Hidden cost of AI layoffs: Why firms may lose in the long run

The rush to replace human labor may create unintended economic consequences not just for workers, but for firms themselves. A new study finds that companies adopting AI-driven automation may collectively undermine demand in the economy, creating a self-defeating cycle that reduces profits even as productivity rises.

The study, titled “The AI Layoff Trap,” presents a formal economic model explaining how firms’ incentives to automate labor can generate a broader demand collapse. By focusing on task-based production and competitive market dynamics, the research shows that while individual firms benefit from cutting labor costs, the aggregate effect of widespread layoffs reduces consumer purchasing power, ultimately feeding back into weaker demand for goods and services.

The findings challenge the dominant narrative that automation is unambiguously beneficial for firms and economic growth. Instead, the study argues that under certain conditions, the expansion of AI can lead to a structural trap in which firms continue to automate even when it becomes collectively harmful.

Automation incentives drive firms into a collective demand trap

Firms make automation decisions based on private gains, not on the broader economic consequences of those decisions. When a company replaces workers with AI, it reduces labor costs and improves efficiency. From the perspective of that firm, automation is a rational and profitable choice.

However, when many firms make the same decision, the effects extend beyond individual balance sheets. Workers who lose their jobs or face reduced wages have less income to spend. This decline in purchasing power reduces demand across the economy, affecting the revenues of all firms, including those that automated.

The study models this dynamic as a demand externality. Each firm captures the cost savings from automation but bears only a fraction of the resulting decline in aggregate demand. Because firms do not internalize this broader impact, they continue to automate even when the overall effect is negative.

This creates what the researchers describe as an automation trap. Firms are locked into a competitive cycle in which failing to adopt AI puts them at a disadvantage, but widespread adoption leads to lower demand and reduced profitability. The result is a misalignment between individual incentives and collective outcomes.

The analysis shows that this trap is not a marginal effect but a structural feature of competitive markets under automation. It emerges from the interaction between task-based production, wage dynamics, and demand formation, suggesting that the problem is embedded in the logic of modern economic systems rather than in specific industries or technologies.

Stronger AI and more competition can worsen the problem

As opposed to expectations that better technology leads to better outcomes, the study finds that advances in AI can intensify the automation trap. As AI becomes more capable and cost-effective, it becomes easier for firms to substitute human labor across a wider range of tasks.

This expansion increases the scale of job displacement and amplifies the reduction in aggregate demand. In turn, firms face even greater pressure to cut costs, reinforcing the cycle of automation. The model suggests that improvements in AI do not necessarily lead to balanced growth but can instead deepen the underlying distortion.

Market competition plays a similar role. In highly competitive environments, firms have strong incentives to adopt cost-saving technologies to maintain margins. Even if managers recognize the broader risks of automation, competitive pressures limit their ability to act differently.

The study shows that as competition intensifies, the equilibrium level of automation increases, even when it reduces overall welfare. Firms are effectively pushed into adopting AI not because it is socially optimal, but because it is necessary to survive in the market.

This dynamic challenges the assumption that competition always leads to efficient outcomes. In the context of AI-driven automation, competition can exacerbate inefficiencies by accelerating the adoption of technologies that reduce aggregate demand.

The research also highlights the role of wages in this process. While lower wages can partially offset demand losses by reducing production costs, they do not fully compensate for the decline in consumer spending. As a result, wage adjustments alone are insufficient to restore equilibrium.

Common policy responses fall short of addressing the core distortion

The study evaluates several policy interventions that are often proposed to address the impact of automation, including wage subsidies, unemployment support, and redistribution mechanisms. While these measures can mitigate some of the effects of job displacement, the analysis finds that they do not fully resolve the underlying demand externality.

Because the core issue lies in firms’ incentives to automate, policies that focus only on workers do not address the root cause of the problem. As long as firms continue to capture private gains from automation while externalizing demand losses, the automation trap persists.

The researchers identify a Pigouvian automation tax as the only policy in their framework that fully corrects the distortion. By imposing a cost on automation that reflects its broader economic impact, such a tax aligns private incentives with social outcomes.

Under this approach, firms would still adopt AI where it generates genuine productivity gains, but the excessive level of automation driven by competitive pressures would be reduced. The tax effectively internalizes the demand externality, encouraging firms to consider the broader consequences of their decisions.

The study acknowledges that implementing such a policy raises practical and political challenges, including how to measure the impact of automation and how to design a tax that does not stifle innovation. However, it argues that without addressing the incentive structure directly, other interventions are unlikely to achieve lasting solutions.