The AI Layoff Trap: Why Companies Can't Stop Automating Even When It Hurts Everyone

Even when firms see the cliff ahead, competition forces them to race toward it anyway.

---

Research Reference: This blog is based on the working paper "The AI Layoff Trap" by Brett Hemenway Falk (University of Pennsylvania) and Gerry Tsoukalas (Boston University), published on arXiv on March 21, 2026. Read the full paper: arxiv.org/pdf/2603.20617

---

About the Researchers

Brett Hemenway Falk works at the intersection of cryptography, computer science, and economic mechanism design at UPenn. His formal modeling background is what gives this paper its mathematical teeth — the results aren't hunches; they're proven propositions.

Gerry Tsoukalas is a professor at Boston University's Questrom School of Business, specializing in AI strategy and platform economics. He has studied how AI pricing algorithms can spontaneously learn to collude — work that sits squarely within his broader focus on unintended market-level consequences of AI.

Together, they build on the task-based automation framework of Acemoglu and Restrepo, but pivot the lens from labor markets to product markets. That pivot is where the externality lives. The paper is a working paper (arXiv:2603.20617v1, econ.TH) — rigorous and empirically grounded, though not yet peer-reviewed.

---

The Central Question

In February 2026, Block laid off nearly half its 10,000-person workforce. Jack Dorsey said AI made those roles unnecessary and predicted most companies would reach the same conclusion within a year. Over 100,000 tech workers were laid off in 2025, with AI cited as the primary driver in more than half the cases.

None of this is hidden. Every CEO knows it. So here's the question the paper asks:

If firms can see that mass layoffs erode the consumer demand they depend on, why are they racing toward it anyway?

The answer: knowing isn't enough. The structure of competition itself makes over-automation a rational dominant strategy — even for firms with perfect foresight.

---

The Trap: A Demand Externality

The core mechanism is simple. Workers are also consumers. When a firm automates, displaced workers lose income and spend less — reducing revenue for every firm in the sector, not just the automating one.

Here's the asymmetry that creates the trap: a firm captures 100% of its cost savings but bears only 1/N of the demand destruction, where N is the number of competitors. The rest lands on rivals.

So automating is always the right private move — even if every firm doing it simultaneously makes all of them worse off. The paper calls this a demand externality, and shows it produces a Nash equilibrium where firms over-automate relative to the cooperative optimum.

In the frictionless limit, the game becomes a textbook Prisoner's Dilemma: full automation is the strictly dominant strategy for every firm, even though mutual restraint would leave everyone better off. Communication is cheap talk — no voluntary agreement is self-enforcing.

---

Both Sides Lose

This is the counterintuitive punch. Over-automation is not a redistribution from workers to owners. It is a deadweight loss that harms both.

Workers lose wages through displacement. Owners lose too — collective demand destruction drives revenues below what cooperation would have produced. A social planner who cares only about firm profits, with zero weight on workers, would still want to reduce the automation rate. The arms race destroys value on both sides.

---

What Makes It Worse

Better AI widens the gap. When AI boosts output per task — not just cutting costs — each firm gains an additional motive to automate: market share. But at the symmetric equilibrium, market share gains cancel out. The demand destruction does not. The authors call this a Red Queen Effect: firms run faster to stay in place, while collectively making things worse.

More competition amplifies it. A monopolist fully internalizes the externality — no rivals to absorb the damage. As markets fragment, each firm's share of the demand loss shrinks, weakening the private incentive to restrain. The most competitive sectors suffer the widest automation gaps.

---

Why Popular Policy Responses Fall Short

UBI raises living standards but adds a constant to demand — it doesn't touch the per-task margin where the automation decision lives. Capital income taxes scale the profit function by a constant, which cancels from the first-order condition. Automation rates are unchanged. Worker equity narrows the wedge by recycling some demand back through profit shares, but can't close it entirely without giving workers more than 100% of profits. Coasian bargaining between firms fails because automation is a dominant strategy — no non-binding deal holds.

---

The Only Fix: A Pigouvian Automation Tax

The paper's solution is a per-task automation tax set equal to the uninternalized demand loss:

τ* = ℓ × (1 − 1/N)

Each firm already bears 1/N of the damage from its automation. The tax charges it for the remaining (1 − 1/N) imposed on rivals. The arms race stops. The cooperative optimum is restored.

Tax revenue directed toward retraining raises the income-replacement rate η, which shrinks the demand loss ℓ, which reduces the required tax rate in future periods. Done right, the tax is self-limiting — it funds the conditions that eventually make it unnecessary.

---

My Take

I find the core logic compelling. The demand externality mechanism is clean, the math is sound, and the intuition maps clearly onto what we're actually seeing.

But I hold it with genuine uncertainty. I believe the structural argument is correct. Whether it plays out in full is another question.

The model is deliberately simplified — one sector, one period, symmetric firms. Real economies are messier. New task creation could absorb displaced workers faster than the model assumes. A unilateral automation tax could push adoption offshore. AI-adjacent sectors could generate enough new high-paying roles to keep income replacement healthy.

The outcome hinges on a race between three speeds:

1. How fast AI displaces workers — accelerating
2. How fast displaced workers find comparable jobs — historically slow; early evidence suggests AI is hitting entry-level workers hardest
3. How fast policy responds — currently, very slowly

The paper's real contribution isn't a prediction. It's a proof that if displacement outruns reabsorption, no market force will self-correct it. That's the warning worth taking seriously — not as fatalism, but as a reason to act before the cliff arrives.

---

Conclusion

The AI layoff trap isn't about greedy corporations or reckless executives. It's about how individually rational decisions — made by companies that can see exactly what's happening — still produce collectively irrational outcomes when competition is the context.

The policy conversation today focuses on the aftermath: retraining, income support, safety nets. These matter. But they don't slow the arms race. Only a tax on automation itself changes the calculus that drives it.

The trap is visible. The solution is identifiable. The clock is running.

---

📄 Paper Details

All results cited here are drawn from the authors' work. This blog is a lay interpretation with personal commentary — any errors in summary are mine, not theirs.

---