Findings suggest policy should address not only the aftermath of AI labor displacement, but also the competitive incentives that drive it

A new study reveals that if AI-driven layoffs outpace the economy’s ability to reabsorb workers, the lost paychecks would erode the consumer demand every firm depends on, and that knowing this will not be enough to stop companies from cutting.

The paper, The AI Layoff Trap, co-authored by Gerry Tsoukalas of Boston University's Questrom School of Business and Brett Hemenway Falk of the University of Pennsylvania, finds that firms do not pay for the economic damage their layoffs cause. A firm keeps the full savings when it replaces a worker with AI while bearing almost none of the cost. Instead, that cost falls on the market, as the worker's lost income reduces spending across the whole economy.

“When a firm lays off workers to cut costs, those savings go straight to its bottom line, but the lost demand those workers represent gets spread thin across the entire economy," says Tsoukalas. "Every firm feels a tiny pinch, but no single firm feels the full hit of its own layoffs, so everyone keeps cutting — and that pull only grows stronger as AI gets cheaper and more capable. Competition creates an automation arms race that no amount of individual foresight can prevent.”

Several findings stand out. The over-automation is not simply a case of workers’ losses becoming owners’ gains; both workers and company owners end up worse off than they would if firms had collectively shown restraint. The effect is also strongest in the most competitive markets, where each firm bears the smallest share of the demand it removes, while a monopolist automates closest to the efficient level. Additionally, the forces usually expected to restore balance — including flexible wages, the entry of new firms, and owners spending their profits — impact when the problem appears but do not remove it.

The paper also tests six widely discussed policy responses, finding only one that truly works. Universal basic income raises living standards but leaves every firm’s incentive to automate unchanged. Capital income taxes do not affect the automation decision, and worker profit-sharing falls short for similar reasons. Voluntary agreements among firms are unstable, because each firm’s best move is to automate regardless of what others do. The one instrument that fully corrects the distortion is a Pigouvian tax on automation, a per-task charge set equal to the demand a firm removes from others, whose revenue can fund retraining that reduces the problem over time.

“Out of six popular policy fixes, we find that five of them fail,” says Tsoukalas. “In our model, only a tax on automation itself actually changes the calculus. Most of the policy debate assumes displacement will happen and focuses on picking up the pieces. That’s all necessary, but none of it deals with firms’ automation incentives.”

The authors are upfront about what the result depends on. It rests on the assumption that displaced workers end up in lower-paid work rather than better jobs. If they are reabsorbed into higher-paying roles quickly enough, the result flips, and firms would automate too slowly rather than too much. The model treats the over-automation regime as the relevant case because AI, unlike earlier tools, may adapt to newly created jobs faster than displaced workers can move into them, cutting off the reabsorption that followed earlier waves of automation.

The model is also deliberately simple, built on a single sector and identical firms. The authors treat that simplicity as conservative: a multi-sector economy, where layoffs in one industry cut spending on every other, would point to a larger problem, not a smaller one. At the same time, detecting the effect would take displacement at a scale beyond what has occurred so far, so the paper identifies a structural vulnerability rather than an active crisis. That restraint also shapes how it meets the public worry about an AI-driven demand "death spiral": the model supplies a mechanism, not a verdict. It suggests the concern is not baseless, while recasting it as a market failure rooted in competitive incentives, the kind a tax could correct in principle, rather than the unstoppable process the popular framing assumes.

Tsoukalas and Hemenway Falk argue the findings should reframe a policy debate that today focuses largely on cushioning the aftermath of displacement. Because a distinct market failure requires a distinct tool, they conclude that policy must also address the competitive incentives driving over-automation in the first place. They caution that a tax adopted by one country alone could push automation offshore, which strengthens the case for international coordination.

About Gerry Tsoukalas

Gerry Tsoukalas is a professor in the Information Systems Department at Boston University Questrom School of Business. Selected for the Thinkers50 Radar, Gerry Tsoukalas is recognized among emerging global business thinkers shaping the future of management and technology. Professor Tsoukalas specializes in Digital Platforms, AI and Machine Learning, and is co-founder of the Crypto and Blockchain Economics Research Forum (CBER).

About Brett Hemenway Falk

Brett Hemenway Falk is a research professor in the Department of Computer and Information Sciences at the University of Pennsylvania. He is the director of the Crypto and Society Lab, which focuses on privacy and security in digital environments, as well as facilitating transparency and trust.

About Boston University Questrom School of Business

Founded in 1913, the Boston University Questrom School of Business is a global top-tier academic research business school. Led by Allen Questrom Professor and Dean, Susan Fournier, Questrom develops business leaders who create value for the world. Questrom redefines transformational business programs, strengthens partnerships with the business community, advances the impact of research on business, and manages the school as a high-performing enterprise committed to excellence with a service mindset. Comprising a renowned full-time faculty of 165 researchers, teaching faculty, and accomplished practitioners, Questrom generates insights to address today’s business challenges and prepares students with the tools they need to succeed from Day 1 in their professional lives. Questrom’s portfolio of academic programs is robust and includes a Top 20 undergraduate program of over 2,200 students; distinctive MBA offerings including 900 students in a full- and part-time MBA, the affordable Online MBA and specialty MBAs in social impact, health, and digital technology; several thriving specialized master’s programs in areas including business analytics, mathematical finance, and management studies; and a rigorous PhD program. More than 50,000 Questrom alumni form a powerful global network of leaders driving value creation that changes the world.

View source version on businesswire.com: https://www.businesswire.com/news/home/20260610408944/en/