Vikram Singh doesn’t walk very fast. He talks fast. Thinks faster. And somewhere between the first and second kilometre, he says something that stops you in your tracks.
We were thirty minutes into our morning walk when he did exactly that. “Here’s what nobody is explaining properly,” he said. “If AI fires everyone, who buys the stuff?”
Seven words. The entire anxiety of 2026 in one sentence.
I’ve since read the economics. A paper by Falk and Tsoukalas at Penn and Boston University builds a formal model around exactly this question. They call it a demand externality. I’ll spare you the equations. Vikram said it better on a footpath in seven words.
The town that ate itself
Think of a small town. One factory, one high street, a few hundred families. The factory owner replaces half his workers with machines. His costs fall. His profits rise. Perfectly rational.
But those fired workers used to buy groceries. Eat at the corner restaurant. Buy their children’s shoes on the main street. One by one, those businesses feel it. Eventually the factory owner notices his own sales falling. His customers — it turns out — were his workers.
He could see this coming. He automated anyway. Because the factory on the other side of town was doing the same thing. If he waited, he’d be undercut and out of business before his competitor felt any pain.
Every firm is rational. Every firm can see the damage. Every firm automates anyway. Because the alternative is dying first.
That’s the trap. A Prisoner’s Dilemma at civilisational scale.
What the economists tried and failed
The paper works through every solution you’d reach for.
Retrain the workers. Helpful, but doesn’t change what firms do. Give workers company shares. Same problem, different pockets. Pay everyone a basic income. Puts money back in circulation but doesn’t stop the automation race. Ask companies to voluntarily slow down. Collapses immediately. Any firm that cheats gains an instant advantage, so everyone cheats.
The only thing that actually works, they argue, is a direct tax on each automated task. It forces firms to pay the full cost of the damage they cause and not just their small slice of it. A Pigouvian tax. Named after Arthur Pigou, a Cambridge economist who devoted his life to one question: what happens when your gain is everyone else’s problem?
Here’s how it works in plain English. When a firm fires a worker and replaces them with a machine, that worker stops spending money. Every other business suffers a little. But the firm that did the firing only feels a fraction of that pain. It kept all the savings. Everyone else shared the damage. The tax reverses that. You automate, you pay the real bill. Suddenly it’s not such an easy call. Nobody is banning robots. But you can’t quietly hand your costs to everyone else anymore.
The only remaining problem is convincing politicians to do it. And politicians, as a rule, are not famous for making brave decisions about things that haven’t quite gone wrong yet.
Where Keynes got it half right
In 1930, deep in the Great Depression, Keynes wrote an essay imagining life a hundred years ahead. By 2030, he said, technology would be so productive that we’d only need to work fifteen hours a week. The rest would be culture, leisure, living well. His worry — genuinely — was boredom. That we wouldn’t know what to do with all the free time.
He saw the same force at work. Technology replacing human labour. He just assumed the gains would spread evenly. That if machines did the work, everyone would share in the abundance.
He didn’t account for the factory on the other side of town.
Falk and Tsoukalas show what actually happens when AI fires everyone: the dishwasher gets bought, the person it replaced gets let go, and the restaurant slowly runs out of customers who can afford to eat there. Keynes imagined we’d all get to stop doing the dishes. The productivity was real. The distribution wasn’t automatic. And a hundred years on, we’re not working fifteen hours a week. We’re working like maniacs.
The leisure never arrived. Neither did the crisis. We’ve been living inside that tension without quite noticing it.
All good ideas come on walks
A critic named McEntire has since checked every line of the Falk-Tsoukalas paper. The maths holds. But he showed that the conclusions depend on specific assumptions. Change a few reasonable numbers and the catastrophe shrinks or disappears. The trap is real. How bad it gets depends on how the world actually works and economists still disagree about that.
So we don’t know how bad it gets. We know the mechanism exists. We know it’s already running.
Vikram spotted all of this in seven words on a morning walk.
Good ideas come on walks because walking has no agenda. You’re not performing. You’re not concluding. You’re just moving, thinking, and occasionally stopping mid-stride because something has become clear.
I need to keep walking with him.
