The Automation Tragedy: How the Free Market Is Overgrazing Aggregate Demand

In the first piece I described what economists like Anton Korinek and Joseph Stiglitz are arguing: cognitive work is now being bought as a depreciable capital asset rather than hired as a human trait. When a company moves coding, logistics or customer service onto AI systems, its operating margin widens the same quarter.

It is tempting to say the law forces executives to do this. It does not. American corporate law, Delaware’s included, asks directors to act in good faith for the long-term value of the corporation, and the business judgment rule gives them wide latitude in how. No statute makes a CEO fire a loyal staff to hit next quarter’s number. Lynn Stout spent a career demolishing that myth.

The market does the forcing instead. Investors reward margin. Activists hunt for companies that leave savings on the table. Executive pay is tied to the share price. And if one CEO declines to automate, the competitor across the street will, then use the savings to cut prices and take the customers. The pressure is not legal. It is competitive, and it is relentless.

That is a Nash equilibrium: a situation in which no single firm can improve its position by changing strategy alone. Automating is the best response to whatever everyone else does. So everyone automates, even though every firm might be better off in a world where everyone went slower. It is the prisoner’s dilemma with a payroll.

Here is where the tragedy takes hold. It is perfectly rational for one firm to replace workers to protect its margin. When every firm runs the same rational strategy at the same time, they collectively eat the pasture.

AI systems do not buy houses. They do not subscribe to software, take vacations or order delivery. Personal consumption is about 68% of the US economy, and most of it is wages being cycled back into the market. Every time a firm replaces a paycheck with a server, it grazes on the shared consumer base without replenishing it. The firm keeps all of the savings. The lost spending is spread across every business the worker used to buy from.

That is Lloyd’s arithmetic exactly, and economists have started to formalize it. In March 2026, Brett Hemenway Falk and Gerry Tsoukalas posted a paper titled The AI Layoff Trap. Its core result is the one above: firms capture the full cost savings from replacing workers but bear only part of the resulting loss of demand, because competitors absorb the rest. The result, in their words, is “an automation arms race, displacing workers well beyond what is collectively optimal.” The idea has older roots. Korinek and Stiglitz argued in 2017 that worker-replacing technology imposes pecuniary externalities on workers: costs that pass through prices and wages rather than smoke or noise, and that no single innovator has a reason to count.

The result is a consumption paradox. Companies are building hyper-efficient production machines to sell to a consumer base whose wages are being engineered away.

The simulator below is a deliberately simple circular-flow model. The economy starts with workers earning 60% of income and owners 40%. Households spend 90 cents of each wage dollar; owners spend 35 cents of each profit dollar. Each year firms automate some share of the human tasks that remain, and next year’s spending depends on this year’s incomes. Pick how hard firms automate, then pick what the rules do.

Three things fall out of it, and the third surprised me.

First, the decline is delayed, then steady. In the first year or two almost nothing seems to happen, which is exactly why no one in the boardroom sees it coming. Second, the faster the herd grows, the barer the pasture: at the “All-in” pace, consumer spending halves in a decade.

Third, and this is the tragedy in its purest form, going harder buys owners as a group almost nothing. Between “Steady” and “All-in,” total owner profit at year ten barely moves, from 48 to 49, while consumer spending falls from 65 to 51. Each firm wins its own race. The class of firms gains nothing from running it faster, and the village loses a fifth of its spending.

Left strictly to the market, the endpoint of this dynamic is a deflationary squeeze. Wages fall, spending falls, revenue falls, and firms respond by cutting the costs they have left, which are the humans they have left. Empty malls and shuttered storefronts become the modern bare-worn common.

You cannot fix this by appealing to the conscience of individual companies, because they are playing by the rules exactly as written. A CEO who holds back alone is not being moral. He is being outcompeted, and his workers lose their jobs a year later from a weaker firm.

Here the history of the commons is more hopeful than Hardin’s famous essay. Elinor Ostrom won the 2009 Nobel in economics for documenting commons that did not collapse: Swiss alpine meadows, Japanese forests, Spanish irrigation systems, some governed well for centuries. They survived because the users wrote rules about who could graze, how much, and what happened to those who took too much. The tragedy is not destiny. It is what happens when there are no rules.

So the question is which rule. There are three serious candidates.

1. Structural tax reform

   Today’s tax code subsidizes the herd. Daron Acemoglu, Andrea Manera and Pascual Restrepo concluded in 2020 that “the US tax system is biased against labor and in favor of capital,” which pushes firms toward more automation than is economically optimal. Falk and Tsoukalas go further: a Pigouvian automation tax, which charges each firm for the demand it destroys, is the one tool in their model that fixes the incentive directly.

2. Redefining fiduciary duty

   Since the law already permits long-term, stakeholder-minded decisions, the real lever is what boards are rewarded for: pay tied to share price, investor stewardship codes, and benefit-corporation charters that write other constituencies into the company’s purpose. Useful, but it asks individual firms to hold back, and the equilibrium punishes the ones that do.

3. Universal basic income

   Re-injecting purchasing power keeps the pasture green, and in the simulator a UBI funded by a profit levy holds spending up better than anything else. But be precise about what it does. Falk and Tsoukalas find that UBI, like wage policy and worker equity stakes, does not remove the incentive to over-automate. It treats the demand loss without changing the race.

My reading is that the three are not rivals. A UBI protects the village from the grazing already underway. A tax on automation slows the herd to a pace the pasture can bear. Governance reform changes what the farmers are paid to want. Ostrom’s commons lasted because they had all three: limits, sharing and enforcement.

Already true

Consumption is about 68% of GDP. Labor’s share of business output is at a record low. Economists have now formalized automation as a demand externality in which each firm keeps its savings and passes on the demand loss.

What has to happen

AI has to displace wage income faster than it creates new tasks, new products and new jobs, and faster than falling prices raise the real value of the wages that remain. That has not happened in any previous automation wave.

Where I am probably wrong

Owners spend too, prices fall as costs fall, and new kinds of work have always appeared. If AI turns out like electricity rather than like a herd with no fence, the pasture regrows faster than it is grazed and this essay will read like Luddism. The model above leaves out every one of those offsets on purpose. The warning stands only if they arrive too slowly.