Chapter 19 The Long Road to the Vector

1. The road

This book has argued that the department was a workaround for a constraint, that the constraint has gone, and that a company can collapse to a chain of agents held by one pair of humans by default at the gates. A reader who has come this far is entitled to ask whether anyone has ever done anything like it, and the honest answer is that everyone has, for two thousand years, a step at a time, and stopped.

The road to the vector is long and it is well marked. At each milestone someone took a skill that lived in a person and moved it into something that was not a person: a drill, a yard, a sequence, a line, a machine that stopped itself, an interface, a model. Each time, the result needed fewer people than before, often spectacularly fewer, and each time the people who were left were reorganised around the new machine in exactly the old shape, with layers above them and handoffs between them, and the collapse stalled a step short of the result. Eight of those milestones are told here, properly, as the stories they are. They are not chosen because they failed. They are chosen because each one succeeded at the thing it attempted and then stopped, and the place where each stopped is a lesson the doctrine has already absorbed.

The thread through all eight is one sentence. Every station moved skill out of people and into a machine or a drill, so every station needed fewer people for the same result. That is also the sentence this book is built on, and the vector is where the sentence finally runs to its end: a chain that holds every skill a result needs, and one pair of humans at its gates. The eight stations show the road; the next chapter hands the reader the keys and a month to drive it.

2. The legion

Late on an afternoon in Gaul, in the first century before Christ, five thousand men stop walking.

They have covered twenty miles since dawn with their armour on their backs, and before they eat they will build a town. Surveyors have gone ahead and planted a white flag where the commander’s tent will stand; from it, with measuring rods and coloured flags, they have laid out the two main streets, the lines of each cohort’s tents, the gaps between them, and the rectangle of the rampart, always the same, always in the same proportion to the same places. Each man arrives and knows where his tent goes without being told, because it went there last night and the night before and will go there tomorrow; Polybius, who watched them do it two centuries earlier, said the whole army took its place like a city’s people coming home, each to his own street and his own house.⁶⁰ Some dig the ditch, some cut the turf for the rampart, some set the palisade stakes they have carried all day for this purpose, and in the time it takes the sun to go down the five thousand are behind a ditch and a wall, in streets, with a gate on each side and a guard on each gate.

Nobody in that camp is a hero. That is the point of it. The Gauls on the other side of the valley have men who have fought since boyhood and can do things with a long sword that no legionary can do with a short one. The legion does not need them to. Its men are farmers’ sons with two years of drill, and the drill has taken the skill of fighting out of each of them and put it into the unit: eighty men to a century under a centurion who has done this for twenty years, six centuries to a cohort, ten cohorts to a legion, eight men to a tent who march, dig, cook and fight as one thing. Josephus, a Jewish general who fought them and lost and then lived among them, wrote that their drills were bloodless battles and their battles bloody drills, and that there was no disorder, no rout, no surprise that could reach them, because nothing they met in battle was anything they had not already done a hundred times in the yard.⁶¹

The legion is the first station because it is the first time a civilisation deliberately took the most prized skill it had, the skill of the warrior, and decided it would rather have it in a drill than in a man. The drilled unit beat the skilled individual every time it met him, for four hundred years, with ordinary men.

What collapsed. The individual skill of the fighter, into the drill of the unit; the knowledge of how to camp, march, build and fight, into a routine every man carried and no man owned.

What stayed. Every skill still lived in a body, and had to be put into each new body by two years of the same drill. Above the eight men in a tent stood the century, above the century the cohort, above the cohort the legion, above the legion the army and the Roman state, a pyramid of rank as steep as any that came after it, with the drill as its base.

The number. Five thousand men who built a fortified town every evening of a campaign, with a gate on each side, before they ate; and a legion of farmers that was expected to beat a host of warriors twice its size, and usually did.

Why it stalled. Because the drill lived in men, it died with them. Four centuries later a late Roman writer named Vegetius compiled a book pleading for the restoration of a drill the army had let lapse, gathering it from older writers and imperial regulations, because the drill itself was no longer practised.⁶² A drill that must be re-taught to every recruit is a machine that is rebuilt from scratch every generation. And a legion’s result scaled only one way: more result meant more legions, each of five thousand men, until the empire could not pay for them. The skill had left the man, but the man was still the unit of output.

3. The Arsenal

In the spring of 1438 a Castilian traveller named Pero Tafur stood in Venice at the gate of the Arsenal, the Republic’s shipyard, and watched ten galleys come out armed between the hours of three and nine.

He wrote down what he saw, which is why we have it. Inside the walls was a long canal, and on each side of the canal stood the warehouses. A galley hull was towed along the water, and as it passed each warehouse the men inside handed out onto the deck whatever that warehouse held: from one, the cordage; from another, the bread; from a third, the arms; from a fourth, the ballistas and mortars; and so down the line, so that by the time the hull reached the end of the canal it carried its oars, its stores, its weapons and its men, and went out of the gate complete. In this manner, he wrote, ten galleys fully armed came out between the hours of three and nine.⁶³

Tafur had seen an assembly line, five hundred years before the phrase, and he had seen something the legion did not have: a machine that was not a man. The Arsenal was a yard, a canal and a set of warehouses laid out so that the work came to the parts rather than the parts to the work, and it ran on standard pieces. Venetian war galleys were built to a pattern so that a mast, a rudder, an oar or a bench from one would fit another, which meant the Republic could keep finished hulls under cover in reserve and fit them out in days. In the summer of 1570, with the Turks landing on Cyprus, the Arsenal fitted out and sent to sea some hundred galleys in about two months, a feat no other state in Europe could have attempted.⁶⁴ The Arsenal in those years employed perhaps two or three thousand craftsmen in ordinary times and several times that in emergencies, and the city thought of it, rightly, as the heart of the state.

The craft of the shipwright had been collapsed into a pattern, a stock of parts and a sequence of warehouses. A hull could be armed by men who had never built a ship and would never sail one.

What collapsed. The shipwright’s craft, into standard parts and a fitting-out line; the knowledge of what a galley needed, into the order of the warehouses along the canal.

What stayed. The men who built the hulls were the arsenalotti, organised in guilds, with positions that passed from father to son and pay that came by the day for attendance. Above them stood the Patroni and the Provveditori, the Senate’s committees of patricians who ran the yard in rotation, and above them the Senate. The design of a galley lived in the head of its master shipwright, the proto, and the Republic lived in fear of losing one.

The number. Ten armed galleys in six hours in 1438; a hundred in two months in 1570.

Why it stalled. The Arsenal had one customer, the state, and it could not go elsewhere. The state bought whatever it made at whatever pace it made it, so its number was never the one a customer would have asked for, and when the Turkish threat receded the yard kept its men and lost its pace. By the seventeenth century the arsenalotti were a privileged corporation paid to be present, the daily roll was larger than the work, and the historian of the yard records that the proportion of the day spent working fell with every decade.⁶⁵ And the design still lived in a man. When a great proto died, so did the galley he had carried in his head, and the Republic’s ships got worse. A machine whose pattern lives in one person’s memory is a department with a canal.

4. The pin factory

In 1776 Adam Smith opened the most influential book on wealth ever written with the smallest manufacture he could think of.

A pin. A man not educated to the business, he wrote, nor acquainted with the machinery it used, could scarce with his utmost industry make one pin in a day, and certainly could not make twenty. But in the way the trade was now carried on, one man drew out the wire, another straightened it, a third cut it, a fourth pointed it, a fifth ground the top to receive the head; making the head took two or three distinct operations; putting it on was a trade of its own; whitening the pins another; putting them into paper another still. Eighteen operations, in some factories eighteen men; in the small one he had seen, ten men, some doing two or three operations each, and those ten men made upward of forty-eight thousand pins in a day. Four thousand eight hundred pins a man, where a man alone could not make twenty.⁶⁶

Smith knew exactly what he had found. The division of labour, he said, was the cause of the greatest improvement in the productive powers of labour, and it worked for three reasons: each man grew dexterous at his one motion; no time was lost passing from one kind of work to another; and a man who did one thing all day was the man most likely to invent a machine to do it for him. Fifty years later Charles Babbage, who built calculating engines and liked to count, went back to the pin factory with a stopwatch and a wage book and found a fourth reason, which has his name: once the work was split, each operation could be bought at exactly the skill it needed, so the master paid for a child’s hands where a child’s hands would do and for a craftsman’s only where they would not.⁶⁷ The pin factory, between the two of them, is the birth certificate of the department. Every function that chapter 2 described, sales and finance and the rest, is a pin-maker’s motion scaled up to a building.

What collapsed. The pin-maker’s whole craft, into eighteen motions that no single man any longer needed to know.

What stayed. A man at every motion. And something new, which the legion and the Arsenal had both kept small and the pin factory made the organising fact of industry: between every pair of men, a handoff. The pin passed from hand to hand wherever one man’s operations ended and the next man’s began, and each pass was a place it could wait, be dropped, or be counted twice.

The number. One pin a day, or at most twenty, for a man alone; four thousand eight hundred for a man in the line of ten.

Why it stalled. Smith saw it himself and put it in Book Five, where fewer people read: the man whose whole life is spent performing a few simple operations has no occasion to exert his understanding and becomes as stupid and ignorant as it is possible for a human creature to become.⁶⁸ The division of labour bought dexterity with attention. And the handoff it created needed a man to watch it, and then a man to watch him, and the pyramid of chapter 2 grew out of the pin factory’s eighteen steps as surely as the pin grew out of the wire. The skill had left the craftsman and gone into the sequence, but the sequence was made of men, and men between men need managing.

5. The line

In the spring of 1913, in a four-storey factory at Highland Park outside Detroit, a group of Henry Ford’s engineers tried something with the flywheel magneto.

Until then one man had assembled a whole magneto at a bench, in about twenty minutes. The engineers broke the assembly into twenty-nine operations, put twenty-nine men in a row, and slid the magneto from one to the next along a waist-high rail. The time fell to thirteen minutes, then, when the rail was raised eight inches so that nobody stooped, to seven, then, when a chain pulled the work at a fixed pace, to five.⁶⁹ They tried the same thing with the engine, and then, in October, with the thing itself. A chassis that a team of men had assembled in one place in twelve hours and twenty-eight minutes was pulled along the floor by a rope and windlass past stations of men who each did one thing, and by the following spring the chassis took one hour and thirty-three minutes.⁷⁰ The price of a Model T, which had been over eight hundred dollars in 1908, would fall, for the runabout, below three hundred by 1925, and Ford would build more cars than every other maker in the world combined.

Then the men left. In 1913 Ford had to hire more than fifty-two thousand workers to keep a workforce that at any moment numbered fourteen thousand, a turnover of several hundred per cent, because a man who had been a mechanic would not stand in one place for nine hours doing a seven-second motion for a wage anyone would pay him.⁷¹ Ford’s answer, in January 1914, was the five-dollar day, more than double the going rate, and it stopped the bleeding overnight. But it came with a condition that is the most instructive part of the whole story. Nearly half the five dollars was paid only to men who met the company’s standards for clean living, and to find out who did, Ford created a Sociological Department whose investigators visited the workers’ homes, asked about their drinking, their marriages, their savings and their boarders, and recommended who was fit to be paid.⁷² Ford had moved the skill of making a car into the line. He then tried to move the attention of the men who stood on it into a department that inspected their lives.

What collapsed. The mechanic’s craft, into a sequence of motions and a moving line; the skill of assembly, into the machine tools and the pace of the chain.

What stayed. The man as a part of the machine, with his hands at a station and his attention, which the line did not need and could not use, unmeasured except by whether he came back on Monday. The foreman above him and the pyramid above the foreman, which grew at Ford faster than at any company before it. And a company that, having found no way to read a man’s attention from his work, read it from his home instead.

The number. Twelve hours and twenty-eight minutes to one hour and thirty-three. Fifty-two thousand hires for fourteen thousand jobs.

Why it stalled. The line was a machine for hands, not for attention, and the attention of the men on it went where unmeasured attention goes. Ford’s own company became, within a generation, the most layered and least nimble of the great American firms, so clotted with departments that after the Second World War a team of young outsiders had to be brought in to reorganise it, in departments.⁷³ And the surveillance was the doctrine’s warning two generations early: a company that cannot read attention from the result will read it from the person, and the person will leave.

6. The loom that stopped

In 1924 Sakichi Toyoda, a carpenter’s son who had spent his life on weaving machines, finished a loom that stopped itself.

The Type G automatic loom shut down the instant a thread broke or ran out, by itself, without a hand on it. That sounds like a small thing and was the beginning of everything that followed. Before it, a weaver stood at a loom to catch the break; after it, one worker could watch many looms, because the looms did their own watching and called for help only when they needed it.⁷⁴ Sakichi called the idea jidoka, automation with a human touch, and his family carried it from looms into cars. The inspector had been moved into the machine.

Thirty years later Taiichi Ohno, running a Toyota plant that could not afford the inventories of the Americans it was trying to catch, took the loom’s lesson to the whole floor. A cord hung above every station, and any worker who saw a defect pulled it and stopped the line, so that a fault was found where it was made and not three departments later. Parts moved only when the next station asked for them, by a card, a kanban, that Ohno said he took from watching an American supermarket restock a shelf; the schedule left the planner’s office and became a signal on the floor. Every task was written down as standard work by the team that did it and improved by them week by week, so that the drill, for the first time since the legion, was written and revised rather than remembered.⁷⁵

The result was measured in 1986 by a team from the Massachusetts Institute of Technology who compared General Motors’ plant at Framingham with Toyota’s at Takaoka. Framingham needed about thirty-one hours of assembly labour per car and shipped a hundred and thirty assembly defects per hundred cars; Takaoka needed about sixteen hours and shipped forty-five. Framingham kept two weeks of parts beside the line; Takaoka kept two hours.⁷⁶ Half the assembly hours, a third of the defects, and a sliver of the line-side inventory.

What collapsed. Inspection, into the machine and the cord; scheduling, into the card; the manager’s knowledge of the work, into standard work written by the team itself; a great deal of inventory, into a signal.

What stayed. The hands. Toyota’s plants were full of people, and the person remained the unit of production, with a team leader for every five or six, a group leader above the team leaders, and the ordinary Japanese hierarchy above that. Sales, design, purchasing and finance stayed as departments, as they were everywhere. And the drill that had been written down still had to be learned by people and lived in their habits, which the company itself called the Toyota Way, as if it were a culture rather than a method.

The number. Thirty-one hours against sixteen; a hundred and thirty defects against forty-five; two weeks against two hours.

Why it stalled. Toyota’s system reached well beyond the shop floor, into product development and suppliers, but it kept the office’s hierarchy and its human coordination, and the office is where this book began. More deeply, the method was carried in people, and people transplant slowly: when Toyota taught it to an American plant in California in the 1980s it worked, and it took years, and most of the companies that tried to copy it from books failed, because the drill was in the habits and the habits were in the men.⁷⁷ A loom that stops itself was the first agent. A floor of them, each holding its own standard, each stopping its own line, each improving its own work without a person’s habits to carry it, is the vector.

7. The interface

In 2002 Jeff Bezos sent a memo to Amazon’s engineers that one of them later summarised from memory, because it was short enough to remember.

All teams would henceforth expose their data and their functions through service interfaces. Teams would communicate with each other through those interfaces and in no other way: no direct links, no shared databases, no reading another team’s data, no back doors. Every interface would be built, from the start, as if it might be opened to the outside world. Anyone who did not do this would be fired.⁷⁸ At the same time Bezos reorganised the company into teams small enough to be fed by two pizzas, each owning one thing and answerable for it. What he had done, in effect, was to take the handoff that the pin factory had invented, the point where one group’s work passed into another group’s hands, and replace it with a contract that a machine could enforce. One team did not ask another team for anything. It called an interface, and the interface either answered or it did not, and the record showed which.

A decade later Amazon did to its warehouses what it had done to its software. In 2012 it paid seven hundred and seventy-five million dollars for a company called Kiva, whose squat orange robots carried shelves of goods to a picker standing still, instead of the picker walking miles a shift to the shelves.⁷⁹ By 2023 Amazon had more than seven hundred and fifty thousand robots of many kinds, and in 2025 it announced its millionth, in a company that employed about one and a half million people, with nearly as many machines on its floors as people.⁸⁰

What collapsed. The handoff between teams, into an interface; the walking of the picker, into a robot; a thousand small negotiations between departments, into contracts that answered or did not.

What stayed. The people, in unprecedented numbers, and the layers above them. Amazon’s teams owned services, not results; a customer’s order still crossed dozens of teams between the click and the doorstep, and no team owned the whole of what the customer felt. Headcount rose with revenue over two decades, from the tens of thousands to a million and a half, which is the opposite of what a collapse does. The company became the largest and best-run department store in the history of the world, with the emphasis on department.

The number. Seven hundred and seventy-five million dollars for the first robots; a million of them thirteen years later, beside a million and a half people.

Why it stalled. The interface collapsed the message, not the department. Teams that can only talk to each other through contracts are cleaner than teams that talk in meetings, but each still owns a piece of the work rather than the result, and a company organised around pieces grows people as it grows. The doctrine takes Amazon’s interface and gives it to the agents inside one chain, so that the only human boundary left is the gate where the result becomes the customer’s. Amazon built the agents’ language. It kept the departments that spoke it.

8. The assistant

On 27 February 2024 a Swedish payments company called Klarna announced that an AI assistant it had switched on a month earlier had handled 2.3 million customer conversations, two-thirds of all its customer-service chats, in thirty-five languages, around the clock. It was doing, the company said, the work of seven hundred full-time agents. Resolution times had fallen from eleven minutes to under two. Repeat enquiries were down a quarter. The company expected it to add forty million dollars to that year’s profit.⁸¹

Chapter 2 cited that announcement as proof that the constraint had gone, and it was. A department’s work had moved into an agent in a month. For the first time on this road, the skill that moved was not the skill of hands or of sequence but of conversation and judgement, the skill the office had always believed was its own. Klarna froze hiring and let attrition take its course, and over the following year its headcount fell by about a fifth, to around three and a half thousand.

Fifteen months later the same company’s chief executive told an interviewer that the company was hiring human customer-service staff again. The AI-first approach, he said, had produced lower quality; it had not been the right path; investing in the quality of human support was the way of the future for the company, and it was critical that a customer always knew a human was there if they wanted one.⁸² Klarna later told the American securities regulator that the remark about lower quality had been misreported and concerned a pilot of human service, not its AI, which it said had handled 66 per cent of its chats in the twelve months to March 2025. The reader should weigh both accounts. The public record does not settle what happened inside the company, and what follows is my reading of it, not a finding.

This is the most important station on the road, and it is important precisely because of the reversal. On my reading, nothing had gone wrong with the agent; what had gone wrong was the organisation around it. As far as the public account shows, Klarna had dropped an agent into a department and measured it by the department’s number, handling time and cost, and the agent had optimised the number it was given, as agents do. Nobody owned the result the customer felt. There was no vector with a human at its gate reading the record for the thing the number did not show; there was a customer-service function, with an agent inside it where the people had been, and when the company wanted better service it said it would hire more people for customer service, the old remedy. The agent existed. The collapse had not been designed.

What collapsed. Two-thirds of a department’s conversations, into one agent, in a month; the skill of answering a customer, which the office had thought could not leave a person, into a model.

What stayed. The department. The agent was placed where the people had been and measured as they had been, by cost and speed, with no human at a gate owning the result and no record read for what the number missed.

The number. Seven hundred people’s work; eleven minutes to two; and then, fifteen months later, a hiring notice.

Why it stalled. An agent without an aligner optimises what it is measured by. The doctrine’s answer to Klarna is not fewer agents or more people. It is one pair at the gates of a vector whose result is the customer’s, reading the record for quality every morning, paid on the result and not on the handling time, with the agent inside the chain where the skill belongs and the human where the judgement belongs. Klarna proved the agent in a month and disproved the department in a year, and the doctrine thanks it for both.

9. The cut

On 26 October 2022 Elon Musk walked into Twitter’s San Francisco headquarters carrying a sink; the next day he completed his purchase of the company. Within a week he had dismissed about half of its roughly seven and a half thousand employees. By April 2023 he told the BBC that the company had about fifteen hundred people left, down from just under eight thousand, a cut of around eighty per cent.⁸³

The service kept running. There were outages, and there were days when features broke and were fixed by whoever was left, but the thing a few hundred million people opened every morning opened every morning, posted, replied and scrolled, with a fifth of the people who had been held to be necessary to it. Musk had a method for this, which he had used in his rocket and car companies and which his biographer wrote down as a five-step algorithm: question every requirement and find out which person asked for it; delete any part or process you can, and if you are not adding back at least ten per cent of what you deleted, you have not deleted enough; only then simplify and optimise what remains; then accelerate the cycle; then, last, automate.⁸⁴ Applied to a company, the algorithm deleted managers, meetings, teams whose work nobody could trace to the product, and whole departments whose results nobody missed when they stopped. Of all eight stations this is the one that most resembles the collapse this book describes, because it was done to the office, not the floor, and it was done in weeks, and the thing still worked.

And it stalled, for a reason the doctrine can name exactly. Eighty per cent of the people went and nothing was built in their place. On the public record, and on my reading of it, there were no vectors, no playbook, no record that anyone but the owner read, and no gate that carried a name other than his. Every decision that the deleted layers had once made rose to one man, who made them at all hours in public, and its advertising revenue, which depended on advertisers who needed to be able to predict what the service would do tomorrow, fell by about half within a year, by Musk’s own account.⁸⁵ A collapse by subtraction leaves a smaller company with one master and no chain. The doctrine’s six are a design. Twitter’s fifteen hundred were a remainder.

What collapsed. Eighty per cent of the headcount, with the service still running; the layers of management between the owner and the engineers; the departments whose results nobody missed.

What stayed. Every signature in one man’s hands. No vectors, no record anyone else held, no gates with other names, so that the company’s judgement was one person’s and its attention was his.

The number. Just under eight thousand to about fifteen hundred, in six months; advertising revenue down by half.

Why it stalled. Deletion is half the doctrine. The other half is the chain, the gate, the record and the pair, and without them the company that remains is not collapsed but merely small. Musk proved that four-fifths of a modern office was a workaround, which this book had argued but no one had demonstrated on that scale. He did not prove what to build in its place, because he did not build it.

10. The tally

Eight stations, two thousand years, one sentence. Here is the sentence, station by station.

Read down the second column and the thread is plain: at every station the skill left the person and went into something that could hold it without him, a drill, a yard, a sequence, a line, a self-stopping machine, an interface, a model. Read down the third and the consequence is plain too: fewer people, every time, often by a factor that would have sounded absurd the day before. Read down the fourth and you have the reason this book exists. Every station kept the shape of the pin factory around the new machine: people at the stations, handoffs between them, layers above them. The skill collapsed; the organisation did not.

And this is where the road has brought us. The figures chapter 2 laid out are the state of the road today, two centuries after the pin factory and one after the line: in the American economy one manager, supervisor or administrator for every 4.7 other workers, and the cost of excess management reckoned at three trillion dollars a year;⁸⁶ knowledge workers spending fifty-seven per cent of their time communicating and forty-three per cent creating;⁸⁷ twenty-one per cent of employees worldwide engaged in their work, by Gallup’s count, down two points in a year;⁸⁸ and consultants at one of the world’s best firms, handed a language model, completing twelve per cent more tasks a quarter faster at forty per cent higher quality, with the bottom half of the firm gaining most, and doing worse on the one task chosen to lie beyond the model’s reach.⁸⁹ The office has had its loom and its line, in the model, and has not yet had its Takaoka. The vector is the office’s line. The eight stations show every piece of it being invented, one at a time, and the doctrine is the first design that uses all of them at once and stops the organisation from growing back around them.

11. Rules for reading the road

1. Every great gain in output in history came from moving a skill out of a person and into something that was not one. There is no exception on this road, and the vector is not one either.
2. Each move needed fewer people for the same result, and the factor was always larger than the people of the day believed possible.
3. Each move was organised, afterwards, in the shape of the pin factory: people at stations, handoffs between them, layers above. That shape is what stalled every station, not the machine.
4. A drill that lives in people must be re-taught to every generation and transplants slowly. A drill that lives in the machine does not. Toyota is the proof of the first; the vector is built on the second.
5. A machine with no customer feeling its result loses its pace. The Arsenal is the proof. Every vector has a customer and a Monday test.
6. A machine for hands that cannot read attention from the work will read it from the person. Ford is the proof, and the doctrine forbids it.
7. An interface between teams collapses the message, not the department. Amazon is the proof. The doctrine gives the interface to the agents and keeps one human boundary: the gate.
8. An agent dropped into a department optimises the department’s number. Klarna, on my reading, is the example, and the fourth form of chapter 4. The doctrine puts the agent in a chain and a pair at the gate, paid on the result.
9. Deletion without design leaves a smaller department store. Twitter is the proof. The six are built, not left over.
10. Nobody on this road was wrong. Each stopped where the organisation of his day stopped him. The reader has no such excuse, because the pieces are all on the table, and the next chapter is where he picks them up.

12. A note on sources

The eight stations are told from primary sources where they exist, and they exist for all but the legion’s drill, where the sources are the two ancient writers who saw it. Polybius described the Roman camp and the legion’s organisation in the sixth book of his Histories in the second century before Christ; Josephus described the drill in the third book of The Jewish War, having fought the legions and lost. Pero Tafur’s account of the Arsenal is in his own travel book, written in the 1450s about a visit of 1438; the 1570 mobilisation is from Frederic Lane’s histories of Venetian shipping, which also record the Arsenal’s decline. Smith’s pin factory is in the first chapter of The Wealth of Nations and his warning about its cost in the fifth book; Babbage’s extension is in his own book of 1832. Ford’s times are his own, from the autobiography he wrote with Samuel Crowther; the turnover and the Sociological Department are from Stephen Meyer’s history of the five-dollar day. Toyota’s method is from Taiichi Ohno’s own book and the Massachusetts Institute of Technology study published as The Machine That Changed the World; the Framingham–Takaoka figures are that study’s. The Amazon memo is known from Steve Yegge’s 2011 recollection of it, which this book attributes to him, and the robot figures are the company’s own announcements. Klarna’s figures are from its press release of 27 February 2024 and its chief executive’s interview with Bloomberg in May 2025. Twitter’s headcount is Musk’s own, to the BBC in April 2023; the algorithm is from Walter Isaacson’s biography; the revenue fall is Musk’s own statement. The chapter 2 figures are re-cited at their original sources. Where a figure is an estimate or a range in the sources, the chapter uses the more conservative end and says “about”.