Better at Everything

That sentence feels like news. It is not. It is the oldest question in economics, asked and answered before the telegraph existed. The reasoning everyone runs — it is better than me, therefore it replaces me — has been tested in the one arena where it could be tested for two hundred years, which is trade between countries. It has never once held. Not because the better party shows mercy, but because better turns out to be the wrong variable. Nothing in the arithmetic of who does what ever reads it.

The country that was better at everything

In 1817 David Ricardo published a two-country example that has been breaking intuitions ever since. England and Portugal each make cloth and wine. A unit of cloth costs England the labour of a hundred men for a year; wine costs it a hundred and twenty. Portugal makes the wine with eighty men and the cloth with ninety. Portugal is better at both. Flatly, measurably better at everything on the table.

Instinct says Portugal should make everything and England should make nothing. Ricardo showed the opposite: Portugal should put every hour into wine and buy its cloth from England — the worse weaver — because every Portuguese hour spent at a loom is an hour its vineyards would have turned into something worth more. The cloth does not cost Portugal ninety men. It costs Portugal the wine those ninety men did not make. England’s hours, outclassed everywhere, give up almost nothing to weave. So the weaving flows to England, not as charity but as arithmetic. The word better appears nowhere in it. The only quantity doing any work is what each hour forgoes.

A century and a half later, the mathematician Stanisław Ulam challenged Paul Samuelson to name one proposition in all of social science that was both true and non-trivial. Samuelson’s answer was this one. True, he said, need not be argued before a mathematician; non-trivial is attested by the thousands of intelligent men who could never grasp it, or believe it once it was explained.

You already live by it. A surgeon may be the best blood-draw in the hospital. She still never draws blood, and nobody calls that her obsolescence at the needle — her hour in the operating room is simply worth too much to spend there. The phlebotomist’s job is not protected by being better. It is created by the price of the surgeon’s hour.

The machine’s other hour

Now run the fear through the theorem. Grant the whole premise — grant that the machine becomes better than you at everything you do, which is an argument about the future, not yet a record. The conclusion so there is nothing left for me needs one more assumption, and nobody ever says it out loud: that the machine’s hour is free. It is not. The machine’s hour is made of chips, power, and buildings, and right now it is so far from free that it is rationed in plain sight — usage caps, priority tiers, rate limits, enterprises deciding which work gets the frontier model and which gets the cheap one. A queue in front of intelligence is a price wearing a different coat.

As long as that hour has a price, Ricardo’s arithmetic runs. Every hour of compute spent on your task is an hour not spent where the machine’s advantage is overwhelming — the protein fold, the grid balance, the merger, its own successor. So intelligence flows to where its edge is largest, exactly as Portugal’s hours flowed to the vineyard. And everything else falls back across the table to us — including an enormous amount of work the machine would do better than the person who ends up doing it.

That is the consequence almost everyone misses. You will keep work the machine does better, and it will not be sentiment, regulation, or mercy. It will be the same arithmetic that kept England weaving. Your work is not protected by what you are good at. It is assigned by what the machine’s hour is worth somewhere else — which means the number that sets the wage of thinking is no longer in your résumé. It is the meter running in the datacenter.

1. You will keep doing work the machine does better.

   The tell is already audible: why doesn’t the AI just do this? For the next decade, the honest answer will usually be a price, not a capability. People will read that as the machine’s weakness. It is the machine’s opportunity cost.

2. The wage of thinking starts tracking the price of intelligence.

   Cognitive wages will begin to decouple from skill and couple to the going rate for the machine’s hour. When a new model launches and eats the world’s capacity, human work will rebound for a season — and almost nobody will connect the two events.

3. The machine-hour gets a ticker.

   Rationing is opportunity cost made visible, and today it is published as rate limits and usage caps. Someone will quote the price of an hour of frontier intelligence the way oil and freight are quoted, and businesses will plan hiring against it.

4. Substitution follows the gradient, not the level.

   Jobs will not disappear where the machine is good. They will disappear where its advantage is largest and persist — even multiply — where its advantage is merely large. That is why entry-level evaporates in text first while hands, presence, and liability hold, and it is the theorem’s signature, not a paradox.

5. “Too valuable to use” becomes a normal thing to say about AI.

   Firms will routinely decline to point their best models at easy work, the way no hospital staffs its surgeon at the blood-draw chair. The refusal will look like stinginess. It will be allocation working.

I live inside this arithmetic every day. The machine I publish with writes faster than I do, and in some registers it writes better; it builds what I could not build and checks what I would not catch. What I keep is not what I am best at. It is the work where my hour gives up the least: deciding once what the business means, signing what has to be sworn to, saying yes. I wrote about handing over the doing in You Do It. You Decide. and about where judgment retreats in The Second-Cheapest Way to Find Out. This piece is the arithmetic underneath both.

How this could be wrong

Already true

Intelligence is rationed today. The labs cap usage and sell priority because their capacity is finite, and they point their own scarcest compute at their own highest-value work — training the successor — while selling the remainder. Enterprises already triage which tasks get the frontier model. Allocation by opportunity cost is not a forecast; it is this quarter’s invoice.

What has to happen

Demand for intelligence has to keep outrunning the supply of compute, so the machine’s hour keeps a price. Chips, power, and land have to stay scarce enough to meter, and prices have to stay the thing that allocates them. Every signal I can see — the datacenter buildout, the power contracts, the waiting lists — says demand is winning by miles.

Where I am probably wrong

Ricardo’s theorem assumes the better party’s capacity is finite. If the marginal machine-hour ever becomes too cheap to meter — genuinely free, not just cheaper — then nothing is forgone when the machine does everything, and comparative advantage stops assigning humans paid work at all. My bet is that abundance keeps receding, because every drop in the price of intelligence has so far created more demand than it satisfied. And a second, harder honesty: the theorem promises assignment, not a living. It guaranteed Portugal’s weavers work in the vineyards. It never promised the vineyard paid what the loom had.

Go back to the laptop at 9:14. The draft on the screen is better than yours, and that fact decides nothing, because it decided nothing in 1817. Somewhere in a building you will never visit, a meter is running on what that machine’s hour could have earned doing something else, and for as long as that meter reads above zero, the work comes back across the table to you. The question was never whether it is better.

It was never about better.