In February 2026, Block cut nearly half its 10,000-person workforce. Jack Dorsey said AI had made the roles unnecessary and predicted that most companies would reach the same conclusion within a year.
He was not being reckless. He was being rational. That is the problem.
Over 100,000 tech workers were laid off in 2025 alone, with AI cited as the primary driver in more than half the cases. Salesforce replaced 4,000 customer support agents with agentic AI. Goldman Sachs deployed Cognition's Devin, an autonomous coding agent that lets one senior engineer do the work of a five-person team. Anthropic's own CEO, Dario Amodei, warned that AI-driven displacement would be unusually painful, broader and faster than any previous technological shock.
Every executive making these decisions can see the cliff. Displaced workers are also customers. Fire enough of them and the purchasing power that funds your revenue disappears. This is not a subtle point. It is not hidden behind complexity or buried in quarterly projections. It is obvious to everyone in every boardroom where these decisions are being made.
So why can't they stop?
A new paper from Brett Hemenway Falk at the University of Pennsylvania and Gerry Tsoukalas at Boston University does something I have not seen anyone else do. It takes the obvious question and runs it through a formal economic model to prove, mathematically, why rational firms cannot stop automating even when they know it will hurt them.
The paper is called The AI Layoff Trap and the core finding is this: AI-driven layoffs create a demand externality that functions as a Prisoner's Dilemma. Each firm that automates captures the full cost saving from replacing a worker. But it bears only a fraction of the resulting demand destruction. The rest of the damage lands on every other firm in the market.
The math is clean. A firm that automates a task saves the difference between a worker's wage and the cost of AI. But that displaced worker was also spending money, and some fraction of that spending flowed to the automating firm's competitors. Because competitive pricing spreads revenue across all firms equally, each individual company absorbs only 1/N of the demand loss from its own layoffs. The other (N-1)/N falls on rivals.
That gap between private cost and social cost is the trap. Every firm rationally automates more than what is collectively optimal because the math of self-interest makes it the dominant strategy. Not the best response to what competitors are doing. The dominant strategy. Meaning it is the right move regardless of what anyone else does.
In the frictionless limit of their model, where every task is equally easy to automate, the game sharpens into a textbook Prisoner's Dilemma. Every firm displaces its entire workforce even though collective restraint would raise all their profits. The resulting loss is not a transfer from workers to firm owners. It is a deadweight loss. Both sides are worse off.
And here is what makes the paper genuinely unsettling: more competition makes it worse. The standard economic intuition is that competition disciplines firms to act in consumers' interests. In this model, competition does the opposite. More firms in a market means each firm bears a smaller share of the demand it destroys, which weakens the incentive to restrain. A monopolist fully internalizes the damage. A fragmented market exhibits the widest gap between what firms do and what they should do.
I have spent two decades in the staffing and workforce industry. I have watched hiring cycles, restructurings, offshoring waves, and technology shifts play out across every sector you can name. And I have never seen a dynamic that felt quite like this one.
The executives I talk to are not stupid. They are not short-sighted. Many of them can articulate the exact problem this paper describes. They know that firing their workforce eliminates their customer base. They know the math does not add up at scale. They tell me this over coffee, in confidence, with the kind of resigned clarity that comes from understanding a problem you cannot solve by yourself.
Then they go back to the office and approve the next round of automation.
The paper names something I have been circling for months. They call it the Red Queen effect. When AI gets more productive, it does not fix the demand problem. It makes it worse. Each firm sees a market-share advantage in automating faster than competitors. But at equilibrium, everyone has automated equally, the market-share gains cancel out, and the only thing left is a bigger hole in aggregate demand.
Better AI does not save the economy from this trap. It digs the trap deeper.
For the 65 million Gen X professionals at the peak of their careers, this is not an abstraction. These are the customer support managers, the operations leads, the middle managers who make organizations actually function. They sit in exactly the roles the paper identifies as most vulnerable: experienced positions with tasks that are susceptible to large language model automation, in fragmented industries where the demand externality is widest.
The most important section of the paper is the one that evaluates proposed solutions. The researchers tested six policy instruments against the externality. Five of them fail.
Upskilling and retraining. Narrows the gap but cannot eliminate it.
Universal Basic Income. Raises the floor on living standards but does not change a single firm's incentive to automate. UBI adds a constant to demand without touching the margin where the externality operates. It changes payoff levels but not the payoff differences that drive strategic behavior.
Capital Income Tax. Same structural limitation. Scales the entire profit function without altering the per-task automation decision.
Worker equity and profit sharing. Narrows the gap but cannot eliminate it. Closing the gap entirely would require profit-sharing rates above 100 percent of firm earnings. And voluntary profit-sharing will not arise on its own because the marginal cost to each firm exceeds the marginal demand benefit — another Prisoner's Dilemma layered on top of the first one.
Coasian bargaining. Fails for four reasons. Automation is a dominant strategy so no agreement is self-enforcing. The externality is multilateral and diffuse. Automation rates are not contractible between firms. The decisions involve large irreversible costs. This is not a coordination failure where firms just need to talk to each other. It is a structural trap where talking changes nothing.
Pigouvian automation tax. The only instrument that works. It charges each firm for the uninternalized demand loss per task. Each firm already bears 1/N of the demand loss from its own automation, so the tax charges it for the remaining (N-1)/N imposed on rivals. The revenue can fund retraining that raises income replacement, which shrinks the externality over time, which reduces the required tax rate. In theory, the tax is self-limiting.
I am not an economist and I do not pretend to evaluate the mathematical proofs in this paper. What I can evaluate is whether the dynamics it describes match what I see on the ground every day. They do. Precisely.
The executives who cannot stop automating even though they know it is collectively destructive. The experienced professionals who are being displaced not because they lack value but because competitive pressure makes their displacement individually rational. The policy proposals that feel good but operate on the wrong margin entirely.
Here is the thing. This paper is not arguing against AI adoption. It is arguing that the competitive structure of markets will push adoption past the point where it benefits anyone, workers or owners, unless the incentives are corrected at the margin where the externality operates. That is a precise and testable claim. It is also, based on everything I have seen in two decades of workforce strategy, almost certainly correct.
For experienced professionals and the companies that employ them, the implications are clear. The wave is not going to self-correct. The market forces that are supposed to provide a brake are the same forces accelerating the problem. And the policy responses most people are counting on — UBI, retraining programs, equity participation — are operating on the wrong margin.
The workforce does not need sympathy. It needs someone to change the math. Until then, every firm in every competitive market will keep doing exactly what the model predicts: racing toward a cliff they can all see, because the cost of being the one who stops is higher than the cost of going over the edge together.
Source: Hemenway Falk, B. and Tsoukalas, G. (2026). The AI Layoff Trap. SSRN Working Paper No. 6448898. University of Pennsylvania / Boston University.