The question has hovered over the AI boom for years: if companies automate enough workers out of their jobs, who is left to buy the products those companies make? Two American economists have now built a formal mathematical model around that question, and the answer they found is more unsettling than the question itself. The problem, they argue, is not that companies cannot see the cliff coming. It is that they cannot stop driving towards it.
Brett Hemenway Falk of the University of Pennsylvania and Gerry Tsoukalas of Boston University published their paper, 'The AI Layoff Trap,' in March 2026 as a Wharton School working paper, also available on arXiv and SSRN. The paper has not yet completed formal peer review, but it has attracted significant attention from economists and policymakers amid a wave of AI-attributed job cuts that shows little sign of slowing.
What the trap looks like
The core mechanism is straightforward, even if the mathematics behind it runs to 60 pages. Imagine a single large company that employs most of the people who also buy its products. When a powerful new technology arrives, its chief executive can see the full picture: fire too many workers and you destroy your own customer base. In that scenario, a rational monopolist holds back, automating only enough to gain an edge without gutting demand.
The logic changes completely in a competitive market. Each company in a crowded sector sees the cost savings from automation as almost pure gain, because its own workers were never spending much with it specifically — they were spreading their wages across dozens of rivals. The demand damage from laying off those workers therefore falls mostly on competitors, not on the company doing the firing.
“"If AI displaces human workers faster than the economy can reabsorb them, it risks eroding the very consumer demand firms depend on. We show that knowing this is not enough for firms to stop it." — Brett Hemenway Falk and Gerry Tsoukalas, 'The AI Layoff Trap,' arXiv, 2026”
When every company in a sector reasons this way simultaneously, the collective result is catastrophic. Demand collapses across the board and every firm ends up worse off than if none of them had automated so aggressively. Economists call this a dominant strategy: the action that is individually rational for each player regardless of what others do, but which produces a socially disastrous outcome when everyone plays it. It is the logic behind the prisoner's dilemma, and it has been well understood in economic theory for decades. The authors argue it now applies directly to corporate AI adoption.
The numbers behind the warning
The paper arrives against a backdrop of data that suggests the described process may already be underway. According to figures cited in the paper, over 100,000 tech workers were laid off in 2025, with AI cited as a primary driver in more than half the cases, concentrated in customer support, operations, and middle management. The pace accelerated: nearly 80,000 tech jobs were cut in the first quarter of 2026 alone, with close to half attributed to automation technologies.
Individual corporate decisions illustrate the scale. The paper notes that Salesforce replaced 4,000 customer-support agents with AI systems, while Block cut nearly half its 10,000-person workforce in February 2026. The World Economic Forum's Future of Jobs Report projects that AI and related technologies will displace roughly 92 million jobs globally by 2030 while creating approximately 170 million new ones — a net positive on paper, but one that masks a critical gap: the workers losing roles are not automatically the ones who fill the new positions.
“"Displacement has intensified over the past four decades while the creation of new work has not always kept pace, and early signs suggest the current wave is disproportionately affecting entry-level workers." — Autor et al., 2024, as cited in 'The AI Layoff Trap'”
Why CEOs cannot simply agree to slow down
The most striking element of the paper is its insistence that corporate self-restraint is not a realistic solution — not because executives are reckless, but because the incentive structure makes restraint individually irrational. Even if every major tech company agreed in a room to scale back automation, each firm would face an immediate temptation to defect once it walked out the door, knowing that rivals had committed to holding back. That opportunity — to automate while competitors do not — is also, in the conventional view, a duty to shareholders. Tsoukalas has put the point plainly: waiting for firms to figure it out for themselves is the worst possible thing policymakers can do.
The paper tested a range of proposed remedies against its model. Universal basic income, giving workers equity stakes, capital income taxes, and various forms of Coasean bargaining — where companies negotiate directly among themselves — all failed to correct the underlying incentive. None of them directly addressed the competitive pressure to replace a worker with a machine.
The policy case for a 'firing tax'
The one instrument the paper found effective is what its authors describe as a Pigouvian automation tax — a levy applied to full worker replacement, not to augmenting workers with AI tools. The principle is borrowed from environmental economics: a carbon tax does not ban pollution, but raises its price until the polluter internalises the cost it is imposing on others. The automation tax would do the same for labour displacement, making it expensive to eliminate a role entirely while leaving the use of AI as a productivity aid largely unaffected.
The political obstacles are real. The word 'tax' carries toxicity in most legislative environments, and critics at institutions such as the Tax Foundation argue that the existing tax code already disadvantages capital investment relative to wage costs in several respects, making additional levies on automation doubly distortive. Some economists prefer demand-side approaches: Brookings has leaned towards a consumption-based levy, while proposals circulating in the United States Congress in 2026 range from per-token charges on AI usage to an AI Workforce Reinvestment Fund funded by industry revenues. Tsoukalas has acknowledged that direct implementation of an automation tax may be politically unrealistic, and that subsidies for firms that retain workers could serve as an equivalent instrument without the same political friction.
“"The competitive pressure to deploy agentic AI is structurally decoupled from whether aggressive automation is profitable in aggregate." — Adnan Masood, analysis of 'The AI Layoff Trap,' Medium, April 2026”
The IMF, which hosted a high-level workshop on AI's macroeconomic implications in December 2025, has flagged similar concerns, noting that AI's impact on workers depends on how it alters task composition and that the transition demands policy treatment as a macro-critical event rather than a standard technology shock. The Fund has stopped short of endorsing a specific automation tax, but it has not ruled one out for scenarios of moderate-to-rapid AI acceleration. What both the IMF and the Falk-Tsoukalas paper agree on is that the decision of what to do — and when — cannot safely be left to the market alone.
This article is free to read. It always will be — no paywall, no account, no tracking.
