Why can't companies stop cutting jobs for AI? A new model says each firm keeps the savings and hands the damage to its rivals

Two economists have put the argument that AI layoffs are self-defeating into a formal model. In a preprint posted to arXiv on 21 March 2026 and revised on 3 June, Brett Hemenway Falk of the University of Pennsylvania and Gerry Tsoukalas of Boston University argue that a firm automating a task "captures the full cost saving from automation but bears only a fraction of the demand loss it creates in the product market; the rest falls on rivals". Because the cost is private and the damage is shared, they write, the arrangement "traps rational firms in an automation arms race, displacing workers well beyond what is collectively optimal" - and harms firm owners as well as workers. Knowing this, the paper argues, does not help: more competition and better AI make the excess larger, and wage adjustment, free entry, capital income taxes, worker equity, universal basic income, upskilling and private bargaining all fail to remove it. The authors conclude that only a Pigouvian automation tax, set against the demand loss per automated task, realigns the incentive. Two cautions belong with this. It is an arXiv preprint in theoretical economics, not peer-reviewed work, and it is a model rather than a measurement - it says nothing about how many jobs AI has actually taken. It matters because the policy conversation, including the US robot-tax bill this site has covered, has mostly been about cushioning the aftermath; this is an argument that the incentive to automate is itself the thing to tax.