Share this content:
Technological obsolescence is no longer just a discussion about efficiency. With the acceleration of automation and artificial intelligence, many companies are cutting costs at the expense of destroying some of the demand that sustains the market.
The problem is that none of them can stop on their own without losing competitiveness.
In February 2026, Block—Jack Dorsey's payments company—laid off nearly half its workforce: 4,000 people. The reason given was that AI had made those roles unnecessary. Dorsey added, "By 2027, most companies will reach that conclusion." It wasn't a threat. It was a description.
The question is: why do they do it if they know they're destroying their own demand? If laid-off workers are consumers, and if each round of layoffs reduces the purchasing power on which all businesses depend, shouldn't rational managers stop before they reach the precipice?
Brett Hemenway Falk and Gerry Tsoukalas, from the University of Pennsylvania and Boston University, have just answered that question with a rigorous mathematical model.
His conclusion:Knowing that the precipice exists is not enough to stop.Competition creates a trap that no company can escape alone.
"Each company captures 100% of the savings from replacing its workers, but only feels 1/N of the demand it destroys. The rest falls on its rivals."
It's a Prisoner's Dilemma On an industrial scale, the company that unilaterally slows down suffers the drop in demand caused by its competitors, but loses out on cost savings. The one that automates captures the savings and outsources the damage. The result: everyone automates more than would be collectively optimal, and both workers and business owners end up worse off than if they had cooperated.
It is not a transfer of value from one to another: it is pure destruction.
Global data confirms the scale of the problem. The WEF projects the destruction of 92 million jobs by 2030, and the creation of 170 million, with a net positive balance of 78 million. Goldman Sachs estimates that the impact on aggregate unemployment will be just 0.5 percentage points.
Good news.
The challenge is that job destruction is happening today, all at once, in specific sectors and for particular job profiles. Job creation will be gradual, will require different skills, and will come too late for those who have already lost their jobs. In the first half of 2025 alone, nearly 78,000 tech jobs in the United States were directly attributed to AI. Unemployment among young adults aged 20 to 30 in affected sectors rose 3 percentage points in a year.
The solution? The authors evaluate six policy instruments: universal basic income, capital taxes, employee stock ownership, collective bargaining, and retraining. None of them work. Only one corrects the distortion at the exact margin where the problem lies: a Pigouvian tax on automation, equivalent to the demand damage that each firm fails to internalize.
It's not populism. It's orthodox economic theory when it honestly confronts a market externality. The question isn't whether governments should intervene. It's whether they will do so before or after the damage is irreversible.
