The AI wealth debate is aimed at the wrong balance sheet

In the second week of July, a survey started making the rounds: roughly 69 percent of Americans said they would back a law forcing the largest AI companies to hand half their ownership to a public fund. A month earlier, Bernie Sanders had introduced almost exactly that bill, a one-time tax that would give the government a 50 percent stake in AI firms with more than $200 million in sales, paid out to citizens as roughly a thousand dollars a person. And while that bill sits in a Congress that will not pass it, two nonprofits are already cutting $1,000 monthly checks to a few dozen workers who lost income to AI.

Everyone in that picture agrees on the premise. AI is going to move an enormous amount of money, and the people whose jobs it erases should get some of it back. The argument is entirely about how. And most of the proposals on the table are aimed at the wrong balance sheet.

Where the money actually goes

Sanders’ bill taxes the AI vendor. Take half of OpenAI, half of Anthropic, half of the companies building the models, and pay the public a dividend. It sounds like you are taxing the winner. Mostly you are not.

Here is the part the headline skips. When a business swaps a support team for software, the AI vendor bills a small fraction of what that team used to cost. A seat of software runs a few dollars a conversation. A support rep ran a salary. The gap between those two numbers does not go to the vendor. It stays with the company that bought the software and stopped paying the salary.

Salesforce is the clean example. It cut about 4,000 customer support roles and said it no longer needs to backfill them, crediting its own AI agents. Those savings sit on Salesforce’s books, the buyer, not on the ledger of whatever model it runs underneath. Multiply that across every company quietly not rehiring, and the displaced payroll pools up on the buyer side of the market. Gartner expects half of the companies that cut service staff for AI to reverse course and rehire by 2027, which tells you how much of that saving was real in the first place. Customer service is ground zero for this. It is also where the money leaks out of view.

So a fund built only from AI vendors’ stock is collecting from a thin slice of the money that actually moved. The bigger pool is sitting with the buyers, and no proposal in the current debate reaches it.

A check is the easy part

There is a second problem, and the basic-income people flagged it themselves. The nonprofit sending those $1,000 checks is doing a real thing, but the critique in their own coverage is blunt: it is charity, dependent on corporate goodwill, and goodwill gets cut the first bad quarter. A check is the easiest promise to make and the easiest to stop.

Even Sam Altman seems to have landed here. He has moved off flat cash payments toward ownership stakes and public wealth funds, the idea being that a stake compounds and is much harder to claw back than a monthly deposit. That instinct is right. Ownership outlasts the year money gets tight.

What would actually hold

Put those two problems together and you get a shape. Not a revocable check, but ownership held somewhere a future board cannot quietly undo. Not a tax on the vendor alone, but a contribution from the buyers who keep most of the saved salary. And, because rivals agreeing to a shared give-back looks uncomfortably close to a cartel on paper, a narrow legal safe harbor so companies can coordinate on funding workers without coordinating on prices.

That is the thing we are trying to write down. We build AI that does customer service work people are paid to do, so the fix is partly ours. Our version is called the Dividend Standard: the greater of 5 percent of our revenue or 70 percent of our profit, every year, directed to the workers our software displaces, granted as a real ownership stake rather than a handout. Tied to revenue, it pays from our first dollar, not our first profit. The math on the page is deliberately unflattering: one company acting alone can fund maybe ten to fifteen cents on each displaced dollar. A shared standard, with buyers contributing their verified savings, could reach closer to sixty.

None of that requires a seven trillion dollar federal fund to clear the Senate first. It requires companies that automate jobs to structure the give-back the way you would structure any obligation you actually intend to keep. The survey says the public already wants the gains shared. The open question is whether the money reaches people as a check aimed at the wrong company, or as a stake in the economy that replaced their job.

We think it should be the second. We have written it down in full and put it in force, and we are asking anyone who studies this for a living to tell us where it breaks.