By Austin VornhagenTen million dollars. Every year. For every American.
That was the clip. It came from Daniel Kokotajlo, a former OpenAI researcher, in a long conversation with Steven Bartlett on The Diary of a CEO this July. If robots and AI end up doing most of the work, he argued, the wealth they produce could be shared with everyone, through something he calls a Citizen’s Dividend. The figure he attached to it for 2040 sounded like a typo.
It isn’t one. Inside his model, the number follows from the assumptions. We’ll get to exactly how.
But a different question grabbed me, and it wouldn’t let go: what happens on the Tuesday your employer figures out it doesn’t need you anymore, and the robot money hasn’t shown up yet?
Because it won’t show up all at once. Every serious version of this idea, including his, starts small and grows. Run honest arithmetic on a realistic first version funded by a tax most people would recognize, and the check comes out around $42 a month. That’s a phone bill.
The dividend is the easy part to imagine. The gap before it is the part that can hurt you.
So this essay asks two questions. How would a robot dividend actually work? And how does an ordinary person get across the years between the last paycheck and the first check worth having: first with the government’s help, then without it?
The dividend is made of scarcity
Kokotajlo’s idea lives inside a bigger document. In July 2026 his group, the AI Futures Project, published AI 2040: Plan A, a step-by-step scenario for how the United States and China could coordinate to slow the race to superintelligence, pushing it back to around 2040 instead of letting it arrive this decade. The authors present it as a plan they are arguing for, not a prediction of what they expect to happen.
The dividend is one gear in that plan, and it works in a way most people who shared the clip probably missed.
It isn’t “the government taxes the robots.” It’s closer to cap-and-trade, the approach used to limit pollution, pointed at chips and robots. Governments agree to limit how fast AI hardware and robots can expand. The right to expand is sold as permits, at auction. The auction money flows to citizens.
Here’s the part worth sitting with: the permits are valuable because they’re limited.
Picture a beach town that allows only ten food trucks on the boardwalk and auctions the ten licenses every spring. In a hot summer those licenses are worth a fortune, and the town collects it. If the town allowed unlimited trucks, a license would be worth close to nothing. The scarcity is what turns the boardwalk’s crowds into money for the town.
Swap the food trucks for data centers and robot factories, and swap the boardwalk for an economy growing faster than any in history. That’s the engine. Build it one part at a time.
How a permit-funded Citizen’s Dividend works, assembled step by step
Machines do the work. Owners get paid.
Out of the box, output flows to whoever owns the machines. Nothing about automation, however impressive, sends a dollar to anyone else. This is the starting position, and it’s worth staring at.
Limit how fast the machines can grow
Governments agree to cap how quickly new AI chips and robots can be added. In Plan A this is part of a U.S.-China arrangement meant to slow the race, not just to raise money.
Sell the right to grow
A company that wants more capacity has to buy a permit. Because permits are scarce and the machines are enormously profitable, the bids get very large. The food-truck license, at planetary scale.
A public agency holds the money
The auction revenue goes to a public body rather than into general spending. In Plan A’s economic model, most of the U.S. share of permit revenue, 75 percent in 2035, is paid out to the population.
Every citizen gets the same cut
Everyone holds an equal claim. The bigger the automated economy grows, the bigger the auctions, the bigger the check. You don’t need to own a single share of a robot company to get paid by the robots.
The regulator gets paid to say yes
An agency funded by permit sales earns more when it sells more permits. The body meant to slow the machines is rewarded for speeding them up, and every citizen has a reason to cheer it on. Any real version has to wall the safety decision off from the revenue.
So where does $10 million come from? The economics supplement behind Plan A assumes that AI and robots can do the vast majority of economic tasks and are allowed to roughly double every six months. That produces average real output growth of about 90 percent a year from 2032 to 2037. For a sense of how wild that is, a good year for the U.S. economy has usually meant growth of a few percent.
Run that economy through the permit auctions and the Citizen’s Dividend lands around $1 million per person in 2035 and roughly $10 million in 2040, measured in 2025 dollars. In the interview, Kokotajlo described it starting out around $25,000 a year. Those are outputs of extraordinary assumptions, not a savings projection, and his team says so.
Notice what actually made the money reach you in that diagram. It wasn’t the robots. The robots only ever paid their owners. The public got a cut at exactly one point: when somebody wrote a rule.
Automation doesn’t make you a shareholder. A rule does.
Nobody wakes up owning the machines because the machines got good. Their output belongs to whoever owns them, unless a rule written before the output exists says otherwise. Plan A’s rule is the permit auction. There are other possible rules. But there has to be one, and it has to be in place early, which brings us to the problem.
Your paycheck and your dividend won’t pass each other politely
Zoom in on one household. Every number below is illustrative, built from the arithmetic later in this essay rather than from any forecast.
A worker earns $60,000 a year, which is $5,000 a month. In 2028 the job goes. Maybe to software, maybe to a “restructuring” that nobody calls automation. A robot dividend exists by then. It pays $42 a month.
Now watch what it takes to keep this household on its feet.
The gap between a lost paycheck and a growing dividend, and the planks that bridge it
One paycheck, $5,000 a month
This is the life the household is built around: the rent, the car payment, the braces. Picture it running straight across to 2040.
The paycheck stops. The dividend is $42.
The job is automated away. The dividend is real, and it’s the thin gold sliver at the bottom. Everything in red is money this household used to have and doesn’t.
The dividend grows at the speed of revenue
It steps up as the automated economy grows and the money actually arrives: $167, then $500, then $1,000 a month. Useful. Also years late. For most of a decade, the red barely shrinks.
Catch the fall
Replace 70 percent of prior earnings, capped at $4,000 a month, for up to a year. For this worker that’s $3,500 a month. The worst of the cliff disappears.
Don’t let the rope run out mid-crossing
Unemployment jumps, so benefits extend on their own. When the worker lands a new job at $3,200 a month, half the pay cut is covered at first, tapering to zero over three years.
Or the job never comes back
Swap the new job for nothing. If work stays scarce, temporary help has to become a lasting floor, here topping total income up to $2,500 a month until the dividend grows into it. The gap doesn’t vanish. It becomes survivable.
Look at what closed the gap. Not the dividend. The dividend was the destination, and it arrived exactly as fast as the revenue allowed. The bridge was built from other planks, and every one of them had to exist before 2028.
A promise of dividends in 2040 does nothing for a mortgage payment in 2028.
Six coupons, starting now
Old paper bonds came with coupons down the side. Each one was a dated promise, and you clipped it when the payment came due. Here’s the government path to a robot dividend as six coupons, starting in September 2026.
The dates are planning targets, not predictions. The bill is my hypothetical, not anything that exists.
- September 2026 to 2027
Write the claim down
A nationwide, mandatory dividend can’t start as a private club that declares everyone a shareholder and sends AI companies a bill. It starts as federal law. Call it the American Automation Dividend Act, a name I made up. It would create a legally separate public fund with a defined revenue source; give every eligible citizen an equal right to distributions, including children and new citizens; fund the Treasury to enroll and pay people who don’t file taxes or have bank accounts; require independent audits; and set up separate help for lost earnings before the dividend gets big.
One design choice matters more than it looks: make the share nontransferable. Otherwise someone behind on rent could sell twenty years of future dividends for a few thousand dollars today, and ownership would pile right back up where it started.
There’s precedent. Alaska saves resource revenue in a professionally managed permanent fund, a separate state division pays residents their dividends, and the legislature still has real say over the size of the check. The machinery can exist. So does the politics. There is no technical workaround for winning that fight.
- 2027 to 2028
Start collecting. Pay something.
Don’t make the whole thing wait on a global robot treaty arriving on time. Start with a broad surtax on corporate profits, and whenever public money subsidizes AI infrastructure, take equity or a revenue share in return.
Why tax profits broadly instead of “AI companies”? Because the gains from automation may land at the insurer, the retailer, or the manufacturer that uses AI, while the AI vendor itself earns relatively little. Trying to trace which dollar of profit “came from AI” is an attribution nightmare.
5-point surtax × an assumed $4 trillion profit base$200 billionPaid out after reserves$150 billion÷ an illustrative 300 million recipients$500 a yearPer month$42An illustration, not a revenue estimate. Real companies shift investment and profits; a real bill needs independent scoring.
That’s where the $42 comes from. It’s the right sequence (lock in revenue, collect it, pay some out, build assets) and it replaces nobody’s paycheck.
- Immediately, alongside it
Build the bridge
The worker in the chart needs help in 2028, not a promise. The planks, as a proposal:
- Job disappearsReplace 70% of prior earnings, capped at $4,000 a month, for up to 12 months.
- Hours get cutReplace part of the lost earnings while the person stays employed.
- New job pays lessCover part of the difference for a while, tapering off.
- Health coverage endsAn automatic coverage bridge with funded premiums.
- Self-employment collapsesHelp based on verified lost earnings, with fraud controls.
- The whole town goes quietTemporary federal support for essential local services and real jobs.
Part of this already has a template. Under the short-time compensation programs many states run, a participating employer cuts hours instead of people, and unemployment benefits replace part of the lost wages. Stretching that across a long automation transition would take new policy.
And one rule: nobody should have to prove a robot took their job. The same layoff can be called restructuring, outsourcing, or cost-cutting. Eligibility should follow lost earnings, not the story the employer tells.
It needs its own money, too. Ten million people averaging $2,000 a month for a year is $240 billion, before administration. The same dollar can’t pay for the bridge and also be counted as savings in the dividend fund. Future robot wealth is not money you can spend today.
- 2028 to 2032
Pre-wire the alarm
This is where the “bloodbath” gets prevented or doesn’t. Congress should write the escalation rules before the crisis, so help grows automatically instead of waiting on a vote.
One ready-made national trigger is the Sahm rule: it fires when the three-month average unemployment rate rises half a percentage point above its low of the previous twelve months. Pair it with regional triggers, so a town that loses its biggest employer gets help even when the national numbers look fine, and a hiring-and-earnings check before temporary help expires.
Retraining should point at jobs that demonstrably exist. Sending someone into a new field that’s also shedding workers isn’t a transition. And if work stays scarce, temporary insurance has to turn into a durable floor. Letting it expire on an arbitrary date just rebuilds the gap.
- The 2030s
Raise the check when the money is real
Here’s where most hopeful plans reach for the same answer: start a fund now, let it compound, and by the time the robots arrive it’ll be enormous. Watch what that answer runs into.
The scale of money a dividend requires, compared with the U.S. economy
The first version
$150 billion a year, split 300 million ways. A real program and a rounding error on a rent payment. Every square in this figure has an area proportional to its dollars.
Quadruple it
$600 billion a year buys $167 a month. A grocery run or two.
Triple it again
$1.8 trillion a year, and each person gets $500 a month. That would change a lot of household budgets. It still wouldn’t replace a job.
A thousand a month
$3.6 trillion a year gets everyone $1,000 a month. Keep that number in mind and ask the compounding question: how big would a fund have to be to pay this out of its returns?
The fund that pays it
At an illustrative 4% payout rate, $3.6 trillion a year takes a $90 trillion fund. The dashed square is the entire U.S. economy in 2024, about $29 trillion. The fund would have to hold roughly three years of everything America makes.
And the headline number
$10 million for each of 300 million people is $3,000 trillion a year, in 2025 dollars. Everything you just saw is now the speck in the corner. That payout is only thinkable after years of roughly 90% annual growth. It isn’t a savings plan. It’s a bet on an economy unlike any that has existed.
You can’t compound your way across the gap. Current revenue and the bridge have to carry the early years.
- 2035 to 2040, or later
Grade it at the kitchen table
Don’t declare victory because a year arrives. The test is whether ordinary households can reliably cover essentials through some mix of earnings, dividends, and public services, including during a recession.
That means still managing the unglamorous things: housing that’s scarce, healthcare access, market concentration, the fund itself. A bigger check does less if rent rises right alongside it. And, from the diagram above, a regulator paid by permit sales must never approve unsafe expansion just to keep the checks growing.
Now take the government away
Suppose none of that happens. No law, no fund, no rescue. Ordinary contracts still work. What can regular people build on their own?
Start with the hard limit, because everything else hangs on it. A private system can pay people from exactly three places: customers, investment returns, or someone voluntarily handing over money or assets. Printing shares doesn’t print money.
Inside that limit, the strongest move is the same one as before, scaled down: get a written claim on automated output while your work still earns money.
Own a business ordinary people can own
“Everyone become an AI entrepreneur” isn’t a plan for a population. Most people need something they can join that somebody else organizes.
So picture a worker-owned company with a permanent community stake. A handful of people start a business that handles scheduling, customer follow-up, invoicing, and admin for local businesses. One is great with customers, one can sell, one runs operations, one keeps the books, one can wire up software. They use AI services that already exist. Nobody invents a robot.
Worker cooperatives are an established way for people to pool their abilities, own the business, and share in its success. The permanent community stake is an extra design choice layered on top. And the ownership agreement has to be signed before automation makes the business valuable. It separates four things:
Wages for the hours and responsibilities someone actually handles.
A stake that vests through agreed contributions over time.
A permanent slice held for a broader group of beneficiaries.
Cash kept back for failures, replacements, and investment.
The clause that matters most: when automation eliminates someone’s role, it shouldn’t erase their vested stake. That has to be drafted explicitly. Plenty of ordinary employee-ownership setups don’t keep paying people after they leave.
Pay for it with customers, not hope
The first 90 days look something like this. Founders who can keep their day jobs during the pilot do.
Customer deposits or setup fees can supply some working capital, though they also create obligations to deliver. Someone may still need to put in modest money for formation, insurance, and early losses. And people without savings should be able to earn ownership through paid work, not gamble months of unpaid labor on a vague promise of shares later.
Where the tiny dividend comes from
Suppose that business eventually reaches this hypothetical month. These are assumptions to test, not a forecast.
That’s a robot dividend, built privately and very small. Customers fund the work, machines do more of it over time, and ordinary people own a claim on the surplus. The catch is competition. If someone else sells the same service at half the price, the surplus evaporates. Owning automation doesn’t guarantee profits.
Or buy one that already works
Starting from zero is risky. Another route is buying a profitable business from an owner who wants out, with employees gradually taking ownership. The money can come from private loans plus seller financing, where the seller gets paid over time out of the business’s own cash flow. Employee buyouts already use bank loans, private capital, and seller notes. ESOPs are the best-known version, though they’re retirement plans, not spendable income.
Under the no-government rule, the deal has to pencil out without special tax treatment or government-backed loans, and the seller still needs a fair price and believable management. One more trap: a business that automation is about to flatten is a terrible buy no matter how good its last five years looked.
Then link them together
One company can’t protect a whole community. A network of them can share accounting, purchasing, and technology, and pool a privately funded hardship reserve. A member who loses work at one business gets temporary help and introductions to openings at another. The community stakes can include people who can’t currently do paid work: caregivers, disabled people, people already displaced. But inclusion takes a real transfer of value. Founders, sellers, investors, or donors have to agree to it.
So how far does a network like that go? Try it.
How much does a network of co-ops pay each person?
Set the size of the network. The community fund’s cut of every business’s distributions is split evenly among its beneficiaries.
To pay each beneficiary $30,000 a year at this stake, the network would need $150 million in annual distributions.
Arithmetic only: businesses × distributions × stake ÷ beneficiaries, before fund expenses. It says nothing about whether those businesses can earn that much.
The essay’s example lands at $1,000 a year each. Useful. Nowhere near a salary. A private dividend network can grow into real support, but it can’t responsibly promise anyone an income before it has the assets and the earnings.
Meanwhile, your own household
While institutions like that grow, the practical first move depends on where you’re starting. Pick yours.
Put immediate paid work and cutting unavoidable costs first. Look for ownership you can earn through paid work, not ownership that asks you to give up cash you need for rent.
A share of a future surplus is worth nothing to a household that loses its apartment this spring.
Build a cash buffer while exploring credible employee ownership or a small, paid side venture.
The job is the asset right now. Use the years it still pays to acquire a second claim that doesn’t depend on it.
Consider diversified ownership of businesses, sized to your household’s risks and time horizon.
Diversified investments can lose value exactly when you need them, and an AI boom doesn’t guarantee good returns on AI stocks bought at high prices.
Use automation to improve real margins, keep reserves, and write explicit ownership opportunities for the people who help you get there.
You’re the person the private path depends on. The surplus exists in businesses like yours or it doesn’t exist.
Access depends on family support, privately funded assistance, or ownership someone deliberately sets aside for you.
This is the biggest hole in any private plan, and it’s why a community stake has to be written in on purpose.
An emergency fund, even a modest one, keeps a temporary loss of income from turning into expensive debt. Diversified investments are another route to ownership, with the caveat that they can fall right when you need them. The ideal sequence is gradual: wages first, then wages plus ownership income, then maybe fewer hours as distributions grow. Paced by actual earnings, not by anyone’s forecast for when AGI shows up.
And the honest limit: if displacement moves faster than these institutions can grow, and nobody with resources chooses to share them, there is no private mechanism that protects everyone. That’s the unsolved part.
Write the claim before the value arrives
Step back and the two halves of this essay are one idea at two sizes.
Congress writing a Citizen’s Dividend into law and five people writing a community stake into a co-op’s operating agreement are doing the same thing. They’re deciding who owns the output of the machines before the machines make that output valuable.
The timing is the whole trick. Ownership is cheapest to assign when nothing is worth much yet. Try to write the rule after the robots are wildly profitable and you aren’t designing a system anymore. You’re asking the people who hold the profits to give them up. And the people asking will have just lost the thing that gave them leverage: their work.
The time to decide who owns the robots is before they’re worth owning.
That’s why the $10 million figure, dazzling as it is, is the least useful number in the story. It describes the far cliff. It says nothing about the planks. And the planks (the insurance, the automatic triggers, the nontransferable share, the co-op agreement, the emergency fund, the stake you earn while you’re still getting paid) all have to be laid from the side you’re standing on. Now.
The first plank available in September 2026 isn’t exotic. For a country: a public claim on the revenue and automatic lost-income protection, passed together. For a person: one paid, broadly owned venture with rules, in writing, about who shares the gains when the work gets automated.
This essay grew out of a conversation with an AI that started with a clip from The Diary of a CEO, then got checked against sources. Plan A’s mechanism, its growth assumption, and the $1 million and $10 million figures come from the AI Futures Project’s AI 2040: Plan A and its economics supplement; the $25,000 starting figure is from the interview. Alaska’s structure is from the Alaska Permanent Fund Corporation, work sharing from the Department of Labor, the Sahm rule from FRED, cooperatives from the U.S. Federation of Worker Cooperatives, and ESOPs from the NCEO. The 2024 U.S. GDP figure is the Bureau of Economic Analysis’s. The bill, the benefit design, the surtax arithmetic, the household chart, the co-op month, and the calculator are my illustrations, not forecasts or existing programs.
If “a claim on future output” stuck with you
Nobody Can Save a Nursing Shift
The same idea from the other end: your retirement account holds a claim on work nobody has done yet, and that one fact rearranges the whole pension fight.
Read the essay →Essay · Content PilotsWe Automated the Apprenticeship
What automation deletes first is the entry-level work, which was also how people learned. A close look at the on-ramp the gap starts on.
Read the essay →Essay · Content PilotsStop Watching the Stock Market
Why layoffs, frozen spending, and slow responses feed each other, and why automatic stabilizers like the ones in coupon 04 matter so much.
Read the essay →Primary source · APFCHow Alaska’s fund actually works
The closest real-world machine to a citizen’s dividend: who manages the money, who pays the checks, and who decides how big they are.
Read the explainer →Building the small version of the robot dividend?
Every plank on the no-government side begins the same way: a real business that customers can find, trust, and pay. I build websites for service businesses at Content Pilots, with scheduling and payments built in, so a new, broadly owned company can take its first bookings on day one. And if you think one of my coupons is wrong, tell me. I’ll clip a better one.