The Automation Trap Is Real. A Tax Won't Be Enough

Two economists have now shown, in a formal model, how rational firms can automate beyond what the economy can bear even when everyone can see the damage coming.

The paper is The AI Layoff Trap, by Brett Hemenway Falk and Gerry Tsoukalas, at the University of Pennsylvania and Boston University.

It is not a manifesto. It is a model. And it arrives at a conclusion the Extractive Economy has spent decades arranging not to see.

The mechanism is simple enough to state in one breath.

A firm that replaces its workers with AI keeps the full saving. But the workers it displaces were also customers, theirs and everyone else's. When their income disappears, demand contracts across the whole sector. The firm that automated bears only a sliver of that loss. The rest lands on its rivals.

Once the private saving exceeds the fragment of demand loss each firm bears, every firm faces the same move: automate. Not out of greed. Out of arithmetic.

Holding back means eating the demand collapse your competitors cause while forgoing the savings that would offset it. Automating is the dominant move no matter what anyone else does. The result is a race that overshoots the collectively efficient level and, in the severe case, leaves workers and owners worse off than restraint would have.

This is not a transfer from labor to capital. It is destruction on both sides of the table.

What they ruled out

The interesting part is not the trap. It is the list of things that do not defuse it.

The authors run the standard toolkit against the problem, one instrument at a time.

Universal basic income raises the floor on living standards but leaves every firm's incentive to automate exactly where it was.

A tax on capital income scales profits up or down without touching the margin where the damage happens.

Giving workers equity narrows the gap but cannot close it.

Firms bargaining with each other to slow down cannot hold the line, because defection is always individually rational and never observable in time to punish.

Every familiar answer operates on the wrong surface.

One instrument survives their analysis: a tax levied directly on each act of automation, set equal to the demand each firm destroys for everyone else. A Pigouvian tax. The textbook remedy for a textbook externality.

That is where the paper ends and the real question begins.

The problem with the only answer

A tax that corrects this externality has to do three things at once.

It has to set a rate calibrated to a sector-wide quantity. It has to observe automation firm by firm, inside organizations that have every reason to obscure it. And it has to hold across borders because, as the authors themselves concede, a tax applied in one place simply pushes the automation somewhere else. Their own remedy requires multilateral coordination or border adjustments to function at all.

So the single working fix starts to require something close to global policy coordination: multilateral tax alignment, border adjustments, or cross-border enforcement capacity. It must also rely on states that can audit internal corporate decisions and maintain a coordinated rate across jurisdictions.

That authority does not exist. It has never reliably existed, and the history of assembling one for any externality that crosses borders is not encouraging.

There is a second problem, quieter than the first. Even where the authority exists, the revenue flows through a government that then decides how to return it to the people whose income was destroyed. A tax can be calibrated with precision; its proceeds rarely are. Bind the revenue to the harm — paid to the displaced, capped at what the levy raises — and it becomes a real corrective. Let it disappear into general spending and it is just a tax with a story attached.

None of this makes the tax wrong. It makes it external. A Pigouvian tax prices the leak; it does not close it. The correction is applied from outside the productive structure, after the damage is already moving. Which leaves standing the deeper question the model never asks: could the structure be built so the value never leaked at all?

It was never a productivity problem

Strip the model down and the externality rests on a single fact: when a worker is displaced, their lost spending leaves the system and does not come back.

The authors carry a parameter for exactly this : The fraction of displaced income that returns through reemployment, transfers, or anything else. When that fraction is below one, the trap is open. The lower it goes, the wider the trap gets.

That is not a productivity problem. It is not even a coordination problem.

It is a circulation problem.

This is the flaw the Extractive Economy never had to face because labor closed the loop by force. Production paid wages. Wages became spending. Spending sustained production. Value returned by dependency, not design. AI removes that dependency while extraction remains. Value is still created broadly, captured narrowly, and allowed to leave.

Two economists modeling layoffs in 2026 have independently rederived the claim that systems collapse when value leaves faster than it returns. They reached it from competitive game theory rather than from first principles about circulation. They arrived at the same place.

The exit they encoded but did not name

The same parameter that opens the trap also closes it.

Push the fraction of returned income toward one, and the wedge shrinks. In the limit — displaced contribution reabsorbed at a value equal to or above what it replaced — the trap closes entirely, and can even invert into firms automating too slowly. History has rarely reached that limit. Displaced workers have mostly landed at lower wages, which is precisely why the trap is open. The point is not to assume reabsorption happens. It is that the rate of return is the lever — and it is the one variable an architecture can actually move.

The lever is not restraint. It is not regulation. It is return.

But the model is precise about one thing, and the precision is the whole game. Value returning through owners spending their profits cannot close the gap under any realistic assumption as owners would have to recycle more than they save. The return has to flow back through the people whose income was lost. Through contributors. Not through the people who captured the surplus.

That is the exact distinction the Commitment Economy was built on.

Infrastructure does what the tax was trying to do

The trap exists because value leaks out of the system and because the customer sits outside the firm, so the firm never feels the demand it destroys. Both of those are not laws of nature. They are properties of how the infrastructure is built. Change the infrastructure and the externality has nowhere to form.

Circulatory Finance is built-in return : value moving back through the people and assets that produced it instead of exiting through ownership. It is an attempt to move that return parameter structurally: not by hoping owners spend more, and not only by taxing firms after the leak, but by designing return paths through contributors, assets, and commitments from the start.

Sovereign Assets hold infrastructure in stewardship that cannot be captured, sold for extraction, or redirected for private yield. Value that enters cannot terminate in absentee ownership, because there is no owner to absorb it. And because they are bound to concrete needs rather than abstract yield, the value they hold has nowhere extractive to flow.

Collaboratives are the coordination unit that operates those assets, bound by commitment, accountable through participation. Not firms agreeing to automate less. Structures where the question of automating away your own demand does not arise, because the people and the value are not on opposite sides of a boundary.

Impact Certificates let capital enter without becoming ownership. Capital is remembered through a defined return path and fulfilled as value circulates back. It's the financial instrument that lets the loop close without anyone needing to own the asset for their contribution to count.

The Ledger of Consequence records the return: what was committed, what was produced, what continues to exist because of it.

None of this requires a tax authority. None of it requires observing a competitor's internal decisions or holding a coordinated rate across borders. It requires building systems where value was never going to leak in the first place.

The boundary the model assumes away

There is one more assumption holding the trap together, and it is the load-bearing one.

The externality only exists because the customer is outside the firm. The firm captures its saving and externalizes the demand loss precisely because the person whose spending vanishes is a stranger. Someone else's problem, a fraction diluted across rivals.

Collapse that boundary and the externality collapses with it.

This is easiest to see locally. The laid-off worker in a global market is a statistical stranger whose lost spending vanishes into the crowd; the laid-off worker down the road is the customer you can see. But locality is only the vivid case, not the mechanism. What closes the externality is the collapsed boundary itself, wherever it is collapsed.

When the buyer is also inside the loop, committing, contributing, carrying value forward rather than absorbing it at the end, the actor capturing the saving and the actor bearing the demand loss begin to converge. The model has a name for the case where a single actor feels the full consequence of its own automation: the monopolist, who internalizes everything and never over-automates.

The Commitment Economy reaches that same internalization without the monopoly. Not by concentrating control, but by removing the terminal node where value was supposed to die.

Two ways to make value return

Strip the argument to its frame and the tax and the architecture are after the same thing. Both target non-return. Both try to make value come back to the people the machine displaced. They differ only in method. The tax compels return from outside, after the leak. The architecture builds return in, so the leak never forms.

These are not rivals. They are instruments for different layers.

Where the market is planetary and the actors are beyond counting — frontier models, global compute, the platforms stacked on top of them — there may be no way to build the loop directly, and a correction priced from outside may be the only tool on the table. If it comes, it has to be bound to the harm and capped to it. Otherwise it stops being a correction and becomes another leak.

But most of the economy is not planetary. Most of it is housing and care and food and energy and the local infrastructure people actually live inside. At that layer return does not have to be compelled. It can be designed. Value that circulates by structure does not need to be taxed back into circulation, and a system with no terminal node has no externality to correct.

The mistake was never proposing a tax. The mistake is assuming return must always be forced from above, and waiting for an authority to force it.

Where the loop can be built, build it.

The authority isn't coming. At that layer, it doesn't need to.