AI-Proof: What Machines Can Never Be — Volume III: Transformation, Part VII: The Conscious Blueprint, Chapter 35: The Conscious Startup
Part VII asks what a person builds when machines can build. After working alone, this chapter takes on the first others: what a founder founds when a machine can build the product and copy the founder.
A designer took on three people and got twenty-four copies of himself, work a machine could do for less. What is left to found is a few people free to see differently, held to one promise, on money that shapes what they make.
In Brief For readers who are about to take on their first others, or money to pay them, and wonder what a company of several is for when a machine can build the product. Chapter 35 follows a Mumbai designer through his first project with other designers, whose first round of work turns out to be copies of his own. His client refuses the copies, a machine shows how cheaply a studio’s manner can be imitated, and he has to decide what the others are for, whose name goes on their work, and whose money to take. John Ruskin’s rule against asking workers for copies, and the last hymn of the Ṛgveda, a prayer for one mind and one resolve, help show what a company of several should share and what it must not be made to share.
A machine can build a founder’s product and copy his manner; what is left to found is a few people allowed to see differently and held to one stated promise, and the money they take decides much of what they become.
Opening: Twenty-Four Boards on the Wall
Twenty-four boards were pinned to the wall of the second room, and every one of them was good.
Arjun Mehta had put them up himself, in the last week of June 2026. He is a brand and packaging designer in Mumbai. His studio is two rooms above a printing shop in Parel, and he lives in Bandra. For most of his working life the studio had been one designer and his machines. The boards were a first round for a client with six yoga studios across the city. Three people had made them: eight by Neha, eight by Prashant and eight by Tanvi.
The type was clean and the colours were considered. Each board understood the client’s world well enough to please most clients. And each was, unmistakably, an imitation of Arjun. Nobody had traced his work. But the grammar was his: the muted earth tones he liked, the wide empty space, the serif type he reached for without thinking. Neha’s boards were the most accomplished and the most like his. Prashant’s showed the same habits with less skill. Tanvi’s were the roughest, and, in a way Arjun could not name, the most alive. They were twenty-four boards that looked like Arjun Mehta on a middling day.
The client was Devika. She had built her six studios over twelve years, and she judged design the way she judged a yoga teacher: by what it did to the room. She looked at the wall for a long time. Then she said, “These are lovely. But they’re all the same person. I hired a studio, not a photocopier.” She asked who had made which boards. Arjun told her, and she listened as if the names were news.
She was right, and he could see how it had happened. In May she had asked whether he could give her three designers for two months, and he had said yes almost at once. He had found people who admired his work. He had briefed them with his own references and judged what they made against his own standards. Without meaning to, he had designed a system to produce more of himself. It did.
The wall belongs to June, four months before the October in which Chapter 34 ends. Behind the boards were three people, each on different terms. Neha had worked for Arjun since May as a freelancer. She did the smaller jobs for a hotel chain in Goa, one of his oldest clients: menus, signs by the pool, offers for the monsoon. That spring he had told the chain’s operations director that others might do such work if he looked at it and stood behind it. From June he paid Neha a monthly fee against her monthly invoice. She was not an employee on paper, and she had a desk in the second room. Prashant and Tanvi were paid by the day and worked mostly from home. Tanvi was a recent graduate of the National Institute of Design in Ahmedabad, and this was among her first paid work.
A machine can now build much of a product, and, as this chapter will show, it can copy a studio’s manner. So the chapter asks: When a machine can build the product and copy the founder’s manner, what is there to found, and what decides what it becomes?
The print edition’s answer was that “a company needed multiple forms of consciousness. Not one consciousness multiplied.” It said that every stage of building a start-up “is a question of awareness, not execution.” The wall in Parel refutes it. Everyone who made and judged those boards was conscious. Consciousness did not stop them making copies, and it does not explain why they did. What made the copies was a brief, a tool set up with Arjun’s past work, and a habit of review. All three were Arjun’s choices.
This edition holds something narrower. Once machines can make the product and imitate its maker, what a founder founds is a “we”. It is a few people who share one stated promise and who are allowed to see differently in keeping it. The promise says what they offer, and to whom, and who will explain and put right what fails. The founder’s first acts are deciding what the others are for, whose names go on what, and whose money to take. None of these acts makes a company good. They make it a company, and not one person copied.
The chapter keeps its title, and the title needs an argument. Consciousness cannot tell a start-up from a copy shop, because everyone in both is conscious. The adjective survives here only in its ordinary sense, as knowingly. A start-up founded knowingly is one whose founder can say, before the build, what the others are for and what the money will ask of them. In May, Arjun could not. The other word in the title needs a plain sense too. Here a start-up is any new company at the moment it takes on its first others, whether or not it seeks investors or fast growth. Arjun’s two rooms are a small case of it. In the stricter sense that founders and investors use, a start-up searches for something it can repeat and grow, and a studio that sells its people’s time is an agency. This chapter’s subject comes before either: the moment one person first brings others into making and keeping a promise. A software founder hiring her first engineers faces the same decisions, as 35.1 shows. The print edition also laid out five “phases” of building a start-up. Their five subjects are kept below, but not as phases. They are five decisions that one founder made, in no fixed order, each with a price.
35.1 The Question Before the Build
A prototype is not a company. The sentence holds better now than when it was first written.
As of 2026, a working prototype of many kinds of software can be built in days. It will not yet be secure, reliable and joined to everything it must work with. In November 2025 Collins Dictionary chose “vibe coding” as its word of the year, and defined it as “the use of artificial intelligence prompted by natural language to assist with the writing of computer code”.1 In design the gap is smaller still. Since 2025 a machine has made competent images, layouts and type within a minute or two; 35.2 shows what that means for a studio’s manner. Making has become cheap.
Cheap making sets a trap for founders who mistake creation for founding. They build because building is now effortless, and the speed itself feels like proof. This book has already said what a fast copy of a product shows: that the part you can see is a thin layer, and that any real business lies elsewhere. It has said that a working demonstration shows only that a thing can be made, not that anyone wants it. This section asks a question that comes before both, and that cheap making has made new.
Creating and founding are different acts. To create is to make a thing, and a machine can now do much of that. To found is to bring into being something that other people will rely on, and that must keep its word when you are not in the room. It is a promise to customers, and an arrangement with the people who keep the promise. Making binds nobody. Founding binds other people, and not all of them chose it. That is why the scarce act, in 2026, is the question before the build. The build has become cheap, and the binding has not.
Arjun founded something in May without noticing that he had. Devika had seen a website he made for the Goa chain, and she asked him to her oldest studio, in Matunga. She wanted a new identity within two months, for her six studios and for three more she meant to open. “Can one person do that?” she asked. He said he had a studio. He had a studio of two rooms, one freelancer and a machine. By the end of the meeting he had promised her three designers whom, apart from Neha, he had not yet met. He had sold a “we” before there was one.
There is nothing wrong in that. Every founding promises more than the founder can yet do alone; that is what it is for. The fault was in what he did next. He found Prashant through friends and, a few days later, Tanvi through a design school’s notice board, and set them to work in June. What he did not do was say, to them or to himself, what the three would add that he and his machines would not.
For a founder who already works alone with machines, the question before the build takes a particular form. Before you take on a second person, what will the others be for? The question has a second half, and it belongs to them: what will the company be for them? It means what they will learn, what they may decide and refuse, what they will be paid, and what happens to their work when they leave.
There are two common answers, and they lead to different companies. The first is “more of me”: more hours of my eye, at more desks. In 2026 that answer has a cheaper supplier than people. The second is “people who may see what I do not, and who will answer for what they make”. That answer asks for a different company. The brief must leave room for another eye. The tools must not be set to the founder’s taste. The names on the work must include the names of those who did it. And the money must be able to wait for people who are slower than a copy. There is a third answer, which founders rarely give: that the others are not needed, and the job should be refused or made smaller. Arjun did not consider it.
He was answering a smaller question, which was whether he could deliver three designers’ work in two months. He could. Because he never asked what the three were for, the arrangement answered for him, and its answer was “more of me”.
The same two answers face a founder who writes software. If her first engineers are hired to write the code she would have written, a coding model is, as of 2026, the cheaper supplier. If they are hired to see what she does not, in the customers’ work or in the ways the code fails, she must let them argue with her plan. She must also put their names on what they build.
The traps differ by kind of company. A studio sells its people’s time and judgement, client by client. Its temptation is to sell copies of one look as if each were made for the client. A product company is built to repeat, and repetition is its point. Its temptation is to treat what the founder built in a weekend as finished, so that the first engineers fear to say that customers hate its core. And code made largely by prompting can be code that nobody in the company understands from inside. When a customer’s case breaks it, someone must read it as if for the first time. Making is cheap. Understanding what was made, well enough to keep a promise when it breaks, is not.
Building with a machine feels fast, and that feeling is weak evidence. In the trial told in Chapter 24, experienced programmers using the AI tools of early 2025 believed they were faster and were slower. METR, which ran the trial, now labels that result out of date, and later tools may well help more. But in a survey METR published in May 2026, technical workers said the tools made them three times faster, and METR warned that such answers “are not necessarily grounded in reality”. The feeling is no evidence at all that anyone understands what was built.
What makes a few people one company, and not one person with helpers? Not that they work in one room, or for one client, or under one name. Arjun’s three had all of that in June. It is that they can say “we promised” and mean the same thing by it. A taste is not a promise. It is a habit, and a habit can be copied but not kept. A promise has to be said in words that the others can keep when the founder is not looking, and that they can be seen to break. This book has argued that work which lives in one person’s hands becomes a business only by becoming a different promise. The point here is narrower. Somebody has to put that promise into words, and the others have to hear it before they start.
One sentence from the print edition is half right: “A thin wrapper around a founder’s personality dies when the founder burns out, loses interest, or produces a team that can mimic but not think.” A thin wrapper, in this book, is a product whose main value is access to something the underlying model already does. A team that only copies its founder is a wrapper of that kind, with the founder as the model underneath. But the sentence blames the team. Neha, Prashant and Tanvi could think. They had been asked to copy, and they did what they were asked.
The questions before a build must also be asked again, because the answers on day sixty are rarely the answers from day one. That holds for this question most of all. By day sixty the others are no longer extra hands. They are people with their own reasons. The founder’s answer about what they are for has to change with them, or it becomes a cage.
35.2 Copies of the Founder
This section is about what a company of several builds that its founder alone could not. The claim that some things cannot be copied does not survive the first fact in this section.
Devika’s first word mattered as much as her last. She called the boards “lovely”, and she meant it. Nothing on the wall was badly made. Quality was not the trouble, and more quality would not have cured it. She had come to a studio to get more than one person’s way of seeing a room, and she had been given one person’s way, made three times. Sameness can be well made. It is still sameness.
The copying, in Arjun’s case, could also have been done by a machine. In March 2025 one company gave its chatbot a new way of making pictures, and people used it to make pictures in the manner of a famous Japanese animation studio. Within about nine days, the company said, more than 130 million users had made over 700 million images with it.2 Critics objected that it was copying the studio’s art. The company replied that it would go on preventing pictures “in the style of individual living artists”, and that it would “permit broader studio styles”.3 Read that from Arjun’s wall. A studio’s house style was exactly what the machine was allowed to copy, and as of 2025 it could do so within a minute or two. What a maker’s policy allows is not what the law allows, or what clients think fair; but it shows what the machine would do. Twenty-four boards in Arjun’s manner, made by three paid people over three weeks, were the dearest way to buy what a machine would make. The designers had a client, a brief and rooms to answer to; a prompt has none of these. But what Devika saw on the wall, a machine could have given her. If more of Arjun was what Devika wanted, she did not need a studio.
The English critic John Ruskin (1819 to 1900) saw this danger long before a machine could do the copying. In “The Nature of Gothic”, a chapter of The Stones of Venice (1853), he set the carvers of the old cathedrals against the workers of his own factory age. He wrote: “You must either make a tool of the creature, or a man of him. You cannot make both. Men were not intended to work with the accuracy of tools, to be precise and perfect in all their actions.”4 If you want a man, he went on, let him “begin to imagine, to think, to try to do anything worth doing; and the engine-turned precision is lost at once. Out come all his roughness, all his dulness, all his incapability; shame upon shame, failure upon failure, pause after pause: but out comes the whole majesty of him also”. Later in the chapter he gave three rules for anyone who buys made things. The third reads: “Never encourage imitation or copying of any kind, except for the sake of preserving records of great works.”5 Ruskin wrote about stone carvers and factory hands, not design studios. What I borrow from him is his rule against asking people for copies, and one distinction below. Devika, as I read her, said the same thing in one sentence.
Ruskin also sorted ornament by how freely its makers worked. In what he called “servile ornament”, “the execution or power of the inferior workman is entirely subjected to the intellect of the higher”. In a better kind the worker’s power has “a will of its own, yet confessing its inferiority and rendering obedience to higher powers”.6 Carried into a studio, which is my use and not his, the first kind was the wall in Parel. The second is closer to what Arjun needed, but its “confessing its inferiority” is refused here. Tanvi owed the work her best seeing. She owed nobody a confession.
Ruskin is not a clean witness for a studio in Mumbai. In a lecture in London in January 1858, months after the rising of 1857 began, he said that Indian art “never represents a natural fact” and “will not draw a man, but an eight-armed monster”. He argued that such art showed minds cut off from nature, and he tied it to what he called the “bestial, and lower than bestial degradation” of the acts of “the Indian race” in 1857.7 That judgement was ignorant, and it is refused here. His rule against demanding copies is kept. A thinker can see one kind of domination clearly and take part in another. A rule about freedom has to be tested by whom it leaves out.
An older text points the other way. The last hymn of the Ṛgveda, the oldest collection of Indian sacred verse, is a prayer for concord among people gathered at a rite. In Ralph Griffith’s translation (1896) it says: “Assemble, speak together: let your minds be all of one accord”. And: “One and the same be your resolve, and be your minds of one accord. United be the thoughts of all that all may happily agree.”8 The word Griffith gives as “resolve” is ākūti: an intention, what a person sets out to do. The scholar Joshua Katz has pointed out that the hymn plays on one run of sounds, man or mān, heard both in words for the mind and thinking (manas) and in the word for “common” or “the same” (samāna).9 In its own music, mind and sameness echo each other. The hymn is a prayer for a gathering, not advice for a company. What I borrow is what it says a gathering can share: a resolve, a purpose, a mind, and someone who lays the purpose before it.
The two texts seem to disagree, and partly they answer different questions. The hymn is about what people decide together. Ruskin is about how each worker’s hand works. But, on my reading, they meet on Arjun’s wall. His first round had one mind, his own, and the twenty-four boards all happily agreed. What it lacked was a resolve that anyone had said aloud. Nobody had told the three what the studio promised Devika that Arjun alone could not. The hymn also has a speaker, which is easy to miss. “A common purpose do I lay before you,” says its third verse in Griffith’s translation.8 Someone sets the purpose before the gathering, in words, and the gathering may then share it. On Arjun’s wall nobody had done that. The boards agreed because they shared his habits, not because anyone had set a purpose before them. Carried into a company, which is my reading, that someone is usually the founder, or the founders together when there are several. Whoever lays the purpose before the others answers for it, including for what it asks of them. The hymn does not say how a purpose is chosen, whether the gathering may dispute it, or how its speaker is held to account. Those questions are the company’s, not the hymn’s.
Read together in my way, the two give the narrower answer this chapter needs. A company of several needs one resolve, said aloud and answerable, and room for different eyes in keeping it. Its members should agree about what they promise. They should not be made to agree about what they see. The two texts still pull in different directions. The hymn asks whether people can act together; Ruskin asks whether each may use his own powers. A company that only coordinates becomes a machine for repeating its mistakes, and one that only frees cannot keep a promise.
Being allowed to see differently is not yet being allowed to act on it. A designer may disagree with the founder and still have no authority to change the work, the brief or the deadline. A company that allows difference only as suggestion has not shared its judgement. It has collected suggestions for one person to approve. Sharing judgement need not mean a vote on everything. It can mean a few decisions that are theirs, named in advance, which the founder does not overrule. And a common resolve can become another name for obedience, if disagreement is treated as disloyalty. The others must be able to ask whether the promise is being kept, whether the means are fair, and whether the promise itself should change. A gathering that cannot question its purpose may stay united, and united in its mistake.
Ruskin’s rule is wider than I can follow. A junior often learns by copying a master before she finds her own hand. And some clients want a house style, and are right to want it. What I keep of the rule is narrower. Do not ask people for copies as the work itself, least of all when a client has asked for more than one way of seeing. Devika had asked for that, and Arjun’s brief, tool and review had asked for one.
Arjun’s own check had missed the sameness. In 2024 the Goa chain rejected the first website he made for it. From his second design for the chain, in the spring of 2025, he wrote the case against his own work before showing it. He wrote it in the voices of the people who would live with the work. In June he had written those voices for the strongest boards on the wall, one board at a time, and each board had passed. The check was built for one designer’s work. It asked whether a design would serve the room. It never asked whether twenty-four designs were one person’s. A check made for working alone does not see what goes wrong when the work is shared.
For the second round Arjun changed what he could change. He took his references out of the brief. He gave each designer the brief, a summary of what Devika’s members said about her studios, and the addresses of all six, and sent them out separately. There was no shared mood board and no palette of his.
The summary came from a machine. Devika’s chain had more than a thousand reviews online, and Arjun had asked a model to read them. One complaint came up more often than any other: the rooms felt cold, “like a clinic”. The machine found it in a minute, before any designer had walked through a door, and it was right. It named the problem. It did not say what the rooms should become.
Three weeks later the wall held three directions, and they were not alike. Tanvi had noticed what the rooms already held: terracotta pots by the windows, and raw cotton on the bolsters. Her boards were warmer and rougher than anything Arjun would have chosen. Prashant’s were interesting in parts and incoherent as a whole. Neha’s were exact and flat. Without his references to lean on, she had fallen back on a competent blankness.
That last fact about Neha belongs to Arjun before it belongs to her. Since May he had paid her to match his eye on the hotel chain’s menus. In June he had handed her his references and judged her boards against them. She had learned what the studio rewarded. Not all of it was his. Some of her flatness was caution in front of the person who paid her each month. And eyes do different work. Some see the room first, and some make a chosen direction exact, as Neha would in August. A founder who demands that every hire be an original makes the opposite mistake to one who demands copies. This book has borrowed, once before, the Chinese story of the man who pulled up his corn to help it grow, and whose son found it withered.10 Arjun’s mistake was different. He had sown only his own seed, and then praised the field for looking like him.
So one designer out of three had produced something that justified a studio of several. One. And the one who did it was the one Arjun had hired most recently and influenced least. That is true of this round. It is not a law. One round, one brief and one client’s eye are not a test. The next brief might need Neha’s exactness more than Tanvi’s warmth. What the round showed was smaller. A studio that wants other eyes must make room for them, and the room has a price.
The price was paid at once. The second round took three weeks that Devika’s budget had not foreseen. Arjun paid Prashant and Tanvi their day rates for those weeks from his own savings, and Neha’s fee ran on. He also had to stand behind a direction that was not his, and that he would not have chosen. Forming another person’s eye, as this book has said before, is slow and someone pays for it. Here it was Arjun. Each such round cost about three weeks of two day rates and of Neha’s fee, and his savings could bear only a few. Freedom without support can also be abandonment. A recent graduate given room to see needs feedback, someone to stand between her and a client’s first no, and paid hours to learn in. Without them, talk of growth only dresses up cheap labour. And Arjun had his own learning to do: to tell quality from his own preference, to give feedback that did not ask for himself, and to hear disagreement as information.
35.3 Where Intelligence Lives
There was a second copier in the studio, and Arjun had built that one too.
The studio’s image generator could be given examples to draw its style from. In June he had loaded it with a library of his own past projects. Whatever the designers asked it for came back already leaning towards his look. Part of the sameness on the wall was in the tool. He had set it up, and nobody had asked him why.
For the second round he emptied the library. The boards changed, but less than he had hoped. The generator still favoured familiar conventions. Its training, its settings and the way it was asked all play a part, and nobody in Parel could say which did most. This book has already separated two problems that look alike: work shaped by one creator, and many people’s work growing alike because they use the same tools. In June the studio had the first problem. In July it still had the second. Sameness also has causes no tool explains: a founder who praises the familiar, and clients who reward what they have seen before.
So where did the studio’s intelligence live? Some of it was rented. The model belonged to a company in another country, and its next version could change the studio’s work without asking anyone in Parel. Some of it lived in the settings, and the settings were Arjun’s. Some lived in each designer’s walk through Devika’s rooms. And some lived in Devika. She had supplied the problem, the rooms and the members, and she judged a design by standing in a room and noticing what it did there.
The print edition said that the studio’s value “lived in the filter”. The machine generated hundreds of options, and people chose. Without that human judgement, it said, the studio would be “a random number generator with a design degree”. It is a good joke and a dated claim. As of 2026 a model asked to choose among boards for a brief will choose, and it will give its reasons. In the autumn of 2025 a client of Arjun’s tested his design for a lending app against designs a machine had made. One of the machine’s designs won on every measure the client counted. Choosing is not what is left over for people.
Nor, quite, is noticing. Tanvi saw the terracotta pots and the cotton because she stood in the rooms. As of 2026, had Arjun given a model photographs of the six studios, it might well have proposed earth and cotton too, since both were there to be seen. The rooms, not Tanvi, held the colours. What mattered was not that only she could see the colours. It was that the studio could ask her to answer for the direction, and give Devika someone to question. Arjun, who held the contract, stood behind the whole. Tanvi argued for the colours with Devika across a table, and she put her own work behind them. And whether her direction did anything to the rooms is not yet known. Devika’s members will find out over months, in the rooms themselves. Choosing a direction well and its working are two different things. A studio that never goes back to the rooms will mistake approval for success.
What the studio added was not the filter alone. It was that someone could be asked why this board and not that one, and that someone would put it right when it failed. A default in a tool is someone’s decision. The library of references was Arjun’s decision. It turned three designers into copies of him, so explaining that and putting it right were his to do. He did both in the second round, at his own cost.
The tools will change, and so will what they can do. What a founder sets into them will stay his: the examples they learn from, the defaults the others work inside, and which of the machine’s suggestions count as the studio’s. A studio of several can be one person’s taste carried silently by its software. Arjun had not seen this until the wall showed it to him. Setting the tools is a way of governing the people who use them.
35.4 Being Chosen
Devika chose Tanvi’s direction. Over two more rounds in August, Neha and Prashant worked on it with her. Neha made the type and the signs exact. Prashant drew a set of small marks for the class timetables, the best thing he made all summer. The identity went to Devika at the end of August.
In September she told two people about it: a woman who runs a group of ayurvedic clinics, and the owner of a small chain of cafés. Both telephoned Arjun. Both asked for “what you did for Devika”. Both meant the terracotta.
Who is chosen when a studio is chosen? The two callers had chosen a name, and the name was Arjun’s. The work they wanted was Tanvi’s, made better by Neha and Prashant, under his direction. This book has said that a recommendation carries the experience of the person who makes it, and also the limits of what she has seen. Devika’s carried a room she liked and a name she knew. Neither caller knew that there had been a Tanvi.
A name on a studio’s door promises that someone will stand behind the work, whoever did it. Earlier in 2026 the operations director of the Goa chain had asked Arjun whose name would go beside hers once others did part of her work. That question has a second half inside the studio. What does a company owe the person whose work its name carries?
It owes her at least two things. The first is her name, given to the client and not kept back for the studio’s convenience. An employee, this book has said, can at least keep her name on her own work. A freelancer paid by the day can lose even that, unless someone gives it to her. In September Arjun wrote to Devika that the direction had been Tanvi’s, and he added Tanvi’s name to the credits of the brand book. It cost him something. A client who came for his name now knows that another hand made the thing she liked. A name is a beginning, not the whole of what is owed. Neha made the type and signs exact, and Prashant drew the marks, and a direction’s name must not swallow the work that followed. Chapter 20 kept three senses of “yours” apart: what the law lets you stop others doing, where a work came from, and who answers for it. Credit should describe what each person did. Pay, ownership and the right to show the work in a portfolio are separate questions, and none stands in for another.
Ruskin’s rule, as I read it here, protects a worker’s freedom to see. It says nothing about the worker’s name. In a studio that sells everything under one name, the name is the first thing a freelancer loses, and the last thing anyone notices is missing. Nobody had asked Arjun to add Tanvi’s, and nobody would have noticed if he had not.
The second is a say in what happens to her idea. The law of copyright will not give her that say. India’s Supreme Court held in 1978 that there is no copyright in an idea, only in the form, manner and arrangement in which it is expressed.3 A palette of terracotta and raw cotton is, as I read it, nearer an idea than a drawing. So whatever say Tanvi has here, the studio must give her. A written agreement is the place to give it: who may reuse a direction, and whether its originator is asked or paid. Both callers wanted Tanvi’s terracotta for their own rooms. A clinic is not a yoga studio, and a café is not a clinic. But the easiest way to serve them was to copy the look Devika had paid for and change the name on it. That would make the studio a copier again, with a new original. Being chosen for a look pulls a studio back towards sameness, because a look is what strangers can see and ask for. It becomes a loop: a success is asked for again, repeated, recognised and asked for again.
The machines had already shown how cheap a look is. In September a rival chain opened a studio in Bandra with terracotta walls and cotton blinds. Its designer, Devika heard, had worked mostly with a generator. A look can be copied in weeks by anyone with the same tools. What the callers could not get from a generator was someone who would stand in their rooms and stand behind what was made for them. Whether Devika could stop the rival is a different question, and not one of copyright. Indian courts have let a business stop a rival who dresses its goods to pass them off as the first business’s.3 A combination of colours can also be registered in India as a trade mark, if it tells one business’s goods or services from another’s.3 A general earthy look rarely qualifies for either.
So at the end of September Arjun told both callers the same thing. He would visit their rooms before quoting, and he would not reuse Devika’s look. The clinic owner sent him a brief and the addresses of her four clinics. The owner of the cafés said, kindly, that she wanted what Devika had, and that she would find someone who could make it. She was not wrong to want it. She had seen a look she liked, and a look is a fair thing to buy. A studio may honestly sell a repeatable look, as a template or a licensed system, if it says that is what it sells. What is refused here is selling a copy as if it had been made for the room, and a look that one person found, without asking her. The studio lost a client by refusing to sell her a copy, and nothing in this chapter says that the refusal will pay.
The print edition ended its story of Arjun’s distribution happily. A studio “whose work made people stop and look was its own distribution channel”, it said. That is a happy ending told too early. Being chosen is someone else’s act. Being chosen for the wrong thing is a trap. The studio was walking towards it, pushed by its own success.
35.5 Capital as Alignment
Not every start-up needs external capital. Some are better without it. Whether a founder can raise money matters less than what the money will expect. Strictly, a client’s retainer is revenue, not capital, and a founder’s savings are his own capital, not outside money. This section uses “capital” loosely, for any money that comes with expectations about what the company will become: a fund’s, a bank’s, a large client’s retainer, or the founder’s own savings. Alignment means that what the money expects fits what the company has promised.
The print edition added that in 2026 “a founder with a working product and early traction can almost certainly raise something.” The figures do not bear it out even where such money is most plentiful. In the second quarter of 2026, American start-ups on one large platform for managing shares raised about as many dollars at the earliest stage as a year before. But they did it in just over 11,500 deals instead of 14,825, and in the first half of the year companies in AI took about half of all such money.11 On that one platform, a window on American start-ups and not the whole market, fewer deals were being signed, and the money went where it was already going.
Whose money you take decides what you build. That claim stands, and it needs one addition: no money is neutral, though some carries far fewer conditions than others. Read my way, every kind brings its own ākūti, its own resolve, into the gathering, and asks the gathering to share it.
The kinds act through different rights. A fund buys part of the company and often a voice in it, and it usually brings a clock, though funds differ. The one that backed the first company of Chidi Okonkwo, a founder in Lagos, asked him every month for growth. A bank lends money that must be repaid in a bad month as in a good one. A client’s retainer buys a claim on the studio’s time, and brings her calendar and her taste. A founder’s own savings bring the size of the mistake he can afford, which is usually small, and he bears the loss alone. And money that wants a company to grow fast tends to pay for what can be repeated. A method, a template or a house style can be learned by juniors in a week and sold in many rooms. That last point is my reading, not a law of finance. If it is right, the money that would grow Arjun’s studio fastest would pay for the very copier Devika refused.
Using one’s own money is no escape. The print edition said that “the discipline of self-funding protects the founder’s judgement from the distortions that misaligned capital introduces.” A founder’s own money has its own pressures. It makes him take every job in a thin month. It makes the hours spent forming someone look like waste. Arjun’s savings had paid for the second round. They could not pay for many more.
At the end of September Devika offered money of her own. She was opening three more studios in the coming year, and she wanted Arjun’s studio on retainer. She would pay a fixed sum every month for a year, enough to cover Neha’s fee and Tanvi’s. In return she wanted first call on the studio’s time, and no other client in wellness. It was a generous offer, and it would decide what the studio built. It would steady the studio’s cash and tie it to one client: more stable, and less free. “First call” sounds small. In practice it meant that when Devika rang, other clients waited, the Goa chain among them for as long as its contract ran. So did anything Neha was halfway through. For a year most of the studio’s hours would go to one chain, and most of those to variations of one look. The clinic owner would have to be turned away, since a clinic is wellness too.
Arjun was not the only one the money would bind. Neha’s fee and Tanvi’s chance of more work would rest on it, and neither of them had chosen it. A commitment that binds other people should be made in the open, and those it binds should be told why. So in the first week of October, before he replied to Devika, Arjun told them both. He said what the money would pay for, what it would ask, and what it would rule out. Telling them was not asking their permission. The decision stayed his, and so did answering for it. What telling gave them was time to decide what they would do. Telling is the least a founder owes, not the most. A decision that changes people’s work and income should leave them time to respond and a chance to negotiate what it means for them. It should also leave them free to say no without being punished. Keeping the decision did not give him the right to make them carry all its costs. And the money would bind each of them differently. Neha would have steadier income without a salary’s protections, and Tanvi more work on one look. Prashant, whose day rate had ended in August and who had not been asked back, was not in it at all.
Neha asked whether her monthly fee would become a salary. He said not yet. That answer had a cost, and Neha bore it. A freelancer on a monthly fee has the regularity of a job without its protections. She had no notice period, no paid leave and no share in the studio, and since June she had carried part of its risk. Keeping her freelance kept Arjun’s costs flexible. It also kept part of the studio’s uncertainty on her.
A hymn of concord in the Atharvaveda (3.30), another early collection of Vedic verse, prays for the same concord, and ties it to a common drinking-place and a shared portion of food.8 Read my way, a shared resolve has usually needed something shared in return. A founder who asks for one resolve while keeping every share, and the others on fees and day rates, is asking for concord on a contractor’s terms. The return need not be a share of the company. It can be the protection of a wage, or a part of what the company wins. Arjun had given Neha neither yet.
Tanvi asked one question: would she be making the same thing for a year? He said yes, mostly.
He signed for six months, not a year, and without the clause that ruled out other clients in wellness. Devika agreed, at a lower monthly sum. Before he signed, he wrote one sentence into the schedule of work. It was the studio’s resolve, said aloud for the first time: each new studio would be walked by the designer who drew for it, before anything was drawn. Devika accepted it. It would cost days that the lower sum did not pay for. The price was plain. The smaller sum covered Neha’s fee and part of a second. The months after March were unknown. If the clinic did not become paying work, the rest would come from his savings again. And he had kept the right to work for the clinic owner, which meant he now had to find a way to serve her that was not a copy.
As I read them, the hymn and Ruskin pull together here. Money arrives with a resolve already formed, and asks the gathering to adopt it as its own: one and the same be your resolve. Ruskin’s rule names something that no resolve should ask of the people who carry it out, which is copies. Neither text tells a founder which money to take. This book’s answer is narrower. Before you take money, say whose resolve it carries, and tell the people it will bind. Then you are at least founding knowingly, which is all this chapter’s title can honestly mean.
Ending: A Brief With Nothing Beside It
In the second week of October the clinic owner’s brief was pinned to the wall of the second room, with nothing beside it yet.
Arjun had asked Tanvi to come back for the six months, on a monthly fee like Neha’s. A larger agency in Bengaluru had offered her a job as well, at a better wage, on a team of twelve, with her work under the agency’s name. She had told him so herself.
He had told her what his offer was. Most of her months would go to Devika’s three new studios, which meant her own terracotta, made again for new rooms. Her name would be in the brand book. If the clinic owner signed, some of her time would go to rooms that nobody in the studio had yet seen.
She said she would tell him by the end of the week. Neha, at her desk in the second room, did not ask what Tanvi had said. Arjun did not know which answer he wanted. If she came, he would be paying her, partly from his savings, to make copies of her own idea. If she went, the next round on the wall would be Neha’s and his.
Tanvi has not yet said.
Practice
What the second person is for. Before you take on a second person, or money to pay one, write one page. First, write the sentence your company would promise a stranger that you, working alone with machines, could not. Second, for each person you would bring in, write one thing they may decide without you and one thing they may challenge. Write what they will learn, how they will be paid, and where their name will appear. Third, write whose money will pay for the first year, what it will ask of you each month, and who else it binds who did not choose it. Fourth, write how you will ask them before a decision that changes their work, not after it. Then show the page to the person you would take on, before they say yes. It costs an evening. If you work inside an employer, the money and the names are often someone else’s decision; write down whose. The page will not make the company work, and it promises no clients, no funding and no good hire. It may show you, before anyone else pays for it, whether you are founding a company or copying yourself.
Words to keep
- agency (design): a firm that sells its people’s time and judgement client by client; distinct from a start-up in the investors’ sense, which seeks something it can repeat and grow.
- ākūti: Sanskrit, an intention or resolve, what a person sets out to do; in Ṛgveda 10.191 the hymn asks that it be “one and the same” for everyone gathered.
- brand book: the document that sets out a brand’s type, colours, marks and rules, so that other people can use them.
- day rate: a freelancer’s fee for each day worked, with no promise of further days.
- house style: the look a studio’s work shares, whoever made it.
- passing off: in Indian law, as in English law, a claim against a trader who presents goods or services so that buyers take them for another’s.
- prototype: a first working version, made to show that a thing can be made.
- retainer: a fixed sum paid regularly to keep someone’s time available.
- servile ornament (Ruskin): work in which the maker’s hand is wholly subject to another person’s design.
- thin wrapper: in this book, a product whose main value is access to something the underlying model already does (Chapter 6).
- vibe coding: writing software by prompting a model in plain language; Collins Dictionary’s word of the year for 2025.
Sources
- Lianne Kolirin, “Vibe coding named Collins Dictionary’s word of the year,” CNN, 6 November 2025: Collins defines it as “the use of artificial intelligence prompted by natural language to assist with the writing of computer code.” On builders’ sense of speed (35.1; the trial is told in 24.5): 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 (v1 12 July 2025, v2 25 July 2025), abstract: “16 developers with moderate AI experience complete 246 tasks in mature projects”; they forecast that allowing AI would “reduce completion time by 24%”, afterwards estimated that it “reduced completion time by 20%”, and “allowing AI actually increases completion time by 19%”. METR’s summary (metr.org, 10 July 2025): repositories “averaging 22k+ stars and 1M+ lines of code”, contributed to “for multiple years”; each issue randomly assigned to allow or disallow AI; “a snapshot of early-2025 AI capabilities in one relevant setting”; “We do not claim that our developers or repositories represent a majority or plurality of software development work.” METR, “We are Changing our Developer Productivity Experiment Design” (metr.org, 24 February 2026): the new data give “an unreliable signal of the current productivity effect of AI tools”, after “a significant increase in developers choosing not to participate in the study because they do not wish to work without AI”; its point estimates were speed-ups whose intervals included zero (all checked 31 October 2026). METR, “Measuring the Self-Reported Impact of Early-2026 AI on Technical Worker Productivity,” 11 May 2026 (metr.org): “In February–April 2026, we ran a survey of 349 technical workers”; “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.” (checked 8 November 2026).
- Kyle Wiggers, “ChatGPT users have generated over 700M images since last week, OpenAI says,” TechCrunch, 3 April 2025: according to OpenAI’s chief operating officer, Brad Lightcap, “over 130 million users have generated more than 700 million images” since the feature, image generation in GPT-4o, launched on 25 March 2025; also The Decoder, 3 April 2025. The pictures in the manner of Studio Ghibli spread in the days after the launch. OpenAI’s launch post (“Introducing 4o Image Generation”, 25 March 2025) says images “take longer to render, often up to one minute”.
- “OpenAI responds to criticism of ChatGPT’s Studio Ghibli-style images,” Mashable, 28 March 2025 (republished by Yahoo Tech): OpenAI’s statement, “We continue to prevent generations in the style of individual living artists” and “we do permit broader studio styles”; OpenAI, “Addendum to GPT-4o System Card: Native image generation”, 25 March 2025, §2.4.2: “We added a refusal which triggers when a user attempts to generate an image in the style of a living artist.” Also Matthias Bastian, “OpenAI blocks artist names in image prompts but allows studio styles,” The Decoder, 27 March 2025. The policy is dated; it may since have changed. On the law in India: R. G. Anand v. M/s Delux Films, Supreme Court of India, 18 August 1978, (1978) 4 SCC 118, proposition 1: “There can be no copyright in an idea, subject matter, themes, plots or historical or legendary facts and violation of the copyright in such cases is confined to the form, manner and arrangement and expression of the idea by the author of the copyright work” (Indian Kanoon, doc 1734007). Reading a palette as nearer an idea than a drawing is mine. On passing off: Colgate Palmolive Company v. Anchor Health and Beauty Care Pvt. Ltd., Delhi High Court, 29 October 2003, 108 (2003) DLT 51 (an interim injunction against passing off, on the red and white colour combination of a tooth-powder container). On marks: Trade Marks Act, 1999, s. 2(1)(zb): a trade mark is a mark “capable of distinguishing the goods or services of one person from those of others” and “may include shape of goods, their packaging and combination of colours”.
- John Ruskin, “The Nature of Gothic,” The Stones of Venice, vol. II (London: Smith, Elder, 1853), chapter VI, § 12; text and page as in The Works of John Ruskin, Library Edition, ed. E. T. Cook and Alexander Wedderburn, vol. 10 (London: George Allen, 1904), p. 192 (Lancaster University, Electronic Edition of The Stones of Venice).
- Ruskin, “The Nature of Gothic,” § 17, Library Edition vol. 10, pp. 196–197, the third of three rules; the second reads “Never demand an exact finish for its own sake, but only for some practical or noble end.” Ruskin adds that he is “reserving the enforcement of the third for another place”.
- Ruskin, “The Nature of Gothic,” § 9, Library Edition vol. 10, p. 188: “Servile ornament” and “Constitutional ornament” (the second quoted from “a will of its own”).
- John Ruskin, The Two Paths (London, 1859), Lecture I, “The Deteriorative Power of Conventional Art over Nations: An Inaugural Lecture, Delivered at the Kensington Museum, January, 1858” (Project Gutenberg ebook 7291). In the same lecture Ruskin set “the Indian race”, “these lovers of art”, against the Scottish Highlanders, whose soldiers he praised for “your avenging in the Indies”. He wrote that “nothing has ever been done by it so significative of all bestial, and lower than bestial degradation, as the acts the Indian race in the year that has just passed by” (Gutenberg’s text, which lacks “of” before “the Indian race”). He also said that “From one instance, of however great apparent force, it would be wholly unfair to gather any general conclusion”.
- Ṛgveda 10.191.2, 10.191.3 (in part) and 10.191.4, trans. Ralph T. H. Griffith, The Hymns of the Rigveda, 2nd edition (Benares: E. J. Lazarus, 1896), as on sacred-texts.com. Sanskrit of 10.191.4: samānī va ākūtiḥ samānā hṛdayāni vaḥ / samānam astu vo mano yathā vaḥ susahāsati. In 10.191.3 Griffith renders samāno mantraḥ, usually “common counsel”, as “The place is common”. On the Atharvaveda’s hymn of concord (its own parallel to 10.191 is 6.64): Atharvaveda 3.30.6, samānī prapā saha vo ’nnabhāgaḥ (a common drinking-place, and your share of food together), paraphrased; trans. William Dwight Whitney, Atharva-Veda Saṃhitā, Harvard Oriental Series 7 (1905), revised and edited by Charles Rockwell Lanman.
- Joshua T. Katz, “The Final Hymn of the Rigveda: Some Thoughts,” Journal of the American Oriental Society 144, no. 2 (2024): 417–420 (abstract; the journal is now the Journal of the American Society for Premodern Asia): the last three stanzas are devoted to saṃjñāna, “unity”, with wordplay on sam- and on the sequence mā̆n (short or long), some of whose words come from a root meaning “think” and others from samāna, “common, the same”.
- Mencius 2A:2 (Book II, Part I, chapter 2, paragraph 16), trans. James Legge, The Works of Mencius, The Chinese Classics, vol. II (Internet Sacred Text Archive): the man of Sung who pulled up his corn, and whose son “found the corn all withered”; see Chapter 19. The passage concerns nourishing one’s own nature; the application is mine.
- Hamza Shad, “State of Pre-Seed: Q2 2026,” Carta, 13 August 2026: “U.S.-based startups on Carta raised $3.19 billion across more than 11,500 pre-seed instruments” (instruments are SAFEs and convertible notes), against “$3.22 billion invested across 14,825 instruments” in Q2 2025; AI companies took 49 per cent of pre-seed dollars in the first half of 2026. Carta adds: “The final pre-seed investment volume for Q2 2026 will be revised upward as more data is entered into our system.”
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 “A company needed multiple forms of consciousness. Not one consciousness multiplied.”; that each phase of a start-up “is a question of awareness, not execution”, that “The asking is the strategy”, that “The discipline to keep asking came from consciousness”, and that a studio’s value was “the sum of what multiple conscious minds could see”; that the studio’s system was designed “unconsciously” (read here as unknowingly); that “Eight moats survive the AI era”, that a start-up should be “moated”, and that moats rest on data “no competitor can acquire” and expertise that “cannot be replicated by training on public data”; that a studio without human judgement is “a random number generator with a design degree”; that “model independence” is plumbing above which differentiation is built, with “multi-model routing and API abstraction layers” said, without a source, to be standard from “every major cloud provider”; that trust gives “a distribution advantage that no budget can replicate”, that the verification economy is “among the largest emerging opportunities of the intelligence era”, and that a studio “whose work made people stop and look was its own distribution channel”; that in 2026 a founder with traction “can almost certainly raise something”; that founders should “lead with the moat, not the model”; that revenue-based financing “limits the upside that attracts talent”; that self-funding “protects the founder’s judgement”; that “Some of the most durable businesses of the next decade will be the ones that never optimised for a metric they did not believe in”; that Arjun’s moat was “a quality of attention that could not survive acceleration”; and that the junior’s concept “was better than what Arjun would have produced alone”. It said Arjun had “lost the Goa hotel client”; it called Ria “the first employee” and then gave her boards to Neha; it named a freelancer Kavya, a name another of my books uses; it knocked a partition through the studio; it named a founder’s companies ClarityAI and TradeGrid, names that belong to real companies, and had him convince an investor by structuring his uncle’s judgement into “a proprietary asset”; it told the stories of founders in Medellín, São Paulo and Bengaluru; and it ended on “The photocopier or the studio.” and a bridge to the next chapter.
- What I now hold: when a machine can build the product and copy the founder’s manner, what a founder founds is a “we”: a few people who share one stated promise (what they offer, to whom, and who explains and puts right what fails) and are allowed to see differently in keeping it. Consciousness distinguishes nothing here; the copies were made by conscious people from the founder’s brief, tool and standard. The scarce act is the question before the build, and for a founder who works with machines it is what the others are for. In 2025 a machine could copy a studio’s manner, and one maker’s policy allowed exactly that, so a team paid to copy its founder is the dearest copier there is. Tools carry their setter’s taste, and choosing among outputs is not a human residue. A company owes the person whose work its name carries her name and a say in her idea’s reuse, and being chosen for a look pulls a studio back towards sameness. No money is neutral: each brings a resolve, and those it binds should be told before it is taken. Imitation has a place in learning, and some clients rightly want a house style; what is refused is asking for copies when more than one way of seeing was promised. Copyright protects a drawing’s form and not a look, so a say in an idea’s reuse is the studio’s to give. Telling those a commitment binds is not asking their permission, and it is the least a founder owes, not the most: those it binds should have time to answer, room to negotiate and freedom to refuse. Being allowed to see differently is not yet being allowed to act on it, and a common resolve must not make disagreement disloyal. Credit is not the whole of what a contributor is owed; pay, ownership and reuse are separate. A retainer is revenue, not capital, and savings are the owner’s own capital, not outside money; both shape a company as outside money does. A shared resolve usually needs something shared in return. A studio is an agency in the strict sense; its trap is selling copies as bespoke work, while a product company’s is treating a weekend’s build as finished.
- What changed my mind: the book’s rule that consciousness is never the hinge and that a machine’s abilities carry a date; the March 2025 image-generation episode and its policy on studio styles; Ruskin’s “The Nature of Gothic”, read with his 1858 lecture on Indian art; the last hymn of the Ṛgveda, which asks for one resolve and one mind; this book’s earlier findings that a fast copy shows a thin product, that a demonstration is not demand, that the eight moats are not walls, and that forming someone’s eye costs someone’s hours; Carta’s figures for 2026; the canon of Arjun’s studio, which kept its two rooms through October 2026; a hymn of concord in the Atharvaveda; India’s law of passing off; and three outside readings of an earlier version of this chapter, in October 2026.
- What survives: the twenty-four boards and “Twenty-four boards that looked like Arjun Mehta on a middling day”; Devika, who judges design “by what it did to the room”, and “These are lovely. But they’re all the same person. I hired a studio, not a photocopier.”; the founder who built a system to produce more of himself; round two, the terracotta and raw cotton, and “One designer out of three ... influenced least”, now with its limit; “A prototype is not a company.”; founders “who mistake creation for founding”; “The answers on day sixty are rarely the answers from day one.”; the thin wrapper around a founder’s personality, corrected; “Not every startup needs external capital. Some are better without it.”; “The question is not whether you can raise money ... The question is what the money will expect.”; and capital as alignment, now as “Whose money you take decides what you build”, with the addition that no money is neutral.
- Section titles in the print edition: 35.2 was “Building What Cannot Be Copied”. The other four keep their print titles.