing to cut headcount through automation may be setting themselves up for a slow-motion economic self-defeat.
The paper was published on arXiv in March 2026 by Brett Hemenway Falk (University of Pennsylvania) and Gerry Tsoukalas (Boston University). Its central argument is uncomfortable for anyone who has cheered the efficiency narrative around AI: competitive pressure is pushing firms to automate far more than is collectively optimal, and the result harms both workers and the businesses doing the cutting.
The Math Behind the Trap
Here is how the researchers explain it. When a company automates a role, it captures 100% of the wage savings. But the displaced worker now spends less — and that lost spending is spread across every business in the sector. If there are 20 competitors, each absorbs roughly 1/20th of the demand it just destroyed.
This asymmetry is the trap. Every individual firm has a rational incentive to automate first and fast, but the collective outcome is a market where everyone is earning less and consumer demand has quietly collapsed. The researchers describe this as “deadweight loss” — not a transfer from workers to owners, but a destruction of value on both sides.
More competition makes it worse. Better AI makes it worse. The race to the bottom has no natural floor.
The Scale Is Already Visible
The numbers backing the theory are hard to ignore. Over 100,000 tech workers were laid off in 2025 alone, with AI cited as a contributing factor in more than half those cases. In just the first months of 2026, 92,000+ employees across 98 companies have already been let go.
The researchers also found that common policy fixes — wage adjustments, retraining programs, universal basic income, profit-sharing, or worker equity — cannot solve the core problem. Only a Pigouvian automation tax (a levy on automation proportional to its external cost) would theoretically break the cycle. That solution is politically unlikely any time soon, which means the trap remains very much open.
What This Means for Business
The study is not arguing against AI. It is arguing against using AI primarily as a blunt instrument to cut costs and headcount without thinking through second-order effects.
The businesses that avoid the trap are the ones treating AI as a capacity multiplier rather than a workforce reduction tool. There is a meaningful difference between deploying AI agents to handle work that previously went undone — after-hours calls, data backlogs, repetitive admin, internal knowledge retrieval — and simply replacing people who were doing work that created real economic activity.
When a business automates the former, it gets more done without shrinking its team’s purchasing power or eroding the spending capacity of its local economy. When it automates the latter purely for headcount reduction, it pockets a cost saving while quietly destroying demand — including, eventually, demand for its own products.
This matters especially for small and mid-size businesses. Enterprise companies with diversified revenue and global customer bases can absorb demand shifts more easily. For regional businesses, the trap is faster and sharper.
The Smarter Adoption Path
The research reinforces an approach to AI that prioritises expansion over extraction. Build AI agents to do the work you could never staff. Add capacity in areas where demand exists but delivery has been constrained. Use AI to serve customers better and faster, not just to shrink the payroll.
Enterprise DNA’s work with business owners has consistently pointed toward this conclusion. The most successful AI deployments are the ones that grow revenue and capability — handling more enquiries, processing more data, supporting more customers — while keeping the human team focused on judgment, relationships, and complex work that machines still handle poorly.
The researchers at Penn and Boston University have now put the economic model behind something many operators were already sensing: pure headcount automation is a race that nobody wins.
The full paper is available at arxiv.org/abs/2603.20617.