Julien Renaux Warns of the AI Layoff Trap

Julien Renaux, a Software Engineer and Founder at FrunkFriends, recently posted something that made me stop scrolling: "We are doomed, people. Congratulations, we are officially automating our way to zero customers." He then pointed to a paper titled "The AI Layoff Trap" and called the current wave of automation-driven layoffs "a mathematical death spiral."

That framing is intentionally provocative, but it raises a serious question that business leaders, policymakers, and workers all need to grapple with: if every company races to cut labor costs by replacing people with AI, who is left to buy what those companies sell?

In this post, I want to expand on what Julien is getting at, translate the economics into plain English, and explore what an "automation tax" (specifically a Pigouvian tax) is trying to solve.

The idea behind "the AI layoff trap"

Julien summarized the paper with a simple chain reaction:

The race: companies replace human workers with AI to cut costs.
The income collapse: jobless humans lose income and stop buying stuff.
The demand cliff: consumer demand crashes.
The trap: profits tank, hurting companies too.
The only fix: an automation tax is the only mathematical cure.

Whether you agree with the conclusion or not, that causal loop is worth unpacking because it highlights something many "AI productivity" conversations skip: the macroeconomy is not just a set of individual company P and L statements. It is also aggregate demand, wages, and consumption.

Key insight: If automation increases total productive capacity but reduces wages and employment faster than prices fall, you can end up with more supply than paying demand.

Why individual incentives can create a collective problem

Julien anticipated the most common reaction: "A new tax?! That's socialism!" The pushback is predictable because each firm, acting alone, has a clear incentive:

If AI reduces your marginal cost, you deploy it.
If competitors deploy it, you must keep up.
If layoffs improve short-term metrics, the pressure to do them rises.

From inside any one boardroom, choosing not to automate can look like negligence.

The problem is that these decisions can create a negative spillover. When many firms cut payroll at the same time, the broader consumer base has less disposable income. That is not a problem any single firm can fully internalize, especially if it sells into a broad market.

A simple example

Imagine a city with 100,000 workers whose paychecks support restaurants, retail, housing, and subscriptions. If the major employers automate and lay off 20,000 people, you might see:

fewer restaurant visits
downgraded phone plans and cancelled streaming
delayed car purchases
more price sensitivity in everyday spending

Even companies that did not automate can get hit by lower demand. Meanwhile, the automating firms may also get hit because their customers are the same people whose income shrank.

This is the "trap" Julien is pointing to: the rational micro move (cut labor costs) can contribute to an irrational macro outcome (shrinking markets).

The "demand cliff" is not guaranteed, but it is plausible

It is important to be precise. Automation does not automatically cause an economic collapse. There are at least three reasons the outcome depends on timing and distribution:

Productivity gains can lower prices, letting consumers buy more with less.
New jobs and industries can emerge, absorbing displaced labor.
Profits from automation can be reinvested, creating additional demand.

But Julien's warning becomes more plausible when the speed of displacement exceeds the speed of adjustment.

If AI adoption is fast, layoffs are concentrated, and new job creation is slower (or requires very different skills), then aggregate wage income can drop before the rest of the economy has time to rebalance. In that window, demand can fall even as output capacity rises.

What is a Pigouvian tax, and why does Julien call it "the only math"?

Julien wrote that the only mechanism that stops the arms race is "a literal Pigouvian automation tax" and explained it as "a per-unit charge set equal to the marginal external cost." That is classic economics language, so here is the plain version.

A Pigouvian tax is designed for situations where private decisions impose costs on others that the decision-maker does not pay for directly.

Pollution is the standard example: a factory can dump waste cheaply, but society pays in health and cleanup.
The tax makes the factory face a cost closer to the true social cost.

Applied to automation, the argument is:

A company captures the benefits of replacing labor (lower costs, higher margins).
But society may bear some costs (higher unemployment, lower wages, lower demand, greater need for public support).
If companies do not pay for those external costs, they may automate more than is socially optimal, or automate too quickly.

So the tax is not framed as punishment. It is framed as alignment.

Julien's claim in one line: Make firms pay for the demand they destroy, and the private incentive to automate matches the social cost.

What would an automation tax actually look like?

This is where the debate gets real, because design is everything. "Automation" is not a single measurable unit like a ton of CO2. A workable policy would need a clear tax base.

Here are a few plausible approaches people discuss:

1) A tax tied to displaced labor

A fee when automation directly eliminates roles, potentially scaled by:

number of jobs displaced
wage level of displaced roles
local unemployment conditions

The upside is conceptual clarity: tax the event that creates the externality. The downside is measurement and gaming (reclassifying roles, slow-walking layoffs, outsourcing).

2) A tax on automated output or AI-driven value added

Instead of taxing layoffs, tax the extra output attributed to automation (similar to how VAT taxes value added).

Upside: it does not require proving a specific layoff. Downside: attributing value to AI vs other factors is messy.

3) A payroll tax shift (tax capital more, labor less)

Rather than a new category, adjust existing tax structures so that funding the social safety net is not overly dependent on payroll taxes when payroll shrinks.

Upside: administratively simpler. Downside: less direct connection to the "automation externality" story.

Where should the money go?

If the goal is to prevent the "zero customers" scenario, the use of proceeds matters as much as the tax itself. Options include:

wage insurance or temporary income support for displaced workers
large-scale retraining and apprenticeship pipelines
incentives for companies that create net new jobs
broad demand support mechanisms (for example, cash transfers or expanded tax credits)

In other words, a tax alone does not solve the problem. The tax is a brake and the redistribution is the steering.

The leadership takeaway: optimize for markets, not just margins

What I appreciate about Julien's post is that it challenges a narrow definition of "efficiency." Cutting costs is not the same as building a healthy market.

For executives, there are a few practical questions hiding inside this macroeconomic argument:

If you automate a function, do you have a plan to redeploy talent, not just remove it?
Are you measuring customer health (retention, willingness to pay, budget growth) alongside cost savings?
Are you assuming demand is fixed, or are you actively modeling how employment trends affect your total addressable market?

Even if an automation tax never materializes, the "AI layoff trap" is a reminder that companies ultimately depend on customers with purchasing power.

A balanced conclusion

Julien Renaux used dramatic language to make a point: the automation race can become self-defeating if it collapses the income base that supports demand. The proposed fix, a Pigouvian automation tax, is an attempt to correct a misalignment between private incentives and social outcomes.

You do not have to agree that a tax is "the only" solution to take the warning seriously. But if we keep treating AI adoption as a purely internal efficiency play, we risk missing the bigger system dynamics. In the long run, the best businesses are not just the most automated. They are the ones that can thrive in an economy where people still have money to spend.

This blog post expands on a viral LinkedIn post by Julien Renaux, Software Engineer - Founder at FrunkFriends. View the original LinkedIn post →