The AI Layoff Trap - Why AI Layoffs Can Hurt Companies Too

Thu, 30 Apr 2026 09:00:00 +0000

Overview

The discussion around AI layoffs usually sounds very simple: companies use artificial intelligence, they automate tasks, and some employees lose their jobs. But the paper The AI Layoff Trap makes the problem more interesting and more dangerous.

The paper, written by Brett Hemenway Falk and Gerry Tsoukalas, argues that AI layoffs can become a trap even when companies are acting rationally. The problem is not only that workers may be displaced. The deeper problem is that workers are also customers. If many companies replace workers with AI at the same time, workers lose income, demand falls, and companies may damage the same market they depend on.

In other words: AI automation can reduce costs for one company, but if the whole market follows the same strategy too aggressively, everyone may end up with weaker demand and lower profits.

The arXiv version of the paper was submitted on March 21, 2026 as arXiv:2603.20617 under Theoretical Economics. The paper is 53 pages and includes 4 figures and 2 tables.

The Main Idea: Cost Savings Are Private, Demand Loss Is Shared

The core idea of the paper is a classic economic problem applied to the AI era.

When one firm automates a task with AI, it receives the full benefit of lower cost. That benefit is private. The firm sees it directly in its margins.

But when the worker who used to perform that task loses income, the loss of purchasing power is not felt only by the firm that automated. It spreads across the whole market. The worker may buy less from many companies, not only from the employer that replaced the job.

This creates what the paper calls a demand externality.

The firm captures the full cost saving, but it does not carry the full demand damage. That difference can push firms to automate more than what would be best for the market as a whole.

Why This Becomes an AI Automation Arms Race

Imagine a market with many companies. Each company knows AI can perform some tasks more cheaply than human workers. Each company also knows that competitors may use AI to lower prices or improve margins.

So every company faces the same pressure:

If we automate, we cut costs.
If we do not automate, competitors may become cheaper.
If everyone automates, the market may lose demand.
But if we wait while others automate, we may lose first.

This is why the paper describes the situation like an automation arms race. The company is not necessarily being irrational. In fact, the company may understand the macroeconomic risk. But competition still pushes it toward automation because not automating can be costly in the short run.

That is the trap: every firm has a rational individual reason to automate, but the collective result can be economically harmful.

How the Model Works

The authors build a task-based model. Each company has a set of tasks. Some tasks are performed by workers, and some can be automated with AI.

AI is cheaper than human labor in the model, but automation is not free. As the company automates more tasks, implementation becomes harder. Integration, management, quality control, process redesign, and technical friction all become more costly.

On the demand side, workers spend part of their wages on the sector’s products. Owners of firms spend less of their income inside the sector in the baseline model. This difference matters. If wages disappear and workers are not quickly reabsorbed into other jobs, total demand falls.

The result is clear:

In a competitive market, companies automate beyond the collectively optimal level because each company ignores part of the demand loss caused by worker displacement.

The Paper’s Most Important Results

The most important result is not simply that workers lose. The paper argues that excessive automation can hurt both workers and firm owners.

That is an important distinction. Many discussions frame AI layoffs as a distribution problem: workers lose wages, owners gain profits. This paper says the result can be worse than redistribution. If demand falls enough, owners can also lose because profits decline.

The paper also finds that the problem can become stronger when the market is more competitive. In a monopoly, one firm internalizes more of the demand loss because it owns the whole market. In a fragmented market, each firm pushes part of the demand damage onto competitors.

This means more competition can increase the incentive for excessive automation.

Why Better AI Does Not Automatically Fix the Problem

Many people assume that better AI always means a better economic outcome. The paper challenges that assumption.

If AI becomes cheaper and more capable, each company has an even stronger incentive to automate. A single company may gain a temporary advantage by adopting AI first. But when all companies adopt similar tools, the market-share advantage can disappear.

What remains is the wage displacement and the demand loss.

So better AI can create a paradox:

AI improves	Firm-level result	Market-level risk
Tasks become cheaper to automate	Stronger incentive to replace labor	More displaced income
Competitors adopt similar tools	Temporary advantage disappears	Demand loss remains
Automation becomes easier	Faster rollout	Less time for workers to transition

Why Common Solutions May Not Be Enough

The paper reviews several common responses to AI displacement. It does not say they are useless, but it argues that most of them do not fully solve the core incentive problem.

Upskilling and Retraining

Upskilling can help workers regain income. If displaced workers quickly move into new jobs, the demand externality becomes smaller. This is one of the most practical tools.

But retraining does not fully change the company’s automation decision. The firm may still automate if the private cost saving is large enough.

Universal Basic Income

Universal basic income can support people after displacement. It can protect living standards and stabilize consumption.

But the paper argues that UBI does not directly change the firm’s decision at the moment of automation. The company still compares the cost of human labor with the cost of AI.

Capital Income Taxes

Capital income taxes redistribute profits after they are created. But they do not directly target the task-level decision to replace a worker with AI.

Worker Equity

Worker ownership or equity participation can reduce the conflict between workers and owners. If workers own part of the firm, they share some of the gains.

But this only partially addresses the demand externality because the broader market effect still remains.

Bargaining Between Firms or Workers

The paper also discusses bargaining-style solutions. The problem is that each firm still has an incentive to deviate and automate more if it can gain a private advantage.

The Policy That Targets the Root Cause

The paper’s strongest policy conclusion is that a Pigouvian automation tax is the tool that directly targets the problem.

A Pigouvian tax is designed to make a private decision reflect its social cost. In this case, the tax would apply to automated tasks in a way that reflects the demand damage created by worker displacement.

This is not an argument to ban AI. It is not an argument to stop useful automation. It is an argument that if firms do not pay for the demand externality they create, the market can choose too much automation too quickly.

The revenue from such a tax could be used for retraining, income support, wage insurance, or transition programs. That matters because the paper’s problem becomes smaller when displaced workers recover income faster.

What Business Leaders Should Learn

For companies, the lesson is not “avoid AI.” That would be unrealistic and unhelpful.

The better lesson is:

Do not treat layoffs as the only measure of AI productivity.

A strong AI strategy should ask:

Which tasks should be automated?
Which roles should be augmented instead of eliminated?
How fast can workers move into new higher-value tasks?
Could aggressive layoffs reduce customer demand or brand trust?
Are short-term margin gains creating long-term market risk?

AI can increase productivity, improve service quality, reduce repetitive work, and create new business models. But if companies use AI only as a headcount reduction machine, they may create the exact demand problem described in the paper.

Practical Takeaway

The most useful idea from The AI Layoff Trap is that AI layoffs are not only a labor issue. They are also a demand issue, a competition issue, and a business strategy issue.

If AI replaces workers faster than the economy can absorb them into new income-generating work, the result may not be a clean productivity boom. It may be a weaker market where companies cut costs but also lose customers.

That is why this paper is important for anyone following AI layoffs, the future of work, automation policy, or the economics of generative AI.

The question is no longer only:

Will AI replace jobs?

The harder question is:

What happens when every company replaces jobs at the same time?

Source: The AI Layoff Trap, arXiv:2603.20617

Future of work | Mohamed Mohana