AI-Proof: What Machines Can Never Be — Volume II: Application, Part V: Conscious Strategy, Chapter 24: Speed as a Weapon
Part V looks at strategy. After the decade and the week in Chapter 21, the middle in Chapter 22 and the price of checking in Chapter 23, this chapter asks what the speed of the work itself is worth when everyone has it.
A designer built in nine days what once took eight weeks. Six weeks later the rose on the shelf was the wrong colour. When everyone is fast, speed wins nothing. What matters is where you slow down, and what the slowness checks.
In Brief For readers who work fast with machines and wonder where they should slow down. Chapter 24 asks what speed is still for when everyone has it, and who decides where work must slow down. A designer in Mumbai makes a summer drinks line in nine days and pays for a colour nobody checked in ink; a trainer of health workers on the Kenyan coast can write a module in a night but cannot hurry the reports that tell her whether it works. A Roman emperor’s sayings, Aristotle on good deliberation and the Dhammapada’s word for haste, which is also its word for force, help separate the speed of making from the slowness that committing needs.
When everyone is fast, speed wins nothing by itself; be quick where a choice can still be undone, slow where it cannot, and make the slowness a check, not a rest.
Opening: The Rose in Thane
The photograph showed a fridge in a supermarket in Thane, on the edge of Mumbai, under a row of tube lights. On the middle shelf stood a new bottle, a sparkling rose drink, the first of a summer line. Its label was meant to be a dusty rose, the colour of diluted rose sharbat in an old brass cup. Under those lights it was warmer and flatter, closer to brown.
Under the photograph was one line. “Is this our rose?”
It came at the end of March 2026, from the marketing director of a drinks company. Fourteen months earlier Arjun Mehta, a brand and packaging designer with two rooms above a printing shop in Parel, had designed that company’s rose sharbat label. It carried one image, a brass cup seen from above, and one colour. He had taken the colour from a sharbat maker on Mohammed Ali Road. Now the company wanted a whole summer line from him: three drinks and their labels, a display for shop shelves, and a guide telling its own staff how to use the new look. It gave him nine days, from the end of January.
Three years earlier the same job would have taken him about eight weeks. This time a machine did much of the making. On the first day he described the line to it and asked for directions. By the second evening he had more than he could have drawn in a month: marks for the brand, colour systems, ways of setting the type, labels for bottles of three shapes. Most were competent, and most were like one another. In one of them the machine had set his old rose beside a slate grey he had never tried. The grey made the rose look cooler, more like something to drink than something sweet. He took the pairing and worked inside it. He asked for variations, threw most of them away and kept a few. The machine did the rendering. He did the choosing.
The director’s team read each round on their screens and sent notes the same afternoon. By the middle of February the company had its identity, its three labels, its shelf display and its guide. She liked the work, and so did he. The nine days had felt like flight.
The six weeks that followed felt like gravity. The guide gave the rose only in the numbers a screen uses, a standard for monitors and the web called sRGB.1 It gave no value for ink. A screen makes colours from light and a press from inks on paper, so numbers that describe a colour on a screen do not say how to print it. The company’s label printer turned the screen colour into ink with the standard setting it used for coated paper, called FOGRA39,2 and its software warned nobody. To save two days, the printer had sent a proof on screen. It was Arjun’s own file, returned with its cutting marks, for checking the words and the layout. He looked at it on his calibrated monitor and signed it, and so did the director. The proof showed his rose, because it was his file. It could not show the printer’s. A screen set to imitate the press could have shown roughly how the rose would change in ink. This proof did not imitate the press. The labels were printed in early March, and the bottles reached the shops in the middle of the month.
The fault was his first, and he said so. He wrote to the director that the guide should have given the rose for ink as well as for screens, and that a printed proof would have shown the change. He paid for new labels for the flagship drink, which cost him about what the nine days had earned. The company let the first labels sell out and used the new ones from April. It spent the first weeks of its season with a label it had not chosen.
Arjun’s spring raises a question that everyone who works fast now faces. When everyone can work as fast as you, what is speed still for, and who decides where the work must slow down?
The print edition answered with a slogan. Speed was “the mechanism”, and “Consciousness is what decides when to engage the mechanism and when to disengage it.” It closed: “Speed is a weapon. Consciousness decides where to aim.” This edition cannot say that. Arjun was fully awake for all nine days, and the rose was still wrong. What failed was narrower and easier to name. He signed for something he had not looked at, on a schedule someone else had set, through a setting at the printer that nobody had chosen on purpose.
What this edition holds is plainer. Machines have made making fast, and fast for everyone, so speed is no longer an edge. What stays scarce is the moment when an option becomes an act that other people will rely on: a proof signed, a label printed, a lesson taught in a village. Chapter 1 described owning an act as taking a decision as yours, knowing what it is, with the power to refuse it. An option commits nobody. An act is someone’s, and someone must answer for it. The pace of that moment should be set by what the act can cost and who will bear it, not by the tool or the rival. Slowness is no virtue in itself. A pause is empty unless it holds the particular look the act needs. And slow work used to hold checks that nobody chose. They disappear with the delay unless someone names them and pays for them.
24.1 Why Speed Matters, and Why It Stops Being an Edge
Begin with what is true. Speed is a real good. A team that ships sooner meets the world sooner. It learns what people want, what breaks, and what nobody notices. Delay has costs, and some delay is fear dressed as care. Arjun’s client was not foolish to want nine days. A rival’s summer range was due in the shops in March, and a drinks company that misses the first hot weeks can lose a season’s place in the fridge.
But a speed that everyone has is not an edge. An earlier chapter followed Diane Uwamahoro, who built a software firm in Kigali on speed. Her team could connect an East African business to a payment system in three weeks. In a region where such projects often stalled for months, three weeks was a weapon. Then a competitor of three people, with AI coding tools, could do in four days what her team did in three weeks. Her speed had been real. It had also been measured against a slowness everyone else shared, and when that slowness went, so did the edge. When everyone executes fast, speed stops setting anyone apart. Being fast is the baseline. The professional who cannot tell the difference is not fast. She is merely in a hurry.
Being first was always less of an edge than the stories said. In 1993 two marketing scholars, Peter Golder and Gerard Tellis, studied about 500 brands in 50 kinds of product.3 They found that “almost half of market pioneers fail”, and that the “early market leaders”, who did much better over time, entered “an average of 13 years after pioneers”. Their study is of another era, and of markets, not of tools. It shows only that arriving first and lasting are different things.
The chapter’s title takes its word from war, so it is fair to argue with it in a commander’s own words. Augustus was the first Roman emperor. His biographer Suetonius, writing about a century later, reports that he “thought nothing less becoming in a well-trained leader than haste and rashness”.4 Among his favourite sayings was a Greek one, speûde bradéōs, which means “hurry slowly”. J. C. Rolfe’s translation gives it as an English proverb, “More haste, less speed”: hurrying makes you slower. Another was a line of Greek verse: “Better a safe commander than a bold.” A third was Latin: “That is done quickly enough which is done well enough.” And he warned against staking much to win little. Those who “grasped at slight gains with no slight risk” were like men who “fished with a golden hook”: if the hook was lost, the loss “could not be made good by any catch”.
Augustus was speaking of battles, not of shop shelves, and nothing here depends on his politics. But his maxims undo the easy reading of the title. For a commander, speed is a weapon only when it serves the judgement of when to strike. The worst haste stakes what cannot be replaced to win what can. Arjun’s nine days won a client’s delight. They staked the colour, the first weeks of a season, and a trust built on one rose label. A weapon that everyone holds wins nothing by itself. And its cuts often fall on people who never held it.
When everyone is fast, the clock also tightens for everyone. The client’s nine days were a reply to a rival’s schedule, and the rival’s schedule was a reply to someone else’s. Nobody in such a chain chooses the pace alone. No link wants the whole chain to be careless. Each link takes its pace from the next, and the checks are cut wherever the link is weakest, though no single link chose to drop them. The weakest link is often the one person working alone. He has nobody above him to ask for two more days.
24.2 Iteration Cycles and Rapid Experimentation
Kilifi County lies on the Kenyan coast. The town of Kilifi is about 56 kilometres north-east of Mombasa by road.5 In the county, Amara Ochieng leads a team of three that writes training for community health promoters. A promoter is a member of the community, selected and appointed under Kenya’s Primary Health Care Act of 2023. The Act came into force in November that year, and it is still in operation while a challenge to it is on appeal.6 The Act sets out the work. Among other duties, promoters give households advice on “maternal and postnatal care”. They “render first aid services” and, where necessary, “make referrals to the link facility”, the clinic a village is attached to. Amara’s team works for a small non-profit, alongside the county’s community health office. Its modules teach what the law asks of promoters: the signs of danger, first aid, and how to get a person to the clinic in time.
Early in 2025 a machine drafted her module on bleeding after a birth at home, overnight. It also translated the module into Swahili, and translated it well. Work that had taken her team a week took a morning of reading. One paragraph told the promoter to “prepare a clean, well-lit surface”. It had been written for a clinic. In the villages the floor may be packed earth, and the light may be a mobile phone torch held by a relative. The team crossed the paragraph out. In its place they wrote what a promoter in those homes can actually do. She can recognise how much bleeding is too much, call the transport and the clinic at once, and stay with the mother until help comes. The county’s community health office checked the new wording before any promoter was taught it.
In one way the speed was a gift. A module in Swahili now reached the promoters weeks sooner than it once had. For a mother bleeding after a birth at night, what the promoter beside her has learned, and how recently, can matter a great deal. Here speed served the people the work was for.
Before the machine, her team made a new module about every three months. They drafted it, tried it with one group of promoters, read three months of the promoters’ reports, and revised. From the middle of 2025 the machine let them draft and translate in days, so they began to change the bleeding module every month. Her funder’s report counted modules delivered, and each new version counted as one. The count rose from four a year towards twelve.
Nobody had asked for twelve. The funder’s form had a line for modules delivered and no line for what the modules changed. The non-profit filled in the line it had. Until the end of 2025 Amara did not think the line could mislead anyone, because every module on it was a good one. But she answers for what the promoters are taught, and the funder answers for what its form counts. Neither had asked what twelve changes a year would do to the promoters. The count had been a rough but fair guide only while each new module took three months of trials and reports. Once a module could be drafted overnight, the same count rewarded change for its own sake.
By the end of 2025 Amara saw what the count hid. In the villages her promoters covered, a mother bleeding after a birth at home was reported a few times in a season. A month of reports held one such case, or none. So each change came before anyone could know whether the last one had helped. Promoters trained in October and in November had learned two different versions of what to say on the phone to the clinic. And each new version cost the promoters an afternoon of training, taken from the households they would otherwise have visited.
The machine had made her side of the loop fast: drafting, translating, printing. It could not make the world’s side fast. Births come when they come. The reports that tell her whether a module works arrive at the pace of the events they report, and those events are rare. A learning cycle runs at the speed of its slowest observation. Changing a lesson faster than the world can reply to it only means guessing more often. A new version is not a lesson learned. The rate at which a team can make changes should not set the rate at which it exposes people to them.
From early 2026 Amara changed her rule. Her team still drafts as fast as the machine allows, and tries each draft on the promoters’ own questions in a single morning. That morning shows whether the promoters understand a draft. Only the reports can show whether it helps a mother. What the promoters are taught changes only when the reports give a reason. Even then, a field report says what happened, and the team has to ask why. A mother who came through safely may owe it to the training, the transport, a relative’s quick decision or the nurse on duty. Her next report to the funder will list fewer modules, and she has written a page explaining why. The page says that the count measured her team’s speed, not what the promoters had learned, and it asks for a line on the form for the reports themselves. It may cost her the funder’s goodwill. She sent it anyway. Whether the slower changes help mothers will not be known for a year or more. The reports come at their own pace.
24.3 The Fast Feedback Loop
Speed changes how often you hear back. It also changes whom you hear.
In the nine days, Arjun’s replies came fast. He sent directions at noon and had the director’s team’s notes by late afternoon. This was a strength: small, exact corrections every day, instead of one large correction after two weeks. The notes made the design better.
But notice who was writing them. The fastest replies come from the nearest reader: here, a team in an office, looking at screens. That team had its own launch date and its own manager. Its notes were honest, and they were about what could be seen on a screen by the end of a working day. The slowest replies come from the people the work is for, and from the world it must survive in. A shopper meets a label in a fridge, under tube lights, for perhaps two seconds. A press meets a colour on one paper with one set of inks. Neither writes notes by late afternoon. In a fast loop the near readers speak many times before the far ones speak once. So speed shortens the loop, and it also decides whose voices the loop is made of.
Arjun had a habit built for this very problem. In 2024 a hotel chain in Goa had rejected a website of his, because it was made for guests the hotel no longer had. Since the spring of 2025, before he shows any design, he writes the three strongest arguments against it in the voices of the people who will live with it. During the nine days he wrote them: a shopper in a hurry, a shopkeeper filling the fridge, the director’s own manager. None of the three voices was a press, a sheet of paper or a tube light. The habit had been built to imagine people. The rose failed in the world. Writing what a shopper might say keeps the shopper in mind. It is not the shopper’s reply.
Amara’s loop had the same shape. Her funder’s count came every quarter, and it was always good news. The mother who reaches the clinic in time, or does not, sends no report of her own. Silence from the far end tells her nothing. A count of modules says nothing about mothers.
So a fast loop needs a slow voice built in on purpose. Someone, or something, must speak for the far reader, at the far reader’s pace. For a label, that is a printed proof under the light where the bottle will be sold. For a health module, it is the reports from the field, read as they come and not overtaken by the next draft.
24.4 Compounding Knowledge: When Speed Builds on Itself
In the nine days, each round of making and choosing taught Arjun something about his own taste that the next round could use. A hundred such rounds in a year teach more than twenty-five. What you already know makes the next thing easier to take in, and the next thing becomes what you already know.
But what compounds is conditional. An earlier chapter put the conditions plainly. The world must reply. The next problem must be like the last. And the learner must not lie to himself about luck. Arjun’s nine days met the first condition for taste, because the director’s team replied every afternoon. They did not meet it for ink, because no round reached the ink. A hundred rounds in which the press never speaks teach nothing about printing. They teach confidence about printing, which is worse. Amara’s monthly drafts met none of the conditions for bleeding after birth, because the world had not replied between them.
Here the machine side often keeps more than a person does. The label printer’s software held the colour data for the standard its press worked to, fixed since 2006. It also held the measurements it had taken from its own press sheets, thousands of them, to keep the press within that standard. Given Arjun’s file, it could have shown on a screen roughly what the rose would become. Design software has offered such previews, called soft proofs, for years.7 It could also have warned that the colour was given for screens. Printers’ checks of incoming files, called preflight, have looked for that kind of problem since the 1990s.8 The printer’s measurements pile up across every job it has run, and they are kept. A designer’s memory of presses gathers across a few jobs, and fades. When the printer’s setting converted the colour in silence, the better record was in the room, and nobody had asked it to speak.
The print edition went on to claim that compounded knowledge is “durable, because it is embedded in the consciousness of the people who accumulated it. It cannot be copied.” That is withdrawn. Some of it can be copied, as the printer’s record shows. And what cannot be copied is not, for that reason, an advantage. What compounds in a person is a habit of reading the replies. It compounds only where the replies come.
Compounding has a second edge. It favours whoever is already inside. Arjun’s rounds began from the rose label he had made fourteen months earlier for the same client. A designer the company had never used would have had no rose to begin from, and no client willing to give her nine days. Speed lets an insider’s knowledge grow faster. It does nothing to let a newcomer in. The edge here is the insider’s place, and speed only makes it larger.
24.5 Speed Traps: When Moving Fast Means Moving Unconsciously
In this section’s title, “unconsciously” has an ordinary sense. It means unknowingly: without knowing what you have stopped checking. An earlier chapter read “consciously” the same way, as “knowingly”. Arjun was awake for every hour of the nine days. What he did not know was that a check had gone.
When everyone executes fast, speed stops setting anyone apart, and moving fast unknowingly becomes the trap. The trap has several forms.
The first is the last mile. The machine does the early stages with great competence: the making, the drafting, the assembly of something that looks finished. The last stage, checking the details against the world, needs the same attention it always did. But the finish of the output creates an illusion. The work feels done. The polish is so convincing that the remaining review feels ceremonial. It is not ceremonial. It is where the errors live. Not every error starts there. But it is the last place to catch the errors that the earlier stages missed. Arjun’s proof was his last mile. It looked final because it was his own file, finished and correct in every word.
The second form is harder to see, and it is the one that caught Arjun. Slow work holds checks that nobody chose. In the old eight weeks there was always a printed proof, because there was time to wait for one. A second look at the colour happened on another day, with rested eyes. The printer had a week in which to ring and ask whether the rose was meant to be so pale on screen. None of these checks was in anyone’s plan. They happened because the work was slow. When the tools removed the wait, they removed the checks that lived in it, and nobody decided to remove them. Not every delay held a check. Much of the old slowness was only waiting. The task is to find the few delays that held a look at the world, and keep those. One way to find them is to ask, of each wait the tools removed, what used to happen during it, and who did it.
So the trap is a delay taken away without anyone asking what the delay had been holding, not speed itself. That is still a decision, even when nobody makes it on purpose. The client set nine days. A screen proof was chosen to save two of them. The printer’s software was set to convert screen colours without a warning, and someone had chosen that setting. Arjun signed a proof that could not show what he was signing for. Each of these was small. Together they removed every check that would have caught the rose.
Responsibility follows control, so each party had a share of what went wrong, though not the same share. Arjun’s was the largest, because the guide was his work: he had given the colour for screens only, and signed a proof that could not show the ink. The printer had sent that proof to save two days, and converted the colour without a word. The client had set the clock and signed the same proof. What each paid did not follow the size of each share. Arjun paid because he said the fault was his. The printer’s contract said that files are printed as supplied, so it paid nothing, and nobody asked who had chosen its setting. The silent setting kept the queue moving, and the clause left the risk with whoever sent the file. A default like that can be a decision, made in advance, about who pays. The client paid with the first weeks of a label it had not chosen. Whoever sets a job’s clock should pay for the pause the job needs.
Arjun answered for the rose in this book’s full sense. He explained what had gone wrong, he put right what could be put right, and he paid. Some of it could not be put right. The first weeks of the season had gone, and the people who bought those bottles had bought a label nobody had chosen. The new labels stopped the harm from going on. They did not undo it.
The third form is stopping early. When making is fast, it is tempting to take the first good direction and commit, because committing looks cheap. Here the machine looks like the safer party. Asked to keep widening the search, it does not tire. But more output is not a wider search. A model can return the same few ideas in different forms, as the competent, similar directions of the second evening showed. The danger is the person who stops because the next round feels like more of the same, and calls that judgement, without first asking for directions unlike the last. Arjun took the grey on the second day and spent seven more days inside it. That may have been right. Whether he should have searched wider is a question the nine days left no time to ask.
The fourth is volume taken for value. A machine produces output at a pace that makes people feel productive. Twelve competent presentations in a day can feel like more work than one problem understood. The hours a tool saves are often filled again with more output. For most employees, as an earlier chapter said, those hours belong to the employer anyway. Volume measures the tool. Value is what the work does for the people it is for.
Plot twist
The feeling of speed can mislead the person who has it. In early 2025 METR, a research group that measures what AI systems can do, ran a trial with 16 experienced developers. They worked on open-source projects they knew well, with about five years on each. Each of 246 tasks was assigned at random to allow or forbid AI tools. The developers expected the tools to cut their time by 24 per cent. With the tools, they took 19 per cent longer. Afterwards they still believed the tools had saved them 20 per cent.8
The trial was small, and METR said what it did not show. It covered one kind of work, with the tools of early 2025, and METR called the result “a snapshot”. In February 2026 METR reported that a later round could not measure the effect reliably. Many developers would not take part if it meant working without AI. METR thought the tools probably helped more by then than in early 2025, but its new data gave “only very weak evidence” of how much. In May 2026 METR published a survey of 349 technical workers. Their median answer was that AI tools had made them three times faster, and METR warned that such answers “are not necessarily grounded in reality”.8
The trial’s lesson for this section is narrower than its headline. The people using the tools were poor judges of their own speed. When the finish of the work tells its maker that the work went fast, that feeling cannot decide which waits are safe to cut. Only a clock that someone reads can show what a tool has saved.
The last form is the deepest trap, and it is about other people. Speed squeezes hardest the time other people need in order to reply. A printer needs a day to print a proof, and someone needs time to look at it. A client’s staff need time to carry a sample to a shop. A promoter needs time between versions to learn the last one. A fast schedule treats these replies as delays, and cuts them first. Often they are the only checks that look at the world. The fast party records the time it saved. The time it took from others rarely appears on anyone’s report. And sometimes the squeeze is the point: a contract sent an hour before it must be signed, terms changed overnight, a vote called before anyone can read the papers. Then speed is a weapon in the title’s plain sense, aimed at other people’s chance to reply.
24.6 Storm and Still
In April 2026, after the reprint, Arjun began to call his two ways of working “storm” and “still”. The words are his, and they are worth keeping, with one change to where the line falls.
The storm is the making: many options, quickly, from the machine and from him. In the storm nothing is yet anyone’s act. An option commits nobody, and nobody has to answer for it. It can be thrown away at no cost but the hours spent on it. So the storm can be as fast as the tools allow, and the pairing of rose and grey came out of one. Choosing happens in the storm too. Arjun chose the grey there. The print edition was wrong to call judging during the storm “destructive”. What must not happen in the storm is committing.
The still begins where an option becomes an act that others will rely on. For Arjun that point was the proof. When he signed it, the rose stopped being an idea on his screen. It became an instruction to a press, and then thousands of labels on bottles in shops. Speed and stillness are one discipline: fast where the act can still be undone, slow where it cannot. Chapter 18 set the pace of an organisation’s decisions by three things: how much is at stake, how far a decision spreads, and how far it can be undone. Inside one person’s work the same rule draws a line through the middle of a job. Before the signature, undoing is cheap. After it, undoing costs a reprint, a season and some trust.
Not every act shows how far it can be undone. A label can be reprinted, at a price. A lesson cannot be untaught. Amara’s drafts were options, and she could change them every week without harm to anyone. A module taught to forty promoters was an act. When a new version replaced it, the old one did not leave their heads. At night, in a hurry, on the phone to the clinic, a promoter might say either. Her monthly changes had treated an act as if it were an option. That is why her rule now falls where it does: drafting stays in the storm, and teaching waits for the still. Waiting can also be a choice that cannot be undone. Had her team held the first module in Swahili back for the weeks the machine had saved, promoters would have gone those weeks without it. A mother might have paid for the wait. So the still is weighed against what the wait itself costs, and it should last no longer than its check needs.
At first Arjun misread his own lesson. In April he tried to build a pause between making and checking. He walked down the stairs, past the printing shop, round the block and back. He came back rested. But the rose would have been wrong after the walk too. No rested eye, looking at his own file, can see what a press will do to it. The walk was a pause, and a pause is not a check. Earlier in this book, U Nandaka, a monk who teaches meditation outside Yangon, admits the years he wasted mistaking a quiet mind for awareness. The book puts his warning plainly: stillness can be its own form of sleep. A still that holds nothing is that sleep.
The still has to hold the particular look the act needs. Some truths can be checked inside a text or a file. Others can be checked only in the world. The rose’s truth was in the world: in ink, on one paper, under one light. So for a label the still is a printed proof on the label paper. It is looked at in a viewing booth under the standard light the printing trade uses for comparing proofs.9 Then it is looked at again under the kind of light it will be sold under. Two colours that match under one light can look different under another.10 For Amara’s module the still is the field reports, read before the next version is taught. Each act has its own still, because each meets a different world. A still is worth what it finds out and what it keeps open, not the hours it takes. And someone who can stop the work must see what it shows.
Two old texts help to say what is wrong with haste, and they disagree about where the wrong lies.
In the sixth book of his Nicomachean Ethics, Aristotle writes about good deliberation, the excellence of thinking well about what to do. In W. D. Ross’s translation11 he says that it is not “skill in conjecture; for this both involves no reasoning and is something that is quick in its operation, while men deliberate a long time, and they say that one should carry out quickly the conclusions of one’s deliberation, but should deliberate slowly.” The last clause is a saying he reports (“they say”), not a rule he lays down, and he does not leave slowness there. He adds that “it is possible to attain it by long deliberation while another man attains it quickly”, and that “in the former case”, the slow man’s, “we have not yet got excellence in deliberation”. Good deliberation is rightness “in respect both of the end, the manner, and the time”. Slowness is no virtue. Being right about the time is. Nor does a good result prove good deliberation, for “it is possible to attain even good by a false syllogism”, by a faulty route.
Aristotle wrote of deliberation in general, not of print runs, so the application is mine. Arjun’s nine days fit that case closely. The result delighted the client. The route had skipped the ink. Psychologists call the habit of judging a decision by its result outcome bias,12 and, read this way, Aristotle’s sentence already warns against it. Machines have also changed the order he reports, slow thinking and quick doing. They have made part of deliberation itself fast: the search through possible answers, which Aristotle counted as inquiry. What has to take its proper time is the last part, where a person decides which option will become an act. His remark about the man who reaches the right decision quickly suggests that the proper time may be short.
The Dhammapada, the early Buddhist collection of verses, makes the other point. Its nineteenth chapter describes the just person, the dhammaṭṭha, the one who stands in the dhamma, what is right. Verse 256 begins na tena hoti dhammaṭṭho yenatthaṃ sahasā naye: one who settles a matter sahasā is not thereby just.13 The translators divide on that one word. F. Max Müller, in 1881, wrote: “A man is not just if he carries a matter by violence; no, he who distinguishes both right and wrong, who is learned and leads others, not by violence, but by law and equity, and who is guarded by the law and intelligent, he is called just.” Acharya Buddharakkhita, in 1985, wrote: “Not by passing arbitrary judgments does a man become just.” Thanissaro Bhikkhu wrote: “To pass judgment hurriedly doesn’t mean you’re a judge.”
Violence, arbitrariness, hurry. The translators are not careless. Each has taken one side of a small family of words that holds all three. The Pali Text Society’s dictionary gives sahasā as “forcibly, hastily, suddenly”, and also “inconsiderately”. Its related noun, sāhasa, is “violence, arbitrary action”.14 Read this way, to decide in haste is to decide by force. The verses are about a judge settling disputes between people, not about a designer or a trainer of health workers. I borrow only two things from them: the word, and the demand to examine both right and wrong before deciding something for other people.
So the two texts agree that slowness is no virtue. Thanissaro’s word for the next verse’s opposite of haste is “unhurriedly”. On my reading, what the Dhammapada sets against haste is weighing both what is so and what is not, and judging impartially. Where they differ is in where the wrong of haste lies. For Aristotle it lies in the one who deliberates: quickness skips the reasoning. For the Dhammapada it lies in what haste does to the people it lands on: it is force. Arjun’s spring needs both. Aristotle explains how a delighted client and a faulty route could live together. The Dhammapada explains why the route mattered to people who were never in the room: the shoppers, the client’s staff, the printer’s crew who ran the labels twice. I would add a point of my own. The time people are given to reply to a decision is part of its justice, and more than a delay in carrying it out. On this reading, the title’s word was in the Pali all along. Haste, seen from the side of those it lands on, is force.
A still costs something. From April 2026 Arjun’s quotes include a printed proof and two more days, and he no longer offers nine; a studio that offers seven may win the next job.
There is a strong case against all of this, and part of it may win. As of 2026, soft proofs on a calibrated screen are already close for many colours, and the software keeps improving. A printer’s system may soon show, with fair accuracy, how a label will look on its paper under a shop’s lights. If it does, much of the still’s looking will pass to a machine, and a designer who insists on paper will be slow for nothing. That should be welcomed where it is true. But that would move the still into the machine, and the still would remain. Someone must still decide to rely on the machine’s look, and for which jobs. Someone must still answer when the look is wrong: explain what happened, and pay. The Prologue put this as a commitment, not a fact about machines: a machine’s makers and users will never be free of what it does. The check can be delegated. The act stays someone’s.
The chapter’s question had two parts. The first asked what speed is still for. Not for winning, since everyone has it. It is for the storm: more options seen, and, when someone asks for it, a wider search than any person could make alone. It is for getting a good thing to people sooner, as the module in Swahili reached the promoters. And it is for meeting the world’s reply sooner, where the world replies quickly. Where the world replies slowly, speed is for nothing much, and it can do harm.
The second part asked who decides where the work must slow down. Neither the tool nor the rival decides. The pace of making can be the maker’s. The pace of the act belongs to those who will answer for it and those who will bear it. Here that was the designer who signs, and the company whose product and season it is. His part is to say what his signature needs before he gives it. Hers is to decide whether to pay for it, knowing what it may cost to go without. Being able to ask for time is not the same as having the power to get it. A freelancer who asks may lose the client, and an employee may not be able to ask at all.
Ending: The Grey Under the Tube Lights
In September 2026 the drinks company came back. It wanted a gift pack for the festive season in the same rose and grey, with the labels in the shops by the middle of October.
Arjun’s quote had two more days in it than it would have had in January, and a printed proof. The director agreed to the days. The company would not pay for the proof, so he paid for it himself.
He went to the printer to see it. In the viewing booth, under the standard light, the rose was the rose. Then the foreman carried the sheet to the packing tables and held it up under the tube lights there, the kind the shops have. The rose held. The grey did not quite. It leaned towards green.
He could not tell whether a shopper would notice, and nobody could without asking shoppers. There was no time to ask. He rang the director and told her what he saw.
She came to the printer that afternoon. She looked at the sheet in the booth, and then under the tube lights, for a long time. She asked the foreman what a new grey would cost and how many days it would take. It was her product and her season. She said she would tell them in the morning.
Arjun took the sheet back to Parel. She had not yet said.
Practice
Find the moment it leaves your hands. Choose one piece of work this week that other people will rely on. Mark the point where it stops being an option and becomes an act: sent, signed, printed, taught or switched on. Beside that point, write what could not be undone after it, and who would bear it if it went wrong. Then name one check that looks at the world rather than the file. It might be a printed proof under the light where the thing will be used, or a reply from someone who will live with it. Ask who set the deadline, and whether the check fits inside it. If it does not, tell whoever set it, before the work starts and in writing, and let them decide. If they will not pay for it and you work for yourself, you may decide to pay for it, although it should not be yours to pay. A small price can show a large risk in time to the person whose decision it is. Note whose hours the check costs. If you work for yourself, they are probably unpaid. If you are employed, they are your employer’s, and only if your employer agrees. This exercise promises no client. A faster rival may win the job.
Words to keep
- act and option: in this chapter, an option is something made and not yet committed to, which can be thrown away at little cost to anyone else; an act is an option someone has signed, sent or set going, which others will rely on.
- community health promoter: in Kenya, a member of the community appointed under the Primary Health Care Act 2023 to advise households, give first aid and refer people to a clinic.
- dhammaṭṭha: in Pali, “one who stands in the dhamma”; in the Dhammapada, the just judge.
- FOGRA39: a standard description of how colours print by offset on coated paper; printers use it to turn screen colours into ink.
- outcome bias: judging a decision by how it turned out rather than by what was known and how it was made.
- preflight: a printer’s check of an incoming file for problems, such as colours given for screens, before anything is printed.
- printed proof: a sample printed before the run so that colour can be judged in print rather than on a screen; for a test under other lights, it must be printed with the real inks on the real paper.
- sahasā: in Pali, “forcibly, hastily, suddenly”; the word the Dhammapada uses for deciding a matter in haste. The related noun, sāhasa, means violence or arbitrary action.
- soft proof: a preview on screen of how a file will look when printed on a particular press and paper.
- speûde bradéōs: Greek, “hurry slowly”, a favourite saying of Augustus.
- sRGB: a standard set of colours for monitors and the web.
- storm and still: Arjun’s names for making options quickly and for the moment, given the time its check needs, when an option becomes an act.
Sources
- Michael Stokes, Matthew Anderson, Srinivasan Chandrasekar and Ricardo Motta (Hewlett-Packard and Microsoft), “A Standard Default Color Space for the Internet - sRGB,” version 1.10, 5 November 1996 (w3.org, marked obsolete and kept for historical purposes): an RGB colour space “well suited to” monitors, television, scanners, digital cameras and printing systems, proposed as a default for the Internet; it has since been standardised as IEC 61966-2-1. The printing systems meant are those that accept RGB files; an offset press prints in CMYK inks, so an sRGB colour has to be converted.
- International Color Consortium, CMYK characterization data registry, “FOGRA39” (registry.color.org): Fogra characterisation data of December 2006 for “Offset commercial and specialty printing according to ISO 12647-2”, “paper type 1 or 2 (gloss or matte coated offset, 115 g/m2)”, ISO 12647-2:2004/Amd 1.
- Peter N. Golder and Gerard J. Tellis, “Pioneer Advantage: Marketing Logic or Marketing Legend?”, Journal of Marketing Research 30, no. 2 (May 1993): 158–170, abstract (“Approximately 500 brands in 50 product categories are analyzed”; “almost half of market pioneers fail”; “early market leaders have much greater long-term success and enter an average of 13 years after pioneers”).
- Suetonius, The Lives of the Caesars, “The Deified Augustus” 25.4, trans. J. C. Rolfe (Loeb Classical Library, 1913), via LacusCurtius (penelope.uchicago.edu); the Greek σπεῦδε βραδέως and the Latin “sat celeriter fieri quidquid fiat satis bene” from the Latin text on the same site. “Hurry slowly” is my rendering of the Greek.
- “Kilifi,” Wikipedia (accessed 23 October 2026): “a town on the coast of Kenya, 56 kilometres (35 mi) northeast of Mombasa by road”.
- Primary Health Care Act, 2023 (Kenya, No. 13 of 2023), assented to 19 October 2023, in force 2 November 2023: s. 2 (a community health promoter is “a member of the community who is selected and appointed in accordance with section 9”); s. 11(2)(f)(iii) (“maternal and postnatal care including advice on breastfeeding, immunization, child health care and family planning”); s. 11(2)(g) (“render first aid services to an assigned household and where necessary, make referrals to the link facility”) (Kenya Law, new.kenyalaw.org). The Act does not define “link facility”; the gloss in the text is mine. On 12 July 2024 the High Court declared the Act unconstitutional for want of public participation and suspended its order (Aura v Cabinet Secretary, Ministry of Health, Petition E473 of 2023, [2024] KEHC 8255); on 20 September 2024 the Court of Appeal stayed that judgment “pending the hearing and determination of the appeal” (Cabinet Secretary, Ministry of Health v Aura & 13 others, [2024] KECA 1195, Kenya Law). I found no decision on the appeal itself, so the Act is in operation as of October 2026.
- Corel, CorelDRAW Help, “Soft proofing”: “Soft proofing provides an on-screen preview of a document as it will appear when it is reproduced by a specific printer” (product.corel.com). Other design software offers the same function.
- “Preflight (printing),” Wikipedia (accessed 23 October 2026): checks that “image files are of the correct color format” and that “required color profiles are included”; the term “was first used in a presentation at the Color Connections conference in 1990”. For the Plot twist in 24.5 (added in this edition): Joel Becker, Nate Rush, Elizabeth Barnes and David Rein, “Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity,” arXiv:2507.09089 (METR, 10 July 2025; on arXiv from 12 July 2025), abstract (16 developers with about five years on their projects; 246 tasks, each “randomly assigned to allow or disallow usage of early 2025 AI tools”; a forecast of 24 per cent less time; “allowing AI actually increases completion time by 19%”; afterwards an estimate of 20 per cent less); METR’s summary of the same date calls the result “a snapshot of early-2025 AI capabilities in one relevant setting”. METR, “We are Changing our Developer Productivity Experiment Design,” 24 February 2026 (metr.org): the new data give “an unreliable signal”, partly because of developers “choosing not to participate in the study because they do not wish to work without AI”; METR believes “it is likely that developers are more sped up from AI tools now”, but its data are “only very weak evidence for the size of this increase”. METR, “Measuring the Self-Reported Impact of Early-2026 AI on Technical Worker Productivity,” 11 May 2026 (metr.org): 349 technical workers surveyed from February to April 2026; “The median self-reported speed change (which we expect to be higher than value change) is 3x”; “survey results are not necessarily grounded in reality”. As of November 2026 METR had published no results from its redesigned experiment, and its July 2025 post now warns: “These results are out of date.”
- ISO 3664, Graphic technology and photography: Viewing conditions (2009 edition; replaced in July 2025 by ISO 3664:2025), as described in “The Key to Accurate Color Viewing,” Printing Impressions, 12 October 2015 (piworld.com): “D50 light, which represents natural daylight, is used to view prints”, and comparisons across the supply chain should be made in conditions that comply with the 2009 standard.
- “Metamerism (color),” Wikipedia (accessed 23 October 2026): illuminant metameric failure, where “two material samples match when viewed under one light source but not another”.
- Aristotle, Nicomachean Ethics VI.9 (1142a31–1142b33), trans. W. D. Ross (Oxford: Clarendon Press, 1908), The Internet Classics Archive (classics.mit.edu).
- Jonathan Baron and John C. Hershey, “Outcome bias in decision evaluation,” Journal of Personality and Social Psychology 54, no. 4 (1988): 569–579.
- Dhammapada 256–257 (chapter XIX, Dhammaṭṭhavagga). Pali from Ānandajoti Bhikkhu’s edition, ancient-buddhist-texts.net (written there with ṁ, here with ṃ). Translations: F. Max Müller, The Dhammapada, Sacred Books of the East 10 (Oxford: Clarendon Press, 1881), chapter XIX, “The Just” (sacred-texts.com); Acharya Buddharakkhita, The Dhammapada: The Buddha’s Path of Wisdom (Kandy: Buddhist Publication Society, 1985), “The Just” (Access to Insight); Thanissaro Bhikkhu, Dhammapada: A Translation, “The Judge” (Access to Insight).
- T. W. Rhys Davids and William Stede, The Pali Text Society’s Pali-English Dictionary (Chipstead: Pali Text Society, 1921–1925), “Sahasā” (p. 701: “forcibly, hastily, suddenly”; “inconsiderately”), “Sāhasa” (p. 707: “violent, hasty”; “violence, arbitrary action”), “Asāhasa” (p. 88: “absence of violence, meekness, peaceableness”; as an adverb, “without violence, not arbitrarily”, citing the Jātaka, with the instrumental asāhasena, the form in Dh 257, given the same sense); via wisdomlib.org.
Named people in this chapter are composites, and their conversations and circumstances are constructed; the studies, laws and figures cited are real and sourced.
Revised from the print edition (2026)
- What the print edition claimed: that speed is “the mechanism” and “Consciousness is what decides when to engage the mechanism and when to disengage it”; that “Speed is a weapon. Consciousness decides where to aim.”; that “Speed serves learning. Learning requires consciousness. A cycle that lacks a conscious pause is not a learning cycle.”; that interpreting a metric needs context and “Context requires consciousness”, “the contextual awareness that this book has been calling consciousness since its first page”; that a machine “cannot tell him” why a client rejected a palette and had made “an assumption it cannot see”; that the exporter’s ear was “an instrument that no AI could replicate”; that “Financial returns compound automatically”; that compounded knowledge “is durable, because it is embedded in the consciousness of the people who accumulated it. It cannot be copied”, and needs “the one thing AI cannot provide: a self that learns”; that “Evaluation during the storm is not just unnecessary. It is destructive.”; that Amara’s modules had “the lowest complication rate of any community health programme in the coastal region”; that Arjun’s sharbat label was “the best-selling packaging in its category across three Indian cities”; that the exporter avoided the relationship trap “through identity”. It told an exporter’s phone call to his Hamburg buyer, “I needed to hear his voice”, as an override the forecast could not match, with the buyer named Klaus; it put Kilifi south of Mombasa, wove “cotton-jersey” (a knitted cloth) and measured it by thread count; it recalled a Ghanaian publisher’s review gates, a story that now belongs to Chapter 21, and “compound interest on conscious attention”, which no chapter contains; it gave Arjun a “deceleration protocol” of a walk, a call and a book; it ended with a roll-call of scenes that no chapter tells; and it opened at a screen at night, with a machine producing “a direction every forty seconds” and cold chai.
- What I now hold: machines have made making fast and fast for everyone, so speed is no edge (Diane; Golder and Tellis), and a commander’s own maxims argue with the title (Augustus: “That is done quickly enough which is done well enough”; the golden hook). What stays scarce is the moment an option becomes an act others rely on. Its pace is set by what the act can cost and who will bear it: fast where it can still be undone, slow where it cannot (Chapter 18’s rule, drawn here through one person’s work). A learning cycle runs at the speed of its slowest observation; fast replies come from the nearest reader. What compounds is conditional, and the machine side often keeps more than a person does (a printer’s fixed standard data and its own measurements). Slow work held checks that nobody chose, and taking away the delay takes them away unless someone names them; that removal is a decision, shared by whoever set the clock, whoever chose the proof, whoever set the printer’s software and whoever signed. “Unconsciously” in 24.5 means unknowingly. A pause is not a check; the still must hold the look the act needs, in the world, not in the file. Aristotle: slowness is no virtue, rightness includes the time, and a good result does not prove good deliberation. The Dhammapada: the word for haste is also the word for force, and its opposite is examining both right and wrong. The still costs hours, and someone must pay for them. The pace of the act belongs to those who answer for it and those who bear it. What each party paid did not follow the size of its share; a silent default can be a decision about who pays. Not every delay held a check. More output is not a wider search, and a new version is not a lesson learned; the rate of making should not set the rate at which people are exposed to change. A field report says what happened, not why; an imagined voice is not a heard one; and silence from the far end tells nothing. Sometimes the squeeze on others is the point, and then speed is a weapon aimed at their chance to reply. Waiting can also be a choice that cannot be undone, so the still lasts no longer than its check needs. Asking for time is not the power to get it.
- What changed my mind: the Prologue’s rule against defining people by what machines cannot yet do; Chapter 1’s account of owning an act; Chapter 18’s pace rule; the audit’s corrections (Kilifi, jersey, the Hamburg buyer’s name, the orphans, the invented sales and complication rate, the false claim about compounding); Kenya’s Primary Health Care Act, which sets what promoters do; Aristotle on deliberation; the Dhammapada’s sahasā and its translators’ split; the printing trade’s own checks; three outside readings of an earlier version of this chapter, in October 2026, which pointed out that the chapter measured Arjun’s share of the failure by what he paid, that the title’s word needed its plain sense (speed aimed at other people’s chance to reply), that more output is not a wider search and a version is not a lesson, and that the printer’s silent setting and its contract fit together.
- What survives: the nine days and the reprint, now dated and paid for; “The nine days had felt like flight. The six weeks that followed felt like gravity.”; “Being fast is the baseline.”; “The professional who cannot tell the difference is not fast. She is merely in a hurry.”; “The work feels done. The polish is so convincing that the remaining review feels ceremonial. It is not ceremonial. It is where the errors live.”; Amara, the crossed-out paragraph and the phone torch; the colour-profile error (sRGB for FOGRA39); premature convergence and volume for value; small, exact corrections every day instead of one large correction after two weeks; each round of making and choosing teaching the next, so that a hundred rounds a year teach more than twenty-five; the last trap, about other people, as the deepest; storm and still, with the line moved to the point of commitment.
- Section titles in the print edition: 24.1 was “Why Speed Matters More Than Ever”, and 24.4 was “Compounding Knowledge: Speed That Builds on Itself”. The other four keep their print titles.