Two economists at Wharton and Boston University built the math.
The math leads to one place: an economy that produces everything. And sells it to nobody.
For the past several years, the dominant reassurance in the AI workforce debate has been this: technology has always displaced jobs, and it has always created new ones. The Luddites were wrong. The textile workers were wrong. The elevator operators and the switchboard operators and the travel agents were wrong. The economy adapted. It always does.
That argument just ran into a peer-reviewed mathematical proof. And the proof does not go where the reassurance wants it to go.
The AI Layoff Trap: What the Math Actually Shows
The paper is called “The Automation Trap.” The authors are Brett Hemenway Falk of the University of Pennsylvania (Wharton) and Gerry Tsoukalas of Boston University. Published March 2, 2026. Peer-reviewed. Game theory and mechanism design ‚Äî not a survey, not an opinion piece, not a thought experiment. A formal economic proof.
Here is what they proved.
When one company automates and reduces its workforce, it gains a competitive advantage. Its costs go down. Its margins improve. This is rational behavior for any individual firm.
But that rational behavior creates a pressure that spreads. Other companies in the same industry face a choice: automate to match the competitor’s cost structure, or accept declining competitiveness. So they automate too.
Now the workers who lost jobs at Company A cannot find comparable work at Companies B, C, and D — because B, C, and D just made the same decision for the same rational reasons.
The workers’ income falls. Their purchasing power falls with it. Consumer demand falls. The companies that just automated to improve margins find that the market for their products is contracting, because the consumers who bought those products were the workers they just displaced.
The loop closes. And then it tightens.
Nobody Is Doing Anything Wrong. That Is Exactly the Problem.
This is the part of the paper that should be required reading for every executive team, every board, and every policy committee in the country.
The automation trap is not a story about corporate greed. It is not a story about bad actors making short-sighted decisions. Every company in the model is behaving rationally. Every individual decision is defensible. Every action is legal, often necessary, and frequently the right call for the organization making it.
The system-level outcome is a structural failure that no individual participant can see coming, because it is the product of individually rational choices aggregating into collective irrationality.
Falk and Tsoukalas model this as a Prisoner’s Dilemma operating at economic scale. In a classic Prisoner’s Dilemma, two parties both defect because defection is individually rational ‚Äî even though mutual cooperation would have produced a better outcome for both. The automation trap is the same logic applied across an entire economy.
Every company automates. Every company defects. The system loses.
The Solutions All Fail. Except One. And Nobody Is Implementing It.
The paper does not stop at the diagnosis. It evaluates every proposed solution.
Retraining programs. The math shows they cannot operate at the speed and scale required. Automation moves faster than workforce retraining can respond, particularly when multiple industries are automating simultaneously.
Universal Basic Income. Addresses consumption capacity but does not solve the structural misalignment between automation rates and human economic participation.
Voluntary corporate restraint. Individual companies that choose to slow automation simply lose competitive position to companies that do not. The rational actor problem reasserts itself immediately.
The only mechanism the paper identifies that could structurally address the trap is a Pigouvian automation tax — a levy on automation that internalizes the social cost of displacement, similar to how carbon taxes internalize the social cost of emissions. This could, in theory, slow the race to the bottom enough for the economy to adapt.
No major government is currently implementing one.
What This Means for Organizations Operating Right Now
The paper is not an argument against automation. That argument would be both futile and wrong — the companies that do not automate will simply be outcompeted by the companies that do.
What it is, is a precise description of the operating environment every organization is navigating in 2026. And that description has two urgent implications.
The first is economic. If the automation trap plays out at scale, consumer demand will contract in the markets where most businesses operate. The organizations that understand this are already thinking about what their customer base looks like in five years — not just what their cost structure looks like next quarter.
The second is human. The people in your organization who are watching automation eliminate roles around them are not paranoid. They are reading the system correctly. The question is whether you are giving them a reason to believe that building capability is worthwhile — or whether the signal they are getting is that no amount of development changes the math.
The Individual Response to a Structural Problem
Policy has not yet solved the automation trap. The paper is clear on that. And waiting for policy to solve it is not a strategy for the professionals and organizations navigating this reality today.
The individual response is the same one it has always been when structural forces are operating faster than institutional adaptation: build the capabilities that the structural forces cannot touch.
Ideas that emerge from genuine synthesis and lived experience. Judgment that operates in ambiguity. Relationships built on trust earned over time. Leadership that reads context, adapts in real time, and makes decisions when data is incomplete. The irreplaceable quality that makes what you and your organization do unmistakably yours.
These are not solutions to the macroeconomic problem the paper describes. The macroeconomic problem will require policy solutions that do not yet exist.
But they are the answer to the question every professional and every organization can actually do something about: what are we building that the automation trap cannot eliminate?
Those prepared need not fear the forces at work.
Distinct or Extinct is available now on Amazon. Download Chapter 1 free at realmikeevans.com
Take the Kryptonite Scorecard at realmikeevans.com/scorecard — the individual response to a structural problem policy has not yet solved.
Source: Falk, Brett Hemenway and Tsoukalas, Gerry. “The Automation Trap.” University of Pennsylvania (Wharton) and Boston University. arXiv:2603.20617v1 [econ.TH]. March 2, 2026. Peer-reviewed.