The AI Layoff Trap: why every company will automate even knowing it will break them
In February 2026, Jack Dorsey laid off nearly half of Block. Around 4,000 people. His justification was that AI made many of those roles unnecessary and that “within the next year, the majority of companies will reach the same conclusion”. In 2025, U.S. employers announced over 1 million job cuts, with AI explicitly cited in 55,000 of them.
Here is the problem nobody is discussing properly: every one of those companies knows that mass layoffs erode aggregate demand. A laid-off worker does not buy things. And yet, every one of them keeps laying people off. Why? A recent paper by Brett Hemenway Falk (UPenn) and Gerry Tsoukalas (Boston University) proves mathematically that this is not a failure of foresight. It is a structural prisoner’s dilemma, and no rational firm can break it alone.
What is happening
Traditional economic theory said automation is self-correcting. Layoffs in one sector open jobs in another, wages adjust, no one is permanently left behind. That logic worked in the 20th century because the pace of substitution was slow enough for reabsorption to happen.
Falk and Tsoukalas show the game has changed. In their model, N symmetric firms compete in the same market. Each one chooses how much of its workforce to replace with AI. The cost saving from automation goes entirely to the firm that automates. But the demand loss generated, because laid-off workers stop buying, gets distributed across all N firms in the sector.
The result is a devastating asymmetry. When you cut a team, you capture 100% of the salary savings. But you only absorb 1/N of the demand drop that causes. The rest lands on your competitors. For you, automating is strictly dominant. For the sector as a whole, it is collectively destructive.
Why you should care
The paper makes three testable predictions that are materializing in 2026 data.
The first is that profit erosion will show up alongside mass layoffs, especially in fragmented sectors. This contradicts the conventional view that cutting costs with AI improves margin. In the model, cutting costs via AI in a competitive market erodes demand faster than it improves margin. The firm ends up worse off.
The second is that better AI makes the problem worse, not better. The authors call this the Red Queen effect, after Alice in Wonderland, where the Red Queen has to run faster and faster just to stay in place. Each firm perceives a market-share gain from automating beyond rivals. In the symmetric equilibrium those gains cancel out. What is left is the extra demand destruction.
The third, and most political, is that almost none of the currently discussed solutions actually work. The paper evaluates six instruments:
| Policy | Reduces the wedge? | Fixes the externality? |
|---|---|---|
| Universal Basic Income (UBI) | No | No |
| Capital income tax | No | No |
| Worker equity participation | Partially | No |
| Coasean bargaining among firms | Partially | No |
| Retraining / upskilling | Partially | Only if η = 1 |
| Pigouvian automation tax | Yes | Yes |
UBI does not change the firm’s marginal decision. If AI saves X per laid-off worker and the government distributes money to everyone, the layoff decision stays the same. UBI raises the floor on living standards (which has value), but it does not stop the arms race.
A capital income tax does not work either, and the reason is more technical. It scales the entire profit function by (1−t), which cancels out in the first-order derivative. The firm’s first-order condition does not change. The tax redistributes gains after the damage is already done.
The only policy that operates on the right margin is a Pigouvian tax on each automated task, at the rate τ* = ℓ(1 − 1/N), where ℓ is the demand loss per laid-off worker. That tax charges the firm for the externality it imposes on competitors. The revenue can fund retraining, which raises the worker reabsorption rate and reduces ℓ over time. The tax is self-funding and tends to zero as the economy adjusts.
How to apply this tomorrow
You are not Janet Yellen, so you will not implement a Pigouvian tax. But the paper has three operational implications you can use immediately.
If you decide on layoffs: the result shows that cuts in a fragmented sector with high AI exposure are likely to erode your own firm’s profit, not just your competitors’. The math of “I will save X in salaries” needs to subtract the revenue drop coming from the aggregate demand contraction. In sectors like customer support, software services and financial back-office (the three the paper highlights as most exposed), that arithmetic can go negative.
If you invest in stocks: the paper suggests a clear empirical signature. Fragmented sectors where multiple firms are laying off simultaneously will show margin erosion that standard competitive models cannot explain. That is an actionable short thesis in some niches.
If you are a worker or entrepreneur: raising η (the reabsorption rate) is where the game is. Work that retrains people into higher-paying roles in the AI-adjacent boom (infrastructure, energy, implementation services) reduces ℓ and, at the edge, can flip the externality from over-automation into under-automation. The paper does not treat this as utopia. It shows mathematically that when η > 1 (reabsorption pays more than the original job), the competitive equilibrium automates too little.
Why this is not just theory
Three 2026 data points the paper cites that should worry any strategist:
- In Q1 2026, business investment overtook consumer spending as the leading contributor to U.S. GDP growth. That had not happened in years.
- The personal savings rate fell to 3.6% in March, the lowest level since late 2022.
- Anthropic CEO Dario Amodei has said AI-driven job disruption will be “much broader and much faster” than previous technological shocks.
All three are consistent with the paper’s mechanism. Labor income contracting, savings falling (because laid-off workers consume their reserves), and firms still accelerating automation.
Glossary
- Externality: an effect of a decision on third parties that is not priced into the decision. Pollution is the classic example. Here, the “demand externality” is the revenue drop your layoff causes at competitors.
- Pigouvian tax: a tax designed to make the firm pay exactly the external cost it imposes on third parties. Named after economist Arthur Pigou.
- Prisoner’s Dilemma: a game structure where the individually rational decision leads to the worst collective outcome. Here, each firm has a dominant incentive to lay off, but all of them end up worse.
- Over-automation wedge: the gap between the automation rate in competitive equilibrium and the rate that maximizes aggregate sector profit. In the paper, this wedge is ℓ(1 − 1/N)/k.
- MPC (Marginal Propensity to Consume): the fraction of each extra dollar of income that immediately becomes spending. Workers have a high MPC. Capital owners have a low MPC. That is why shifting income from labor to capital depresses aggregate consumption.
- Reabsorption rate (η): the fraction of income lost by laid-off workers that comes back via reemployment, transfers or retraining. η = 1 means full reabsorption. η < 1 is the historical case.
- Reinstatement effect: term used by Acemoglu and Restrepo for the creation of new tasks that reabsorb workers displaced by automation. It worked in the 19th and 20th centuries. The open question is whether it will work with AI.
Your next move
If you have decision power over hiring or layoffs: ask one question before the next planning cycle. How many of your direct competitors are running the same automation move right now? If the answer is “several”, the paper suggests you are about to enter an arms race whose equilibrium is collectively destructive. It is worth reading the full paper before signing the next reduction plan. It is 40 pages and is openly available on arXiv.