Severance gets paid once. The saving pays out every year.

On July 31, Fortune put two things side by side: a warning about AI and employment signed by 16 Nobel laureates, and Bernie Sanders’ proposal to answer it with a sovereign wealth fund. The letter went out on July 13 with roughly 200 economists and computer scientists behind it. Its language was hedged in both directions. Major gains in living standards, possibly. Large-scale job displacement, also possibly.

Every argument since has been about how much and from whom. Almost nobody is arguing about the shape of the payment, and the shape is where these plans come apart.

Block paid well. It paid once.

When Block cut about 4,000 people in late February, roughly 40 percent of the company, Jack Dorsey credited the tooling his own engineers were building. The severance was not stingy. Departing staff got 20 weeks of salary plus a week for every year of tenure, six months of healthcare, vested equity, and a $5,000 transition stipend. Compared to the going rate for a tech layoff, that is a real package, and Dorsey took the whole cut in one day rather than bleeding the company through four quarters of rumors.

Then the market opened and the stock rose as much as 21 percent.

Sit with those two numbers together, because they are measuring different things. The severance is a cost that occurs once. The stock move is the market’s estimate of a saving that occurs every year from now on, discounted back to today. Investors were not applauding a one-quarter expense cut. They were pricing a permanently smaller payroll against permanently similar revenue.

So one side of that trade got a perpetuity and the other side got 20 weeks and a stipend.

The retraining fund has the same problem

Verizon is the more sympathetic case, which makes it the more useful one. Dan Schulman has been the most candid CEO in America about this. He has said publicly that AI could drive unemployment to 20 or 30 percent, and rather than deny what his own layoffs were for, he set up a $20 million reskilling and career transition fund and started asking other Fortune 100 CEOs to put in $20 to 30 million each and pool it.

That is better behavior than the industry norm by a wide margin. Now do the division. Verizon announced 13,000 cuts. Twenty million dollars across 13,000 people is about $1,538 each, paid one time, against a labor saving the company will book in every future year it stays automated.

The gap is not caused by stinginess. It is caused by reaching for the wrong instrument. Severance and retraining grants are exit costs. They are budgeted the way you budget a moving expense: size it once, book it, close the file. Automation is not an exit cost. It is a line that keeps not appearing on the payroll every January.

The critics are right about the shape

Watch the cable panels argue about the Sanders bill and something interesting happens. The sharpest objection from the right is not really about the 50 percent. On Rising, a co-host walked through the plan and kept snagging on the same thing: it is a one-time tax, so what happens after the one time? If you hand out a check and nothing structural changed, what did you fix?

That objection is correct, and it applies far past the bill. It is the same objection that should be aimed at every severance package and every transition fund announced this year. A single payment against a recurring gain is not a settlement. It is a discount.

Sanders at least got the instrument half right, because equity is a perpetuity. Stock keeps paying. His bill aims it at the model vendors, which is a much thinner slice of the money than people assume, but the mechanism has the right shape.

What matching the shape looks like

Here is the arithmetic that made us pick the commitment we picked. A support rep in the US runs somewhere near $45,000 to $55,000 fully loaded. Software that handles a meaningful share of that same queue bills a few dollars per resolution. The difference is not a windfall that lands in one quarter and stops. It recurs, quietly, forever, on the books of the business that bought the software.

If the gain is ongoing, the give-back has to be ongoing. That is the whole reasoning behind The Dividend Standard, our commitment to direct 70 percent of profits to the workers whose jobs AI automates. Not a fund sized to a press release. A percentage, tied to the thing that keeps earning, that gets larger exactly when the automation gets more valuable.

We build the automation too. Celeste learns from a company’s own support history, practices privately, and takes over one topic at a time as the owner graduates its trust, which is how the whole thing works. We are not pretending the displacement is somebody else’s doing. We are saying the payment should last as long as the saving does.

There is a simple test for any company announcing a transition fund this year. Ask when the payments stop. Then ask when the saving stops. If those two answers are not the same, it was a discount.