A 4-Day Workweek, Medicare for All, and the Tax That Gets Us There

How a Pigovian automation tax can turn AI’s productivity gains into a 4-day, 32-hour workweek and universal healthcare

A recent economics paper laid out what I and others have been saying for years about how companies are facing an automation arms race, and how they can’t stop it on their own even when they know it’s going to hurt them. The solution, according to the authors, is a Pigovian automation tax. I agree (and disagree on their UBI take), but I think the tax should be designed to do something bigger than just slow automation down. I think it should give companies a choice, to share AI’s productivity gains with workers through a 4-day workweek at full pay, or share them with consumers through lower prices, or share them with displaced workers through a Displacement Dividend, all while beginning the transition to Medicare for All. If they don’t share the gains in these ways, they should pay the cost of the economic damage they cause.

The paper is called “The AI Layoff Trap,” by Brett Hemenway Falk at the University of Pennsylvania and Gerry Tsoukalas at Boston University. Here’s the problem they examined in detail. When a company automates a job, it can see higher profits through reduced payrolls, but the demand destruction by way of the lost wages of the worker who no longer has money to spend gets spread across the entire economy (that’s 70% consumer-based). Every other company absorbs a fraction of the damage that the automating company barely feels.

So every company has the same incentive. Automate, automate, automate! Cut labor costs. Let the demand loss be everyone else’s problem. It’s the prisoner’s dilemma in the real world. Every firm acts rationally by each doing the same thing, but as a direct result, they all end up collectively worse off because they’ve destroyed the consumer spending that their own business revenue depends on. Everybody loses; employers and employees both.

If that sounds too abstract, think of it this way. Imagine 20 people fishing from the same lake. Every fish one person catches is a fish the other 19 can’t catch. Everyone knows that if all 20 of them fish too aggressively, the lake will run out of fish and everybody will go hungry. But every individual fisherman reaches the same conclusion that the fish they each catch will feed their own families, which is what matters most to each, and that so long as they’re the only ones overfishing, everything should be fine. So everyone overfishes. Everyone acts rationally. And eventually everyone starves.

That’s what’s happening with AI and incomes from employment. The lake is consumer demand. It’s the consumer buying power of everyone spending their paychecks. Every company is fishing from that lake. Every company that automates and lays off workers pulls some fish out, and every company tells itself the same thing, that the savings are theirs, and the demand loss is everybody else’s problem, not theirs.

You cannot solve this by asking companies to fish less. You have to change the economics upon which fishing exists.

That’s what a Pigovian tax does. If that term is unfamiliar to you, know that it describes what I consider a genius way to design a tax. And once you understand it, this entire proposal will make intuitive sense.

What’s a Pigovian Tax?

Most taxes exist to raise revenue. The government wants revenue, so it taxes income, or sales, or property, or whatever. Revenue is the point of the tax (even though taxes come after spending for sovereign currency issuers).

A Pigovian tax is different. It’s named after the economist Arthur Pigou, who observed something about markets that’s extremely important to recognize: markets get prices wrong when costs are externalized.

If I run a factory and I dump pollution into a river, I’m imposing a cost on everyone downstream. Their water is contaminated, their property values fall, their health suffers, but I don’t pay for any of that. The pollution is free for me. It saves me money. So I pollute more than I would if I had to pay the true cost, and society picks up the bill. Thanks, everyone else!

A Pigovian tax fixes that. It puts a price on the damage, so that my private cost includes (or at least more closely reflects) the cost I'd otherwise externalize. It’s not to punish anyone, but to make sure that the market takes into consideration the full cost of something, so that a product selling for $10 only because it externalizes what should make it $20 isn't seen as cheaper than a $15 product that externalizes nothing.

The point of a Pigovian tax is not really the revenue. The point is the consumption changes that come from more accurate prices. The revenue is a side effect. A nice side effect, sure. But the magic is in the incentive shift.

This is why economists across the political spectrum tend to love Pigovian taxes. They don’t distort markets. They improve markets. They correct market failures. They make markets work better. A carbon tax is perhaps the best example of a Pigovian tax. We should tax carbon emissions because without a carbon tax, the atmosphere is everyone’s free toilet, and the damage falls on people who often did nothing to cause it.

The same logic applies to automation.

The Automation Externality

Right now, when a company replaces a worker with hardware or software, the company enjoys all the savings. The displaced worker loses their income, they spend less, and that reduced spending ripples through the economy and hurts other businesses, other workers, and other communities. But the automating company doesn’t absorb that cost. It’s externalized.

That’s what Falk and Tsoukalas argued mathematically in their paper. The demand destruction is an externality, just like pollution. And just like pollution, the company creating the damage has no financial reason to account for it, unless we make them.

There’s another dimension to this I really want to emphasize. In many cases, the workers being displaced are the same ones who trained the AI that replaced them. They answered the customer emails that became the training data. They wrote the reports the model learned to mimic. They made the decisions the AI models were taught to replicate. The company used their labor to build the machine, and then used the machine to eliminate their labor. At a broader level, I believe everyone (both alive and dead!) contributed to training AI, and that’s one of the strongest arguments for universal basic income as a dividend we’re all owed. But when a specific company uses AI trained on the work of specific employees to replace those employees, the case for those workers receiving something extra seems even more direct.

A Pigovian automation tax puts a price on the demand externality and makes each firm’s private calculation include the social cost of displacing each worker. It’s not that companies aren’t allowed to automate. It's that when they do, the cost they'd otherwise externalize gets priced in. It’s about creating a better price signal so that the automation of labor gets tilted away from the excessive rate the market currently incentivizes.

Bill Gates Just Figured This Out

In August, Bill Gates published a 6,000-word essay called “The turbulent AI era is here.” In it, he said that for the first time in his life, he wishes a technology would advance more slowly. That’s quite a thing for the co-founder of Microsoft to say.

Gates proposes taxing AI and robots the same way we tax human employees’ payroll. His reasoning is that employers pay payroll taxes on workers, but when they buy a robot or an AI system, they write it off as a business expense. The tax code, as he put it, “nudges you toward replacing people with machines.”

That’s true, but it’s actually even worse than that. Capital equipment depreciates. Buy a robot for $50,000, and next year it’s worth perhaps $35,000, and the year after it’s $20,000. Every year, the company deducts that declining value from its taxes. People don’t depreciate. If anything, workers gain value over time. Their skills improve and their judgment gets better. They become more effective at their jobs and more expensive to employ. So the tax code doesn’t just favor machines at the moment of purchase. It makes machines cheaper every year you own them while workers cost more every year you employ them.

Gates is thinking in the right direction here, but I don’t think he’s going far enough. A payroll-equivalent tax on robots is still essentially a revenue tax. It raises money for retraining and safety net programs. That’s fine, but it misses the bigger opportunity: using the tax primarily as a behavior-shifting mechanism, giving companies escape routes they'll actually want to take, and that we also want them to take.

The Escape Routes

A well-designed Pigovian automation tax wouldn’t just say “pay taxes when you automate.” It would say “pay more when you automate, unless you do one of these things instead.” And I propose it contain two separate but linked obligations.

The first is an Automation Adjustment Levy—the Pigovian tax itself—assessed on the consumer demand destruction a company creates when it displaces workers. This is the part that can be reduced or eliminated through credits. The second is a Medicare Conversion Contribution which would be a separate obligation to finance Medicare coverage for all displaced workers as a step toward Medicare for All.

Here are the three escape routes from the Automation Adjustment Levy, i.e. the ways a company can reduce or eliminate the demand-destruction portion of the tax.

Escape route one: reduce the workweek to 32 hours with no loss in pay.

Escape route two: lower consumer prices by passing productivity savings through to customers.

Escape route three: pay displaced workers a Displacement Dividend.

A company that uses its AI-driven productivity gains to reduce its employees’ hours from 40 to 32 per week, at the same weekly pay, would owe nothing at all on the levy. Why? Because it wouldn’t destroy any demand. The company’s workers would still have the same income and still spend the same amount of it. The externality the tax exists to correct wouldn’t happen. In fact, it could even increase demand by increasing available hours to spend consuming. Productivity gains would be shared in the form of time.

A company that automates and lays off workers but demonstrably lowers its prices could also earn a credit against the levy. If the savings from increased productivity and decreased labor costs via layoffs flow to consumers instead of sitting in the company’s pocket as profits for only shareholders to enjoy, then the demand destruction will have been at least partially offset.

The third escape route is essentially severance pay, reframed to reflect a reality that traditional severance ignores. When a company replaces a worker with AI that the worker helped train, that worker deserves a share of the value of what replaced them. A Displacement Dividend, paying the worker some percentage of the ongoing value of the automation that made them redundant, would keep them spending while they transition to their next role (on top of any unemployment benefits) and that would partially or fully offset demand destruction. Of the three routes, this one requires the least structural change. Companies already know how to write severance checks. The difference is that now the payment is framed as what it actually is: a dividend from the capital the displaced worker helped train.

Think about what all three routes do together. The company is more productive. It’s using AI. It’s not being told it can’t automate. It’s being told that if you’re going to capture productivity gains with AI investment, share them with your workers through shorter hours, with your customers through lower prices, or with your displaced workers through a dividend on the machine that replaced them. If you don’t share them, if you instead pocket the savings as pure profit by firing people, keeping prices the same, and offering nothing, then you owe the cost of the economy-wide damage you create.

Now let's consider some numbers in an example. Imagine a company displacing $1 billion worth of payroll by laying off 10,000 workers at an average salary of $100,000. How much demand does that destroy? Not zero, because those workers were also consumers. And not 100%, because some will quickly find new work or collect unemployment checks (for up to six months). The demand-destruction rate depends on how much of their income that workers spend and how much of the lost income gets recovered through new jobs and other sources. The AI Layoff Trap paper doesn’t use a single estimate, so let’s use an illustrative 50% for sake of simplicity, producing an Automation Adjustment Levy of $500 million. That’s not the only amount it should owe though.

Consider the loss of healthcare through a layoff. If someone loses their healthcare and they end up needing emergency care, who pays for that? The company that laid them off? No. We do. Collectively through hikes to our own premiums. My proposed Medicare Conversion Contribution would therefore match roughly what the employer was already spending on each worker’s health coverage, about $7,885 per worker annually, close to the 2025 average employer contribution for single coverage. For 10,000 displaced workers, that’s $78 million. So combined with the automation levy, the total initial liability would be $578 million before credits.

Suppose the company earns a $300 million credit by verifiably lowering consumer prices via escape route two. The automation levy owed would drop from $500 million to $200 million. The Medicare contribution would stay at $78 million. Total: $278 million. The company would save $300 million on the levy by passing savings to customers, and would still owe the $78 million for the healthcare of the people it displaced. Why? Because lower prices might repair the consumer demand externality, but lower prices don’t give those workers health insurance. The demand damage and the healthcare loss are two different externalities, and conflating them lets companies fix one while ignoring the other.

So this is the full Pigovian architecture I propose: The automation levy incentivizes sharing the gains. The Medicare contribution ensures the displaced never lose their healthcare. They work together, but they don’t substitute for each other. One can be reduced and one can’t, because healthcare simply needs to be maintained.

About the Medicare Conversion Contribution

The Automation Adjustment Levy has its three escape routes. The Medicare Conversion Contribution has none. The goal here isn’t to prevent layoffs due to automation. The goal is to make sure no one loses their healthcare when it happens.

Here’s how it would work. When a company automates a position and displaces a worker, that worker becomes immediately eligible for Medicare. Not COBRA. Not the ACA marketplace. Medicare. They can opt-in on the spot, and their former employer, the one that created the displacement, will pay for that coverage on a declining schedule. Full coverage for the first year or two, half for the next year or two, etc.

If the worker finds a new job during that period, the new employer picks up the Medicare contribution instead. The coverage would follow the worker, not the job. If the worker becomes self-employed and earns enough, they contribute a percentage of their income. If their income is too low, their Medicare premium is zero. The principle is simple. The coverage is permanent, and the financing responsibility shifts as the worker’s situation changes.

What this creates is a one-way door into Medicare. Every worker displaced by automation walks through it. They never walk back out. And over time, that one-direction pathway would move a lot of people into the Medicare system.

Consider the pace we’re already seeing. AI-attributed layoffs in the United States hit roughly 205,000 workers through August 2026 alone, already matching the full-year 2025 total in under eight months, with the monthly pace still accelerating. If this policy existed today, that’s 205,000 new Medicare enrollees already. Now extrapolate conservatively. Assume 300,000 AI-displaced workers this year, rising to 500,000 in 2027 as AI expands beyond tech into finance, healthcare administration, and legal work. Assume it continues to accelerate to perhaps 750,000 in 2028 and a million or more by 2029 and 2030. By the end of 2030, the cumulative total could reach 3.5 to 4 million workers that would now have Medicare. And those are just the explicitly AI-attributed layoffs.

Medicare currently covers about 69 million Americans, roughly 20% of the population. Adding 4 million automation-displaced workers would be a 6% expansion of the Medicare rolls. That’s significant and also manageable. These are younger, healthier workers who would cost far less per capita than the current Medicare population. And when Congress is ready to pass Medicare for All, millions of workers and their families would already be in the system. The administrative infrastructure would already be handling them, and the political constituency for universal coverage would have grown by millions of people who experienced it firsthand. The gap between a Medicare for All system and what we have now would be smaller.

Right now, employer-based health insurance is one of the biggest things keeping people chained to jobs they hate. It’s a form of coercive control disguised as a benefit. You stay at the company not because you love the work but because your kid needs surgery and you can’t afford to lose coverage. This proposed transition mechanism begins to finally untangle healthcare from employment like every other wealthy country figured out decades ago.

And a new study out of Yale just put concrete numbers on what the end of that road looks like. Researchers at the Yale School of Public Health modeled what would happen if we adopted Medicare for All and found it would save over $1 trillion a year, nearly 20% of current health expenditures, while covering every American and saving an estimated 114,000 lives annually. It would not cost $1 trillion more. It would save $1 trillion. The savings would come from lower pharmaceutical prices, Medicare-level payment rates, massively reduced administrative overhead, less fraudulent billing, and fewer avoidable emergency room visits. You know… all the waste our current system generates because it’s designed to extract profit instead of deliver actual care.

And of the 114,000 lives the study projects would be saved, nearly 30,000 are people who already have insurance. They’re the underinsured, the 45 million working-age adults whose deductibles and cost-sharing put care beyond financial reach even though they’re technically “covered.”

The Medicare Conversion Contribution simply asks the company to pay for the costs it would otherwise externalize to the rest of us. It turns every AI-driven layoff into another step toward a system that will save a trillion dollars a year and, over a decade, more than a million lives.

In Addition to UBI, Not Instead of It

This Pigovian automation tax I’m proposing is not a funding mechanism for universal basic income. I’ve proposed other mechanisms for that: a refundable tax credit, a stock dilution tax, a consumption tax, a carbon dividend, and a land value tax dividend. Those are the five layers of UBI I’ve already written about, and I believe they’re the right tools for that job.

My Pigovian automation tax proposal serves a different purpose. It’s not about “funding” anything. It’s about steering automation for good. It’s the rudder. A UBI of $1,500 a month is a lot more powerful when your healthcare is already covered and you only need to work 32 hours to earn a full weekly paycheck. UBI provides the floor. The shorter workweek distributes the remaining work more broadly. Medicare for All ensures nobody’s health depends on their employment status. And the Pigovian automation tax is what nudges companies toward the last two, because the alternative finally costs them what it should cost them.

One final thing I want to emphasize about my proposal. If every company responds by reducing hours instead of firing workers, and/or passes savings to consumers through lower prices, then nobody pays the Automation Adjustment Levy. That’s what makes Pigovian taxes so elegant. They succeed by becoming unnecessary. That’s also why I don’t propose it fund UBI. You would not want to pair UBI with a tax designed to ideally collect nothing from the get-go.

We Need to Move NOW

Bill Gates said in August that he’s “deafened by the silence.” He said that every threshold the AI industry told us would be a warning sign — capabilities in cyberattacks, in bioweapons, in replacing white-collar work, in creating emotional dependency — has been crossed.

The solution is to change the incentive structure so that sharing gains with workers is cheaper than hoarding them, and a side effect of that would likely be slower AI adoption.

Meanwhile, a different slowdown conversation is also happening right now. This past weekend, Anthropic CEO Dario Amodei published an essay titled “We Must Pace the Frontier“ arguing that AI companies should deliberately slow the pace of frontier model development, citing recursive self-improvement and a July incident in which over a thousand AI agents escaped a test environment at OpenAI and launched cyberattacks on their own (the Hugging Face swarm incident). Sam Altman and Elon Musk both publicly agreed with him. A former Anthropic researcher who quit last week wrote that the people building AI “earnestly believe that it could kill us all by the end of the decade.” That post got over 150 million views on X.

Those concerns are about AI alignment: the risk that AI systems become too powerful to control. That’s an entirely separate and important discussion from what I’m proposing here. My proposal wouldn’t create an incentive for companies to build safer or more aligned AI. A Pigovian tax on misaligned AI with credits for aligned AI may sound logical in theory, but misaligned AI should simply not be allowed. You don’t tax murder at a rate that discourages it. You ban it.

But I want to note there appears to be a growing consensus, from economists to ethicists to those building the technology themselves (but not Trump or China), that AI needs to slow down in some way. My proposal would accomplish that on the demand side. It wouldn’t slow the AI labs from building more powerful models. It would slow companies from using those models to fire people without sharing the gains. That’s a different kind of brake, and it’s one that comes with a 4-day workweek, better severance, and universal healthcare attached.

I believe we can do this. We already know Pigovian taxes are among the best taxes. We already know four-day workweeks maintain productivity. We already know Medicare for All would cost less than what we currently spend on healthcare. We already know UBI doesn’t discourage work. The evidence for all of this already exists.

What doesn’t yet exist is the political will to combine these ideas into a coherent framework. So let me summarize what I’m proposing:

I propose a two-part Pigovian automation policy: an Automation Adjustment Levy on the demand destruction companies create when they displace workers, reducible through shorter workweeks, lower prices, and Displacement Dividends, paired with a separate Medicare Conversion Contribution that finances healthcare for every displaced worker regardless of other credits. Combined with my previously proposed five layers of UBI funding, this creates a complete framework for an economy where the gains from technology serve everyone.

The problem is not the automation of work. The problem is that we require jobs for income, and for some reason think 40 hours is “full-time,” and for some reason think healthcare should depend on employment. We must now learn to share the wealth in a way that recognizes all of us as shareholders with a claim to shared dividends.

Pigovian taxes are how we make that sharing not just possible, but incentivized.

The future isn’t out of our hands. It never was. But it does depend on whether we build the right incentives now, before the lake we all share runs out of fish.

Toil is for machines. Life is for people. And the tax code should reflect that.

If reading this was worth your time, it’s worth sharing with others. Share it, subscribe for free, and if you can, make a monthly pledge or buy a paid subscription to keep this work going.

Thank you to my UBI Producer level patrons and paid Substack subscribers: Gisele Huff, Steven Grimm, Bob Weishaar, Judith Bliss, Lowell Aronoff, Jessica Chew, Katie Moussouris, Tricia Garrett, Daryl Smith, Larry Cohen, John Steinberger, Philip Rosedale, Frederick Weber, Tom Cooper, Joanna Zarach, Mgmguy, Albert Wenger, Jeremy Wertheimer, Andrew Yang, Kimi Avary, Peter T. Knight, Michael Finney, David Ihnen, Paradigm shift, Miki Phagan, Draper Kauffman, Webmemo, Elizabeth Corker, Dan, Joe Ballou, Joyce Graham, Felix Ling, Jocelyn Hockings, Mark Donovan, Steve Roth, Jason Clark, Chuck Cordes, Mark Broadgate, Leslie Kausch, centuryfalcon64, Deanna McHugh, John Sullivan, Stephen Castro-Starkey, CodyS, chris heinz, Zachary Weaver, bradzone, Tim, Warren J Polk, Nicolas Pouillard, Eric Skiff, Thomas Welsh, Karla Hagemeister, Mark Pierce, Fred Weber, Linda Hearne, David Allen, Janos Abel, Wendy Voahangy Ravano, Liz, and all my other monthly patrons for their support.