Hello everyone, I'm Tony Bai.
Over the past year, the "layoff panic" brought by AI has swept through nearly the entire tech industry.
Back in February, Jack Dorsey's Block laid off nearly half its workforce. He was blunt: "Because AI made many roles unnecessary."
Salesforce replaced 4,000 customer service agents with AI, and Cognition's AI programmer Devin allows a single senior engineer to do the work of five people.
We seem to be in the midst of an "efficiency revolution" driven by AI. Managers cheer for "cost reduction and efficiency gains," while we workers tremble, worrying that our jobs could be snatched away by an invisible agent at any moment.
But what if I told you today that the ultimate outcome of this seemingly "zero-sum game" layoff frenzy might not be "capitalists win, workers lose," but rather "everyone loses together"?
Just this past March, two scholars from the University of Pennsylvania and Boston University published an extremely rigorous, even chilling economics paper—"The AI Layoff Trap".
Using a highly stringent mathematical model, this paper arrives at a spine-tingling conclusion:
In a perfectly competitive market, all rational companies will descend into a frantic "automation arms race." They will continuously use AI to lay off employees until they completely destroy the overall market's consumer demand, ultimately leading to the collapse of both corporate profits and worker incomes.
Today, let's deconstruct this doomsday prophecy paper and see how we are, step by step, willingly jumping into this "lose-lose" trap.
The Prisoner's Dilemma: Why Are All Companies Still Accelerating Madly Toward the Cliff?
The core of the paper is built on an extremely simple economic common sense: Laid-off employees are also consumers. When they lose their income, the purchasing power of the entire market decreases.
Since even a street vendor selling vegetables understands this logic, why do giant corporations, boasting countless top economists, still race toward the cliff of "zero demand"?
The answer lies in a classic game theory model: The Prisoner's Dilemma.
The paper constructs a simple competitive market model:
- There are N companies in the market, competing with each other.
- Each company can choose to replace a portion of its human employees with AI, thereby reducing costs.
- But every layoff leads to a slight decrease in total consumer demand in the market.
Now, let's make a decision from the perspective of one company's CEO:
Scenario 1: If other companies choose not to lay off employees
In this case, if I choose to lay off, I can reap all the cost reduction benefits (increased profits) brought by AI for myself, while the decline in market demand caused by the layoffs is distributed among all N companies.
For me, laying off is the absolutely optimal strategy.
Scenario 2: If other companies are all laying off frantically
At this point, the market's total demand is already shrinking. If I choose not to lay off, I not only suffer alongside them from the market contraction but also fail to enjoy the cost advantages of AI. My market share will be rapidly eroded.
To survive, my only choice is to lay off even more aggressively than they do.
Do you see it?
No matter what competitors do, for myself, "maximizing automation (layoffs)" is always my optimal solution (a strictly dominant strategy).
And when every company in the market thinks and acts this way, the entire system falls into an irreversible "death spiral." The figure below, through three sets of two-dimensional graphs, visually demonstrates how the shadow area of "over-automation" (representing the degree of the lose-lose outcome) becomes increasingly larger and darker as the number of market competitors (Number of firms N) increases.
The over-automation wedge
Every company makes the most rational decision for itself, but it ultimately leads to the worst possible outcome for the collective. This is the essence of the "AI Layoff Trap."
"Better" AI, Faster Destruction: The "Red Queen Effect"
Some might optimistically think, "It doesn't matter; as long as AI's productivity is high enough, the new wealth it creates will always fill the consumption gap left by the laid-off employees."
But this paper offers an even more despairing corollary: "Better" AI doesn't alleviate this problem; rather, it accelerates the process of destruction.
Because a more productive AI creates a greater illusion of "market share gain" for the companies that adopt it first. This further stimulates all companies to plunge even more frantically into this arms race.
This resembles the "Red Queen effect" from "Alice in Wonderland": You must run with all your might just to stay in the same place.
Eventually, in an equilibrium state where everyone (including AI) is panting from exhaustion, no single company has actually gained additional market share. The entire system is simply hurtling toward that "zero demand" cliff at an even faster speed.
Malfunctioning "Antidotes": Why UBI and Upskilling Can't Save Us?
Faced with this brutal dilemma, several seemingly appealing "antidotes" are circulating in society. But this paper uses mathematical models to puncture their illusions one by one.
Antidote 1: Universal Basic Income (UBI) or Increasing Capital Gains Tax
Conclusion: Completely ineffective.
Because UBI and capital taxes affect a company's "profit level," not the "marginal decision" that drives layoffs.
As long as the cost of replacing an employee with AI remains lower than that employee's wages, no matter how many subsidies you give or taxes you levy on this company, its motive to lay off employees will not change.
Antidote 2: Employee Upskilling or Employee Stock Ownership Plans (ESOP)
Conclusion: Partially effective, but cannot cure the root cause.
Enabling laid-off employees to find higher-paying jobs through retraining, or letting them hold company stock to share in the profits from automation, can indeed partially "recycle" the lost consumer demand.
However, the paper points out that this "recycling" process can never 100% offset the initial loss. Because there is always friction in the flow of information and capital, as long as there is even a little bit of a "Demand Externality," the devil driving everyone toward the cliff still exists.
The Only "Brake": Painful but Necessary "Automation Tax"
After ruling out all seemingly nice "market-based" solutions, the paper ultimately points to a very classical and highly controversial "ultimate weapon" — The Pigouvian Tax.
This concept, proposed by economist Arthur Pigou in 1920, has a core idea: Directly tax behaviors that generate negative externalities.
For example, if a factory emitting one ton of exhaust gas causes $100 of environmental damage to society, then levy a $100 "pollution tax" on it.
In this paper's model, this "tax" is specifically formulated as an "Automation Tax."
Whenever a company replaces a human position with AI, it must pay a tax for this "automation act" itself. The amount of this tax should precisely equal the "consumer demand loss" that this layoff inflicts on the entire society.
Only in this way can the social cost that was "externalized" by the enterprise be "internalized" back into its own decision-making model, thereby forcing it to think twice before laying off employees.
Of course, the authors acknowledge that implementing an "Automation Tax" faces enormous challenges in reality: how to measure it precisely? How to prevent companies from moving production overseas?
But they emphasize that, in theory, this is the only policy tool that can fundamentally hit the brakes on the "layoff arms race."
Summary: What Kind of Future Are We Creating?
This paper, though written in the language of economics, explores a future that every one of us in tech is personally participating in and shaping.
It acts like a mirror, reflecting the cognitive blind spots we have when pursuing the "technological optimum."
We are obsessed with replacing customer service agents with AI Agents, junior programmers with AI Coders, and we cheer for every successful instance of "cost reduction and efficiency gain." But we rarely think about it, when those people we personally "optimize" away lose their spending power, where is the foundation of the commercial edifice we built with our own hands?
The value of this paper lies not in offering a perfect answer, but in raising a higher-dimensional question:
When "individual rationality" conflicts with "collective rationality," what role should we, as system builders, play?
Do we continue running blindly, accelerating this "lose-lose" game?
Or do we pause to think about how to introduce, from the architectural level, new rules that can balance "efficiency" and "fairness" with a more humanistic touch?
This actually goes beyond the scope of economics; it feels more like a profound issue of "architectural ethics."
Resource link: https://arxiv.org/abs/2603.20617
👇 Today's Interactive Discussion:
After reading this paper's deductions, do you also feel a chill about the future of AI? Do you think an "Automation Tax" is a feasible solution, or a utopian fantasy?
Welcome to share your thoughts in the comments section!
If this article has been helpful to you, please like, recommend, and share it!