The machines took the weaving. They did not take the worth of cloth.
The worth moved.
When the power looms spread through England’s textile counties, the weavers who smashed them were right to be afraid and wrong about what they were losing.
Once goods were mass-produced and identical, you could no longer trust them because you knew the maker’s hands. Trust had to travel some other way, so it became a thing you could register and own. Britain’s first registered trademark, in 1876, was Bass’s red triangle: a reputation turned into property, standing in for the face you used to look in the eye (UK Intellectual Property Office).
That is the move every wave of automation makes, and we are inside another one.
AI is collapsing the cost of knowing things the way the loom collapsed the cost of weaving. The error, then and now, is to assume the value evaporates. It does not. Borrow a prior from physics: energy is never destroyed, only changed in form.
Value under automation behaves the same way. Pull it out of the knowing and it does not disappear. It reappears somewhere a machine cannot stand, and across three centuries it has reappeared in the same three places.
Trust is the first, and the looms already showed it. When making outruns the handshake, someone still has to be believable, and believable has only ever attached to a person, or to a name a person is willing to stake. A trademark is a portable piece of a human being’s standing. The model can generate the words on the label. It cannot be the one whose name is ruined if the label lies.
Accountability is the second, and we have never once automated our way out of it. We build more of it. In 1494 Luca Pacioli set down double-entry bookkeeping, not to speed up arithmetic but to make a merchant answerable: every entry meeting a counter-entry, every figure with a person behind it. The notary, the audited ledger, the surety bond are the same invention on repeat, machinery for locating the human who pays when it breaks. You cannot sue a probability distribution. You cannot strike its name from the register.
The third we are shyest to say out loud: people want to be dealt with by people. When the ATM arrived, the forecast was that the teller was finished. The opposite happened. Cash machines made branches cheap, cheap branches multiplied, and the teller’s work shifted from counting twenties to knowing the customer. The number of bank tellers grew through the 1990s and 2000s, faster than the workforce as a whole (Bessen, 2015). The machine took the rote half and handed back the human half: the person at the window who is glad you came in.
There is a clean result from economics underneath this. Two hundred years ago Ricardo showed that when one party gets far better at a task, the other is not made worthless. It moves to its comparative advantage. As the machine takes the knowing, our advantage is turning out to be each other: the trusting, the answering, and the plain want to be served by someone capable of being glad you exist.
The obvious objection is that this time is different: past machines took muscle, this one takes judgment, so the human layer is not safe, only next. But trust, accountability, and welcome are not skills the model happens to lack. They are roles that need a who, not a what. Trust needs someone whose name can be ruined. Accountability needs someone who can be made to pay. Gladness needs someone who can actually be glad. A model can produce the words for all three and occupy none of them, because there is no one inside it to shame, to sue, or to mean it.
You can automate the doing. You cannot automate the being.
So when people ask what survives, the honest answer is not a tidy abstraction. It is a direction the record has pointed for three hundred years, from the loom to the ledger to the teller’s window: away from the doing, toward the trusting, the answering, and the welcome. That is the part I would stake a career on. I have. The looms are loud again. The thing worth listening for is what they hand back.