The Child Was Not the Weaver
In 1804 a machine in Lyon took a child's job. The child was not the weaver.
Before Joseph-Marie Jacquard's loom, a child called a draw boy sat on a plank above the loom, pulling cords by hand every time the weaver called for a thread lifted, all day, for years, until the pattern lived as much in the child's hands as anywhere else. Jacquard's punched cards replaced that child in 1804. A hole in the card lifted a thread. No hole left it down. The weaver kept weaving. The liseuse, whose trade it already was to read a design and translate it into the pattern of holes the card would carry, kept working too.
The reader will bring Luddites and mob violence to a piece titled after a loom. Hold that assumption. The record does not deny the fight had real cost. It puts the cost somewhere other than where popular memory does. The story of Lyon's weavers mobbing Jacquard's own machine comes down without a primary source I can find; Wikipedia's entry on Jacquard records only that silk weavers "fiercely opposed" the invention out of fear for their livelihood, so I am treating the mob as legend, not as evidence.
Here is the actual claim, compressed. Agents entering organizations right now is not a new event. It is the fourth time a pattern that a trained person's judgment produced has been decomposed into steps and written onto a portable artifact: cards in 1804, a priced computing hierarchy in 1832, a planning department by 1911, and now agents. Each time, the value and the risk did not travel together. Cheap execution went to the artifact. Unpriced accountability landed on whoever composed, routed, and checked what the artifact produced. I call this role the middle tier: the people who decide what the card, the formula sheet, the work order, or the prompt should say, and who answer for it when it is wrong.
The first three times, the fight over who owns the decomposed pattern and who absorbs the cost when it fails was settled by wages, by power, and at least once by force, not by good design. My bet, and it is a stated bet rather than a guarantee, is that this time it can be settled by design and bargaining instead. A disclosure before the argument: I have a commercial stake in how the middle tier gets designed, and Part 2 is where I will say exactly what it is.
This is Part 1 of a four part series called The Loom and the Loop, and its only job is to show that the middle tier is real and forming now. Part 2 takes up how to design work around it, Part 3 how to see inside what agents actually do, and Part 4 how what the middle tier learns gets taught forward without decaying.
The card replaced the child. It did not replace the person who decided what the card should say, and finding that person is where this history actually goes.
The Value Was Never at the Bottom
I start where the value and the risk first came apart: de Prony's room, 1791 to 1801. It held three kinds of people. Five or six mathematicians chose which formulas to use. Seven or eight people with what Babbage called "considerable acquaintance with mathematics" turned those formulas into simple steps of addition and subtraction that a less trained person could follow, and checked the results. Dozens more did nothing but add and subtract, all day, with, in the historian Lorraine Daston's words, "no machine more complicated than a quill pen," producing what Babbage's own account calls "seventeen large folio volumes" of logarithmic and trigonometric tables. Gaspard de Prony's own description of the project, in Daston's rendering, was that he meant to "manufacture my logarithms as one manufactures pins."
That is the value and accountability split in miniature. Charles Babbage did not run de Prony's room. He wrote it up in 1832, in On the Economy of Machinery and Manufactures, three decades after the room itself had closed, and his rule for pricing what he found there is worth quoting exactly: "purchase and apply to each process precisely that quantity of skill and knowledge which is required for it." Organizing and pricing are two different acts. A manufacturer no longer had to pay a master craftsman's wage for work a much cheaper person could do exactly as well. Applied to de Prony's own tiers, the bottom section needed almost no training and cost almost nothing, work Babbage called something that "may almost be termed mechanical." Daston goes further: "Human intelligence sunk to the mechanical level, kindling the idea of machine intelligence."
I stop a misreading here before it forms, because it is the natural one. De Prony found that the sheets with the fewest errors "came particularly from those who had the most limited intelligence, [who had] an automatic existence, so to speak." Read quickly, that sounds like praise for narrow work, evidence that simple people did the best job. It is not that. It is evidence that the work above them was well decomposed. The bottom tier made few mistakes because the seven or eight people above them had already done the hard part, cutting a complex formula into steps simple enough that following them correctly required no judgment at all.
Daston's own conclusion is the one worth keeping: "it is the middle tier, those seven or eight 'algebraicists'... who should be the object of our attention. They were the ones who dissected a highly complex formula into simple, step-by-step procedures." That is the middle tier this piece keeps naming, and the value sat there. Whether de Prony's own budget paid for it is a gap here, not a settled fact: no wage figures for his three tiers appear in the sources I have drawn on.
De Prony's own room already held three layers, so the separation this piece is tracking was never one clean break between a whole trade and a machine. It happens at whatever level currently holds the judgment nobody has written down yet, and each era of this story is really naming the level where that cut became visible to everyone outside the room.
The accountability half of the split earns its place in current research too, not just history. Nyilasy and his coauthors, across four experiments with 1,801 people in an April 2026 study, found that a human paired with an AI system is handed about 10 points more responsibility for the outcome, on the study's own scale, than a human paired with another human. Their own explanation: "AI is seen as a constrained implementer, which makes the human the default locus of discretionary responsibility." That is de Prony's room again, the cost unpriced, sitting on whoever answers for the sheet rather than whoever cut the formula. I have argued before that the composition question precedes the automation question, and that the O-ring logic of assembled work, that one weak link caps the value of the whole chain, is really Babbage's own pricing rule with the price of one link driven toward zero. This is where that logic lives: in the middle tier, not the bottom one.
De Prony organized the pattern with nothing more complicated than a quill pen. What he could not have known is that the principle he had organized would turn up decades later, woven into silk, on the wall of the man who had already priced it.