Artificial intelligence allows companies to replace expensive human workers with low-cost software, but this efficiency comes with a hidden economic cost that could lead to a systemic collapse of consumer demand. Research by Brett Hemenway Falk of the University of Pennsylvania and Gerry Tsoukalas of Boston University demonstrates that while individual layoffs save a single firm money, the collective loss of wages across an industry reduces the total pool of money consumers have to spend. Because each firm captures the full benefit of its own cost-cutting but shares the resulting demand destruction with its rivals.
Rational CEOs are currently trapped in an ‘automation arms race’ where they must fire workers to remain competitive. This is even if they know it is collectively self-destructive for the industry as a whole.
3 Key Takeaways
- Foresight cannot stop the automation arms race. Even when firms clearly see that mass layoffs will eventually destroy the purchasing power of their customers, competitive pressure forces them to automate anyway to avoid being undercut by rivals who are doing the same.
- The ‘Red Queen Effect’ means better AI makes the trap deeper. As AI becomes more productive and cheaper, firms perceive even greater market-share gains from over-automating, which intensifies the destruction of consumer demand without providing a lasting profit advantage to any single company.
- A Pigouvian automation tax is the only corrective tool. Standard solutions like Universal Basic Income (UBI), retraining, or capital taxes do not change the underlying competitive incentive to automate; only a per-task tax equal to the uninternalized demand loss can align corporate actions with economic stability.
Read the AI research paper here: The AI Layoff Trap
What is the AI Layoff Trap and how does it create a market failure?
The AI layoff trap is a specific type of market failure where firms automate tasks to save internal costs. But in doing so, they inadvertently eliminate the very people who buy their products.
This phenomenon is driven by a demand externality, which occurs when one company’s private decision to cut staff imposes a broader economic cost on the entire sector by reducing total consumer spending.
In a competitive economy, a firm that automates a task with AI captures 100% of the labor savings. But because revenue is distributed across many firms, that same firm only suffers a tiny fraction of the resulting drop in aggregate demand.
This imbalance creates a structural incentive for firms to automate far beyond what is collectively optimal. If a single firm chooses to keep its human workers to help support the economy, it remains the high-cost producer. While its competitors use AI to lower prices and take its market share. Consequently, every firm is forced to adopt the same aggressive automation strategy to survive. This leads to a ‘race to the demand cliff’ where total industry revenue collapses as the workforce—which also functions as the consumer base—loses its income.
| Model Parameter | Definition | Economic Role |
| Automation Rate (αi) | The fraction of the workforce a firm replaces with AI. | Determines both cost savings and demand destruction. |
| Cost Saving (s) | The difference between human wages and AI operating costs (w – c). | The primary driver of the private incentive to automate. |
| Demand Loss (l) | The spending reduction caused by a displaced worker. | Calculated as ℓ := λ(1 − η)w, is the propensity to consume. |
| Friction (k) | The physical or organizational cost of integrating AI. | Prevents instant total automation; acts as a temporary brake. |
Why is AI automation considered a Prisoner’s Dilemma for CEOs?
AI automation functions as a Prisoner’s Dilemma because it is individually rational for every CEO to fire workers, even though the total result is a lower profit for every firm involved. In this scenario, mutual restraint—where all firms agree to limit layoffs, would lead to the highest total profit because consumer spending would remain strong. However, as long as any one firm can gain an advantage by ‘defecting’ and automating, no agreement is stable.
The research proves that automation is a ‘strictly dominant strategy’ in competitive environments. If your rivals choose to automate, you must automate to avoid being priced out of the market. If your rivals choose to keep their human workers, you want to automate to gain a massive cost advantage and dominate the sector. Because every firm follows this logic, they all automate, leading to a state where everyone has lower costs but no one has customers. This is why simple foresight is not a brake on the process; even if a CEO sees the cliff ahead, they cannot stop the car if they are being chased by competitors.
How does more competition increase the severity of the layoff trap?
Competition acts as a magnifying glass for the over-automation problem. It makes the gap between what is good for a company and what is good for the economy much wider. In a monopoly, a single firm owns the entire market. Therefore, it feels the full impact of the demand destruction it causes. A monopolist would internalize the demand loss and only automate to the point where the cost savings exactly match the lost revenue.
However, in a fragmented market with many competitors (N ≥ 2), each firm only feels 1/N of the demand loss from its own layoffs while its competitors feel the other (N-1)/N portion. As the number of firms in a market increases, each individual CEO perceives less and less of the damage they are doing to the collective purchasing power. This dilution of responsibility means that highly competitive sectors—like retail or generic software services—are likely to hit the demand cliff much faster and harder than sectors dominated by a few large players.
| Market Structure | Internalization of Demand Loss | Degree of Over-Automation |
| Monopoly (N=1) | 100% (Full Internalization) | None (Efficient Automation) |
| Duopoly (N=2) | 50% Internalized | Moderate Over-Automation |
| Perfect Competition (N → ∞) | 0% Internalized | Maximum Over-Automation |
Why does better AI actually make the economic outcome worse?
The researchers identify a phenomenon they call the ‘Red Queen Effect’.
Here higher AI productivity actually amplifies the economic distortion rather than solving it. In a standard productivity model, better tools lead to more wealth. However, in the AI layoff trap, better tools simply increase the speed of the arms race. When AI becomes more capable (ϕ > 1), a firm perceives an even larger market-share gain from automating more tasks than its rivals.
This market-share motive adds a second layer of distortion on top of the baseline demand externality. Every firm races to capture more of the market by deploying more powerful AI. But at the symmetric equilibrium, where everyone has adopted the same “better” AI, these gains cancel each other out.
The firms end up with the same relative market shares they started with, but because they have now automated more tasks, the total pool of consumer demand has shrunk even further.
‘Better’ AI effectively makes firms run faster just to stay in the same place, while the ground of consumer spending erodes beneath them.
Why do standard policies like Universal Basic Income and Profit Taxes fail to stop the AI layoff trap?
Universal Basic Income (UBI) and capital income taxes are ineffective because they address the symptoms of displacement rather than the competitive incentive to automate.
UBI provides a financial floor for workers. This is beneficial for human welfare, but it does not change the marginal calculation a CEO makes. If a robot task costs c and a human task costs w, the CEO will still choose the robot as long as c < w. This is regardless of whether the fired worker is getting a UBI check from the government. UBI changes the level of profit but not the ‘payoff differences’ that drive strategic behavior.
Capital income taxes (taxes on total profit) suffer from the same mathematical failure. Because a proportional tax ‘t’ on profits scales the entire profit function equally, it effectively cancels out of the firm’s optimality condition.
A CEO deciding whether to automate one more task will find that the tax does not change the fact that the robot remains cheaper than the human at the margin.
While these policies can redistribute wealth, they cannot stop firms from racing toward the demand cliff because they do not penalize the act of replacing a worker with a machine.
| Policy Instrument | Mechanism of Action | Effectiveness on the “Trap” |
| Universal Basic Income | Adds to autonomous demand (A). | Ineffective; leaves incentives unchanged. |
| Capital Income Tax | Reduces retained earnings by (1-t). | Ineffective; cancels out of the decision margin. |
| Upskilling | Increases income replacement (η). | Partially effective; narrows the distortion. |
| Worker Equity | Recycles profits to workers. | Partially effective; limited by spending leakage. |
| Pigouvian Tax | Taxes the act of automation (τ). | Fully effective; corrects the externality. |
What is a Pigouvian Automation Tax and why is it the only solution?
A Pigouvian automation tax is a per-unit charge set equal to the marginal external cost of an automated task. This makes the firm ‘pay’ for the demand destruction it would otherwise ignore.
The research argues that this is the only evaluated instrument capable of aligning individual corporate incentives with the cooperative optimum of the industry. By charging the firm for the spending power its layoffs destroy, the government forces the CEO to account for the health of the entire sector’s revenue base.
The optimal tax rate is roughly equal to the portion of the demand loss that the firm externalizes onto its rivals. Mathematically, this rate is defined as:
τ∗ = ℓ(1 −1/N), where ℓ = λ(1 − η)w is the demand loss per displaced worker.
This tax has a unique self-limiting property: if the tax revenue is recycled into retraining programs that help workers find high-paying new jobs, the income replacement rate (η) will rise. As η approaches 1, the demand loss disappears, and the required tax rate automatically shrinks to zero. The tax is not meant to stop progress, but to act as a ‘speed governor’ that ensures automation does not outpace the economy’s ability to reabsorb workers.
How are real-world layoffs at Block and Salesforce illustrating this trap?
Current events in the technology sector provide an empirical signature of the AI layoff trap. It shows firms reporting high profits while simultaneously slashing their workforces.
In February 2026, Block (the parent company of Square and Cash App) announced it was cutting nearly half of its 10,000 employees.
It reasoned that AI had rendered many roles unnecessary. CEO Jack Dorsey warned that ‘within the next year, the majority of companies will reach the same conclusion’.
“Intelligence tools have changed what it means to build and run a company… A significantly smaller team, using the tools we’re building, can do more and do it better. And intelligence tool capabilities are compounding faster every week.” — Jack Dorsey, Block CEO
This signals the start of the massive automation wave predicted by the researchers.
Salesforce has followed a similar path. They cut 4,000 customer-support agents—reducing that department from 9,000 to 5,000 ‘heads. This was done to lean on its autonomous ‘Agentforce’ platform.
“I was able to rebalance my headcount on my support. I’ve reduced it from 9,000 heads to about 5,000, because I need less heads.” — Marc Benioff, Salesforce CEO
While Wall Street initially rewarded these moves with stock price surges, the research warns that these gains are deceptive. If every major tech company ‘rebalances’ its headcount this way, the total purchasing power of the consumer base will collapse. This will lead to a long-term profit erosion that no amount of internal efficiency can fix.
| Case Study | Action Taken | CEO Statement/Rationale | Economic Implication |
| Block Inc. | Cut 4,000 jobs (40%) | “AI made many roles unnecessary” | First major CEO “telling the real story”. |
| Salesforce | Cut 4,000 agents | “I need less heads” | Replaced human touch with Agentforce bots. |
| Cognition | Deployed “Devin” AI | 1 engineer does work of 5 people | Massive compression of the technical labor pool. |
| Snap | Cut 1,000 jobs (16%) | Use AI to “increase velocity” | Prioritizing savings over repetitive work. |
Why can’t endogenous wage adjustments fix the over-automation problem?
Traditional economic models suggest that automation is self-correcting because as people lose jobs, wages fall. This eventually makes hiring humans cheaper than using machines. The Falk-Tsoukalas model shows that this ‘wage channel’ can raise the threshold at which firms start to automate. But it cannot close the gap between individual and collective rationality once the process begins. Wage adjustment changes when the problem ‘bites,’ but not whether it exists.
Furthermore, resolving the trap through wage depression is a ‘Pyrrhic victory’ for the economy. If human wages are driven down to the level of cheap electricity for a robot, workers may keep their jobs. But they will no longer earn enough income to function as consumers. The economy’s aggregate purchasing power collapses through wage depression just as surely as it would through unemployment. A labor market that ‘self-corrects’ only by impoverishing its entire workforce has not solved the AI layoff trap; it has simply replaced mass unemployment with mass poverty, leaving the demand cliff unchanged.
How does the trap impact entry-level white-collar roles and career progression?
The current wave of AI automation is uniquely dangerous because it is ‘swallowing the junior layer whole’. This destroys the learned judgment necessary for senior roles.
This ‘Junior Role Erasure’ means that the messy, low-level tasks that once allowed young professionals to learn how a business works are now handled by autonomous agents.
“My prediction for 50% of entry level white collar jobs being disrupted is 1–5 years, even though I suspect we’ll have powerful AI… in much less than 5 years.” — Dario Amodei, Anthropic CEO
When firms treat automation as a way to skip the ‘learning phase’ of a career. They are effectively liquidating their future human capital for short-term quarterly gains. The loss of these entry-level ‘stepping stones’ creates a long-term risk for firm owners that is not captured in immediate cost savings. Without a pipeline of humans who have learned the basics of the system, there will eventually be no one left with the expertise to supervise the AI or to take on complex leadership roles that require human intuition.
Action points — how to use this information
For Policymakers: Shift to Proactive Intervention
- Evaluate the Pigouvian Automation Tax: Legislators should consider a per-task tax on AI labor replacement to align firm incentives with long-term economic demand.
- Prioritize Retraining over Transfers: Direct tax revenue toward high-intensity human capital development to increase the income replacement rate (η) and shrink the externality over time.
- Address Competitive Incentives Directly: Move beyond cushioning the aftermath of layoffs and focus on the structural ‘automation arms race’ that drives them.
For Firms: Recognize the Long-Term Risk of Efficiency
- Internalize the Red Queen Risk: Understand that automating faster than rivals provides only a temporary advantage that likely reduces collective industry profits in the long run.
- Focus on ‘Elevation Space’: Before cutting headcount, look for ways to transition workers into roles involving judgment, coordination, and trust—tasks where AI cannot easily replace human value.
- Support Industry Standards for AI Deployment: While voluntary deals are hard, participating in sector-wide guidelines for AI transition can help mitigate the most aggressive demand-destructive behaviors.
For Workers: Protect Your Economic Floor
- Seek Roles with High ‘Learning Capacity’: In an age of junior role erasure, prioritize positions that offer mentorship and on-the-job training that cannot be automated.
- Develop Specific, Not General, AI Skills: Generic AI certifications are losing value; focus on specific applications like AI-augmented threat detection or complex product design.
- Advocate for Co-determination: Support labor initiatives that give employees a voice in how AI is integrated into the workplace, shifting the focus from displacement to augmentation.
FAQs on the AI Layoff Trap
1. Is the AI layoff trap just another way of saying ‘jobs are being lost’?
No. It is a specific economic failure where the rate of job loss destroys the customer base that companies need to buy their products, leading to lower profits for the owners themselves.
2. Why don’t smart CEOs just stop the layoffs if they see the problem?
They are stuck in a Prisoner’s Dilemma. If one CEO stops while everyone else keeps firing, the rivals will have much lower costs and take over the entire market.
3. Does this happen because AI is bad at its job?
Actually, the trap happens because AI is too good. The more productive and cheaper the AI, the faster companies feel they must fire workers to stay ahead of the competition (the Red Queen Effect).
4. Won’t new jobs be created to replace the ones lost to AI?
Historically, yes (the reinstatement effect), but the paper warns that AI is different because it can often perform the ‘new’ jobs as soon as they are created, and displacement is currently happening much faster than new work is appearing.
5. How much of the average person’s job is at risk from this?
Estimates suggest that roughly 80% of U.S. workers hold jobs with tasks that could be handled by large language models, especially in offices and customer service.
6. Does Universal Basic Income (UBI) solve the trap?
No. UBI helps people survive, but it doesn’t change the fact that a robot remains cheaper than a human for a company. Firms will keep firing people as long as the machine is cheaper.
7. Why is a “robot tax” better than taxing a company’s total profits?
A profit tax only applies to whatever money is left at the end of the year. A ‘robot tax’ (Pigouvian tax) applies to the act of firing a person, which directly changes the CEO’s decision-making process.
8. Can companies just agree among themselves to slow down?
No, because “cheating” on the agreement is too profitable. Since automation is a dominant strategy, voluntary deals usually collapse as companies try to get a cost edge on each other.
9. What happened at Block in early 2026?
Block cut nearly 40% of its staff citing AI efficiency. Even though its stock went up, the model warns that this move helps Block’s short-term costs while hurting the long-term economy.
10. What is ‘Agentic AI’?
Agentic AI refers to autonomous systems that can handle entire workflows—like customer service or coding—without a human watching them every second.
11. Why do stock prices go up after a company fires people?
Wall Street looks at the ‘private savings’ for that one company but ignores the ‘social damage’ to the whole economy. The model shows these stock gains are often temporary and deceptive.
12. Is the trap worse for big companies or small ones?
It is worse in markets with many competitors. A single big monopoly feels all the pain it causes, but in a crowded market, each firm thinks it can fire people without hurting itself.
13. What is ‘MPC asymmetry’?
It means workers spend a much larger part of their check on the sector’s goods than firm owners do. When money moves from worker wages to owner profits, total spending in that sector drops.
14. Which roles are vanishing the fastest?
Displacement is currently concentrated in customer support, back-office operations, entry-level software coding, and middle-management roles.
15. Is there a natural limit to how much AI can replace people?
The only natural limit is if computing power (the “cost” of AI) becomes more expensive than hiring people. Without government intervention, the race continues until it hits that physical cost wall.
Further reading — 5 resources to learn more about the AI layoff trap
- Acemoglu, D., & Restrepo, P. (2018): “The Race between Man and Machine.” Foundational work on how automation and task creation interact in the labor market.
- Beraja, M., & Zorzi, N. (2025): “Inefficient Automation.” Analysis of why firms automate too quickly when workers face financial constraints during job changes.
- Brynjolfsson, E., et al. (2025): “Generative AI at Work.” Evidence on the productivity gains of AI and its early impact on entry-level career pipelines.
- Autor, D., et al. (2024): “New Frontiers.” A 40-year study showing that current technology is destroying jobs faster than it creates “new work” titles.
- Costinot, A., & Werning, I. (2023): “Robots, Trade, and Luddism.” Research on using corrective taxation to protect the economy during technological shifts.
Check out more AI research paper explainers on Applied AI Tools:
- Anthropic AI Fluency Index: Why Polished Outputs Reduce Critical Thinking
- AI Energy Costs 79x More For Reasoning: Princeton Researchers
- How Recursive Language Models Solve LLM Context Rot Issue [No Jargon Explainer!]
- OpenAI’s Why Language Models Hallucinate AI Research Paper Explained
- Microsoft AI-Safe Jobs Study Explained: Use Insights to AI-Proof Career in 2025
- RAG in Healthcare: Real Adoption Use Case Examples
- MIT ChatGPT Brain Study: Explained + Use AI Without Losing Critical Thinking
Twice a month, we share AppliedAI Trends newsletter.
Get SHORT AND ACTIONABLE REPORTS on AI Trends across new AI tools launched and jobs affected due to AI tools. Explore new business opportunities due to AI technology breakthroughs. This includes links to top articles you should not miss, like this ChatGPT hack tutorial you just read.
Subscribe to get AppliedAI Trends newsletter – twice a month, no fluff, only actionable insights on AI trends:
You can access past AppliedAI Trends newsletter here:
This blog post is written using resources of Merrative. We are a publishing talent marketplace that helps you create publications and content libraries.
Get in touch if you would like to create a content library like ours. We specialize in the niche of Applied AI, Technology, Machine Learning, or Data Science.
Leave a Reply