Early in my career I worked in Japan for several years, for some of its largest companies. They hired me as a designer and then put me in front of their young in-house designers as a teacher. I had long hair and torn jeans. They wore the company jacket. They looked at me as if a rock star had wandered into the office, and I suspect management kept me partly for the entertainment. They also valued the work, which is the only reason the jeans were tolerated.

My sermon never changed. Question everything. Throw the brief away. Be wild and free, look for the answer nobody has tried, and keep going until you find the one that works best, however illogical or strange it looks. The students listened and took notes. Then they went back to their desks and produced designs that all looked the same.

After a while I gave up. I decided there was no creativity there, which is a convenient thing for a young man with long hair to believe. I was wrong, and it took me years to see how wrong.

The nail that sticks out

Japan has a proverb for what I was asking: the nail that sticks out gets hammered down. I wanted twenty-five-year-olds to be the nail. I could afford to be one. I was a foreigner, I would leave one day, and nothing I said in a meeting would follow me for thirty years. They had to come back on Monday.

In the West we treat creativity as one person against the rest. You have the idea, you defend it and you sign it. Japan’s crafts and martial arts work to an older scheme called shu-ha-ri: first follow the form, then break it, then leave it behind. Most people spend decades on the first step and see nothing shameful in that. I walked in and demanded the third step from people who had just started the first. A master would have laughed at me, politely.

Where the creativity was

The creativity I could not find in the sketches was in everything else. It was in the tolerances, in the paint, in the click of a switch that somebody had tuned for a year. At Toyota any worker may stop the production line when something looks wrong. For all our talk about the individual, I know few Western companies that trust a single employee that far.

The shrine at Ise is taken down and rebuilt every twenty years, a custom that began in 690. The rebuilding of 2013 was the sixty-second, and the next is due in 2033. The design does not change. Every carpenter who works on it learns from one who has done it before, so the knowledge of how to build it never has to be dug out of a book. A culture that does this has a different idea of what a designer is for.

The same patience shows in its companies. Teikoku Databank counts 2,371 Japanese firms that turn one hundred this year alone. Elsewhere a company of that age is a curiosity. In Japan it is one of a year’s intake.

And then, out of nowhere

What still surprises me is the other Japan. The country that has repeated one building for thirteen centuries also gave the world the bullet train in 1964, the Walkman in 1979, the QR code in 1994 and the first widely used emoji in 1999. None of them came from a lone rebel. They came out of teams inside a state railway, Sony, a Toyota supplier and a phone company, exactly the kind of place where I had concluded nothing new could happen.

I do not know whether the Japanese way is better or worse than ours. It has a price. People who think differently have a hard time there, and some of the best leave. But I no longer believe mine is the only way that produces new things. Mine produces more noise.

What I would teach now

If I stood in front of those rooms again I would talk less about freedom and ask more questions. How long did you work on that hinge? Who taught you? What would you change if nobody were watching? I would still wear the jeans.

The West could use some of what I failed to notice. We reward the idea and then lose interest in the thousand small decisions that make it good. The founder gets the profile in the magazine. The person who gets the door to close properly gets nothing, and in Japan that person is the hero.

I love the place. I love how its objects are made, and that a wooden temple and a vending machine can share a street corner without either looking out of place. I could see myself growing old there, with no plan except to walk around and find, every day, one thing that exists nowhere else. It would be a better education than the one I tried to give.

I am not a car engineer. I have been told so often enough. I had my first PC at twelve and was online at thirteen, as soon as that was possible, and I never studied computing either. I have taken cars apart since my first Alfa Romeos. I once put the engine of a 164 into a 156. None of this counts as a qualification.

So when I co-founded Piëch Automotive with my friend Anton Piëch, I came to the car from design and from brand. I found an industry stiff with its own success, the German part of it most of all. The thinking went: we do what works, and we do not need to start anything new. It had worked for a century, which is long enough to forget that it can stop.

A copy of everything

Our first car, the Mark Zero, was shown in Geneva in 2019. It is an electric gran turismo with the proportions of the late 1960s, a period of car design I admire without embarrassment: long bonnet, cabin set back, short tail. Designers understood it at once. It won several design awards, among them the energy efficiency prize of Design Prize Switzerland that same year.

The motoring press saw something else. Within days the car was a copy of an Aston Martin, a copy of a Porsche and a copy of a Maserati. It can hardly have been all three. A gran turismo follows the same proportions whoever builds it, because they come from where the engine, the people and the luggage go. The reviewers recognised the type and called it theft.

The German press had also decided who we were: two young men with no idea, building a sports car for fun, with no chance against Porsche. I was reminded in print and in person that I was not an automotive engineer and had no business in the trade.

The offer nobody read

A few years later I founded Quantum Group AG in Switzerland with one purpose. An investor from the Middle East, with the means and the patience, wanted to buy Lamborghini, and he liked what I had done with cars. In May 2021 we sent Volkswagen a letter of intent worth €7.5bn, more than nine billion dollars. It came with a five-year supply agreement with Audi, a five-year job guarantee and a new centre for battery technology in Germany.

Audi’s spokesman said Lamborghini was not for sale and that the group would not consider the offer. That is an owner’s right. What stayed with me is that nobody asked what was in the letter. I was young and had not spent decades on a supervisory board. I hold that against myself less than I used to, given what decades on supervisory boards have produced since.

Then the letter leaked. German journalists called, and what they wrote was a portrait of me more than a report on an offer: a curious investor with no background who did not have the money. I had never claimed the money was mine. I said so every time I was asked, and it changed nothing. Some went as far as doubting that the offer existed.

I took it badly. I withdrew from public life for several years, and it was a long time before I had the nerve to start companies again. I still doubt that I was the real subject of those articles. Mine was the name in them, though.

What certainty bought

At the end of June this year Germany’s car industry employed 691,500 people, the fewest since 2005, according to the federal statistics office. It had shed 42,300 jobs in twelve months. Volkswagen is removing at least 50,000 positions in Germany. I take no pleasure in any of it. I love the product, and I always have.

But an industry that could not listen to an outsider with a sketchbook, or read a letter before refusing it, did not listen to the Californians or the Chinese either. Those it could not wave away.

Expertise is real. I would not board an aircraft designed by an amateur. The trouble starts when a qualification becomes a reason not to listen. Everyone in that industry had the training I lacked, and the training did not save them. Outsiders are wrong most of the time. They also ask most of the new questions, and an industry that has stopped hearing questions has begun to die, however good the quarter looks.

As for the press, I am older and calmer now. Anyone who does things in public will meet people who look for the worst in it and who, finding nothing, write it up so that it sounds bad anyway. I would tell a younger founder what I could not tell myself then. Read it once, check whether any of it is true, and go back to work.

Founders spend years on the product and an afternoon on the company. The product gets drawings, prototypes and tests. The company gets a template from a notary and an organisation chart copied from the last place the founder worked. Then it grows like coral, one hire and one committee at a time, and after ten years nobody can say why it has the shape it has.

I am an industrial designer, so I see a company as an object with parts. The questions are the same as for a coffee machine. How many parts does it need? Where are the joints? Which part can be replaced without opening the whole thing? A good product has few parts and clean joints. So does a good company.

What a firm is for

In 1937 a young economist, Ronald Coase, asked why firms exist at all, when in theory a market could supply every task. His answer was cost. Finding a supplier, agreeing a price and checking the work all cost something, and when that cost is higher than doing the job yourself, you hire. A firm is as large as its transaction costs make it.

Those costs have been falling for thirty years. Software that can read a contract, compare offers and check a delivery is pushing them down again. Much of what sat inside a company because it was awkward to buy can now be bought. What stays inside should be the part nobody else can do.

Own the core, buy the rest

Nvidia designs chips and owns no factory that makes them. In its 2025 financial year it turned over $130bn with about 36,000 people. Volkswagen needed some 680,000 people for €325bn in 2024. The two are in different trades and the comparison is unfair. It still shows what happens when a company keeps the thinking and lets others own the machines.

The opposite mistake costs as much. Boeing handed around 70% of the 787, including much of the design, to partners across the world. The aircraft arrived more than three years late. Boeing had given away the knowledge of how the parts fit together, which was the one thing that made it Boeing. You can buy capacity. You cannot buy an understanding of your own product, and a company that tries ends up as a logo with a purchasing department.

So the first design decision is to name the core in one sentence. In most companies I know it is some mix of the design, the brand and the data. Everything else is a candidate for a contract.

Small units, real owners

The second decision is size. I build small companies that each do one thing, under a holding that supplies capital and keeps out of the way. Ten people who own their result behave differently from a department of ten inside a group of ten thousand. They know their customers by name, and they cannot hide a bad quarter in somebody else’s budget.

The third is ownership. A salary buys attendance. A share of the outcome buys attention. I would sooner have five people who own part of what they build than fifty who are waiting for Friday. I do not do this to be generous. Owners supervise themselves, which makes it the cheapest control system there is.

What I would not copy

None of this is an argument for the company without structure. Someone decides, and everyone knows who. The fashionable alternative, the firm as a flat network of equals, mostly produces meetings.

Nor is small always right. Aircraft, chips and power stations need scale, and somebody has to own the factory. But even there the companies that last are built from small units that answer for themselves. The family firm with three hundred people and half of a world market has understood something the conglomerate forgot.

A company is the tool its founders work with every day, for decades. It deserves at least the care we give a kitchen appliance. Draw it before you build it. Count the parts. Then take a third of them out.

Europe likes to call itself a power. It has 450 million people, one of the three largest economies on Earth and more rules than anybody. What it lacks is the ability to say no. A power can refuse. A customer can only complain about the price.

Look at what Europe buys. Its security comes from the United States: between 2020 and 2024, 64% of the arms imported by the European members of NATO were American, by the count of the Stockholm institute SIPRI. Since it gave up Russian pipeline gas it has turned to tankers, and 57% of the liquefied gas they brought in 2025 was American as well. About 70% of its cloud computing runs on the servers of three American companies. The models that are about to sit in every office are built in California and, more and more, in China.

The summer of 2025

Two meetings in the summer of 2025 showed what this is worth. In June, in The Hague, the European members of NATO promised to raise spending on defence and related infrastructure to 5% of GDP by 2035, because Washington had asked. In July, at a golf resort in Scotland, the European Commission accepted a 15% tariff on most of its exports to America and promised to buy $750bn of American energy. It was presented as a deal. A deal is what you get when both sides can walk away.

I do not blame the Americans. They are doing what any supplier does with a customer who has nowhere else to go. I blame the customer.

How a continent became a customer

It happened through comfort. After 1990 Europe decided that history was settled and that it could concentrate on regulation, tourism and very good cars. Defence was America’s job. Cheap energy was Russia’s, manufacturing was China’s and software was California’s. Each choice made sense on its own spreadsheet. Together they produced a continent that writes the rules for technologies it does not build.

Mario Draghi put numbers on it in 2024. Only four of the world’s fifty largest technology companies are European. No company founded in the European Union in the past fifty years has grown from nothing to a value above €100bn, while America created all six of its trillion-dollar companies in that time. To catch up, he wrote, Europe would have to invest an extra €750bn to €800bn every year. The report was applauded and filed.

What a power makes

I design things for a living, so my test of power is crude. Can you make it yourself, or do you have to ask? A power makes its own energy, chips, weapons and models, or enough of each that asking is a choice. By that test Europe has the parts of a power lying around. It has ASML, without whose machines nobody on Earth makes an advanced chip. It has Airbus. It has nuclear engineers, toolmakers and more good universities than it knows what to do with.

What it lacks is the decision to assemble them. Its armies run more than a dozen kinds of battle tank where America’s runs one. A young company that wants to sell across the continent meets twenty-seven sets of company law. Around €300bn of European savings leaves every year, most of it for America, because nobody at home offers it anything to build.

I write this from Switzerland, which is not in the Union and has its own habit of living off other people’s decisions. I am not asking for a European superstate. I would settle for a continent that can build a reactor, a chip and a rocket without filing a request abroad.

None of this is beyond Europe. It built Airbus when everybody said Boeing had won. ASML started in 1984 in a leaking shed beside a Philips factory. Both began with a decision, taken by people who were tired of asking. A well-behaved customer gets a discount now and then. I would like Europe to want more than that, and to decide soon, while the choice is still its own.

Every product has a bill of materials, the list of all the parts it needs, down to the last screw. Designers learn early to read it for the dangerous line: the part with one supplier. A machine with three hundred parts ships when the three-hundredth arrives. How good the other 299 are makes no difference.

Europe has never read its own bill of materials. It should, because the dangerous lines are easy to find.

One supplier

According to the European Parliament’s research service, the Union gets 98% of its rare-earth magnets from China. These are the magnets in every electric motor, wind turbine and guided missile. China also supplies 97% of the magnesium Europe imports, without which there is no aluminium alloy, and 71% of its gallium, which goes into radar and power electronics. Chinese plants hold about nine-tenths of the world’s refining capacity for critical raw materials, wherever the ore was dug up.

In 2025 Beijing showed what one supplier can do. It made exports of rare earths subject to a licence. A survey by the European chamber of commerce in China found that of 141 applications from its members, 19 had been approved by early September. Car suppliers in Europe stopped production lines for want of a component that weighs a few grams.

A few weeks later came Nexperia. The company, based in the Netherlands, makes the simplest chips there are, the kind that switch a headlight or watch over a battery, and has belonged to a Chinese group since 2018. In October the Dutch government took control of it under an emergency law. China answered by blocking exports from Nexperia’s assembly plant in Guangdong. Honda stopped a factory in Mexico. European carmakers ran on their reserves and warned that the lines would stop. Nobody had thought of these chips as strategic. They cost almost nothing, which is why nobody had bothered with a second source.

It was not always so

The word I care about is again. Forty years ago a single plant in La Rochelle, on the French Atlantic coast, was one of the largest rare-earth refineries in the world, and its output set the world price. Europe let that business go to China because it was dirty and cheap, and because the spreadsheet said so. Today the plant produces about 4,000 tonnes a year. China produces roughly 300,000. The chemists are still in La Rochelle. What went missing was the will to pay for the unglamorous half of industry.

Brussels has since written a law. By 2030 the Union wants to mine 10% of what it uses of each strategic material, process 40% and recycle 25%, and to take no more than 65% from any single country. In February its own auditors looked at the plan and concluded that Europe is unlikely to get there in time. Ten of the materials on the list are not recycled at all.

What independence costs

Independence does not mean making everything at home. No designer wants that and no economy could pay for it. It means a second source for every line that can stop the factory. Some of those sources will be mines in Sweden, Portugal or Greenland, and people will object to them. Some will be refineries, which smell. Some will be recycling plants that pull the magnets out of old motors. All of them will cost more than buying from the cheapest seller, in the way a fire door costs more than a curtain.

The price of not paying is now known. It is a car plant standing still for want of a chip, and a foreign policy that has to be cleared with a supplier first.

I have spent my working life making complicated things simple to build. The rule that has never failed me is to design the supply before the product. If a part has one source, you change the part or you find a second source, and you do it before the launch. Europe is doing it after. That is more expensive. It is still cheaper than finding out in a crisis that your sovereignty was on back order.

Look at any street photograph from the 1950s. A man waiting for a tram, a woman carrying her shopping, workers at a factory gate. Almost everyone in the picture is better dressed than almost everyone in the same street today. They were not richer. Most of them owned a fraction of the clothes we do.

I should declare an interest. My sister is a fashion designer. With a partner she ran the Zürich label RS Hader, which set out to make what it called honest clothes: reduced shapes, exact cuts, prints of its own. Fashion was never a foreign country in our family, and I have opinions.

Why they looked better

The reasons are practical. People had few clothes, so each one had to be good. A coat was bought to last ten winters and was altered when its owner changed shape. Fit came from a tailor or from someone at home with a needle, which is why a cheap suit of 1955 sits better than an expensive one bought online. The cloth was wool, cotton and linen, which hang and age in ways polyester does not.

There was also an idea behind it that we have almost lost. Dressing was a courtesy. You dressed for the people who had to look at you: the client, the teacher, the other guests. It said that the occasion mattered and that you had made an effort for it. By class I mean that quiet effort, and it was never a question of money.

How it ended

The effort went out of fashion in two steps. First the office gave up. In 1992 Levi’s sent a booklet called A Guide to Casual Businesswear to 25,000 personnel managers in America, in order to sell more khaki trousers. It worked better than any campaign should. Within a decade the suit was a costume for weddings and court appearances.

Then clothes became disposable. In the fifteen years to 2015 the world doubled the number of garments it made, by the count of the Ellen MacArthur Foundation, while the number of times each one was worn fell by more than a third. Less than 1% of the material ends up in new clothing. A T-shirt now costs less than a sandwich and is designed to be photographed once.

Freedom was the promise. Wear what you like, be yourself. What we got was another uniform, and an uglier one: the hoodie, the leggings, the trainers the size of bread loaves. When everything is permitted most people choose comfort, and comfort has no shape.

Why it is coming back

I think the tide is turning, and nostalgia has little to do with it.

The first reason is scarcity. When a machine can generate a flawless picture of anyone in anything, the picture stops being worth much. What cannot be generated is a person standing in a room in clothes that fit. How you actually turn up becomes the one signal that is hard to fake, and people will invest in it again.

The second is waste. A generation that grew up sorting its rubbish is working out that the most sustainable garment is the one worn three hundred times. Second-hand shops and alteration tailors are full of people under thirty. They are relearning what their great-grandparents knew: buy less, buy better, have it altered.

The third is boredom. Every fashion reacts to the one before. After thirty years of looking as if we had just left the gym, making an effort is the only rebellion left.

The same principles, other clothes

It will not look like 1955. Nobody is bringing back the hat and the girdle, and nobody should. The form will be new: lighter cloth, softer construction, shoes made to be worn with a jacket and a bicycle. The principles underneath do not change, because they come from the body and from other people. Fit. Proportion. Good material. A sense of occasion. Restraint.

A designer notices that these are the principles of any good object: a chair that fits, in honest material, made for its purpose, without decoration it does not need. For a while we accepted that clothes were exempt. I do not think they are, and I look forward to streets that are worth photographing again.

The Nespresso Pixie is 11 centimetres wide. I worked on it. Into that width go a pump rated at 19 bar and a heater that is ready in 25 seconds. Espresso once needed a café and somebody trained to work the machine. The Pixie needs a shelf.

What happened to coffee is now happening to two far larger machines. Small language models run on a phone, and small modular reactors can sit beside the factory or data centre they feed. The industries that build them still worship scale and treat the small versions as curiosities. They have it backwards. When a thing gets small, somebody else can own it, which makes miniaturisation a transfer of power.

Who owns the big one

Big machines belong to institutions, because only institutions can pay for them and wait for them. The two new reactors at Plant Vogtle in Georgia were planned for 2016 and 2017 at a cost of $14bn. They entered service in 2023 and 2024 at more than $30bn. In December 2023 the state’s regulators let Georgia Power, one of the four owners, raise its rates to recover $7.56bn of its remaining costs. Only an owner with captive customers could have carried a bill like that.

Artificial intelligence has copied the arrangement. In January 2025 OpenAI and its partners announced Stargate, a plan to invest $500bn over four years in computing infrastructure. The largest models live in buildings of that kind, and everybody else rents them by the token. The landlord sets the price and can change the locks.

Computing has been here before. The mainframe sat in a machine room and belonged to the company, until the personal computer put a lesser machine on a desk and gave it to whoever paid for it. Nobody now calls the PC a weak mainframe.

Small language models, small reactors

In April 2024 Microsoft published Phi-3-mini, a model with 3.8bn parameters that its researchers ran offline on an iPhone 14, in about 1.8 gigabytes of memory. By Microsoft’s own tests it rivalled GPT-3.5, the model family ChatGPT launched on in late 2022. In August 2025 Google released Gemma 3 270M, with 270m parameters. In Google’s tests on the chip of a Pixel 9 Pro, 25 conversations used 0.75% of the battery.

Models of this size do less than the giants. They also run on a device the user owns and send nothing to anyone. A law firm can run one over its own files. So can a hospital, the customer least able to post its records to somebody else’s server.

Reactors are taking the same road. The International Atomic Energy Agency defines small modular reactors as those producing up to 300 megawatts of electricity per module, and counts more than 80 designs and concepts. Deep Atomic, where I am a board member, develops the MK60, which is smaller again, at 60 megawatts of electricity per unit, and is built for AI data centres. At that size the buyer can be a single company instead of a state’s ratepayers.

A data centre that brings its own power stops waiting for the grid, and the company that owns its power owns its timetable. It is a fair joke that the first customers for small reactors are the buildings that house the biggest models.

The efficiency objection

The objection is that small things are weaker and cost more per unit, and it is partly right. Economies of scale are real. In November 2023 NuScale and a group of municipal utilities in Utah cancelled what was planned as America’s first small modular reactor plant, after its target price rose from $58 to $89 per megawatt-hour and too few buyers signed up. No model on a phone will match one that fills a data centre, either.

Cost per unit is the institution’s measure. It counts the output and leaves out the price of admission, which is a decade of building and the capital that waits through it. An owner wants to know whether the machine is affordable at all and whether anyone can take it away, and small is a strategy because it answers both. The buyers in Utah were towns shopping for a reactor. The price stopped them, but the size is what let them into the shop.

Small things also get cheaper in a way big ones do not, because they are made in series, and whatever is built by the thousand improves with every run. Vogtle’s builders got two tries. The first small reactors will be expensive, as first products are, and the comparison that matters is with the tenth.

As for weakness, most work is small work. A model that sorts invoices or answers questions about one company’s manuals has no use for the whole internet. Google calls its 270m-parameter model the right tool for the job, which is a polite way of saying that the big one is often the wrong tool.

The Pixie’s water tank holds 0.7 litres. That is a hopeless figure for a café and the right one for a kitchen, and the kitchen is where the espresso went.

Microsoft sells software. On 20 September 2024 it signed a 20-year contract for the electricity from a nuclear reactor. The reactor is Unit 1 at Three Mile Island in Pennsylvania. Its owner had shut it in 2019 because it lost money. Its neighbour, Unit 2, is the one that partly melted down in 1979.

That contract says more about the future of thinking machines than any benchmark. Artificial intelligence does not fail for lack of intelligence. It fails for lack of electricity. AI energy demand is now the hard edge of the industry, and whoever designs the power plant designs the limits of AI.

My training is in design, and I have designed devices for Sony, Panasonic, Samsung and Nespresso. The brief for a device of that kind is about size, price, material and the date it must reach the shops. Power is a plug and a wall socket. For the machines being built now, the first line of the brief is megawatts, and everything else follows from it.

The queue for a socket

The numbers come from the International Energy Agency, which is nobody’s idea of a hype merchant. Data centres used about 485 terawatt-hours of electricity in 2025, 17% more than the year before, while the world’s demand for electricity grew by 3%. Consumption by the centres built for AI rose by half. The agency expects the total to double by 2030, to roughly 950 terawatt-hours, which is about what Japan consumes in a year.

Demand is the easy part. Supply is made of copper and paperwork. In Frankfurt, London, Amsterdam, Paris and Dublin, Europe’s five big hubs, a developer waits seven to ten years on average for a grid connection, according to the same agency. Dublin and Amsterdam have at times paused new projects altogether, because the grid had nothing left to give. A large power transformer can take up to four years to procure, and lead times have almost doubled since 2021.

The agency reckons that grid constraints could delay a fifth of the data centre capacity planned worldwide by 2030. Set that against the calendar of the people who run the AI debate. They are software people. Their argument is about what the models can do and who will own them, and almost none of it is about where the current comes from.

They ship a new model every few months and treat data centre power as somebody else’s problem, a matter for procurement. A transformer has no interest in a release schedule. In America, developers tired of waiting for the grid are parking gas turbines beside their buildings, and orders for turbines jumped by 70% in 2025.

AI energy demand outruns efficiency

Optimists reply that chips and models keep getting more efficient, so the problem will solve itself. The first half is true. By the agency’s account the energy used per AI task has fallen at least tenfold a year in recent years, a pace it calls unprecedented in energy history. In 2025 consumption still went up by 17%.

William Stanley Jevons explained why in 1865, in a book about coal. James Watt’s steam engine got more work out of every ton than the engines before it, and England burned more coal as a result, because cheap work finds new uses. Cheap intelligence behaves the same way.

A question answered in a few lines of text costs little. Video generation and software agents can use hundreds or thousands of times more energy per query, and that is where the industry is heading. Efficiency lowers the price of a task, and the customers order more tasks.

Megawatts on line one

So the technology companies have gone shopping for reactors. By the agency’s count, conditional agreements between data centre operators and developers of small modular reactors grew from 25 gigawatts at the end of 2024 to 45 gigawatts by April 2026. The word conditional matters. The reactors still have to be built.

I am an interested party. Deep Atomic, on whose board I sit, develops such a reactor: the MK60, a compact unit made for AI data centres. Each one supplies 60 megawatts of electricity and 60 megawatts of cooling.

The second figure is the one a designer notices. Nearly every watt that goes into a server comes out as heat, so a hall full of servers has two needs: power to compute and cold to carry away the result. Most power stations sell electricity to whoever sits at the end of the wire and regard the customer’s heat as the customer’s affair. A plant drawn for this customer delivers both.

This is what I mean by a design brief. The power plant and the computer are one object, and they have to be drawn as one. A reactor sized for a town and a data centre bolted on later is two briefs and a long cable. The companies that grasp this will decide where AI is built and how much of it there is. The others will have excellent models and a place in a seven-year queue for a socket.

The first Apple Watch could be had in a case of 18-carat gold, for up to $17,000. That was in 2015. Three years later the company’s own software update, watchOS 5, would not run on it, and in 2023 Apple added the watch to its list of obsolete products. Patek Philippe undertakes to maintain and restore every watch it has made since 1839.

The distance between those two promises is the future of luxury. Gold did nothing for the Apple Watch, because material was never what made a thing last. For a long time luxury was defined by price and by the scarcity of what it was made from. Now that AI makes almost everything fast and cheap, another scarcity takes over: time that refuses to be compressed. Luxury is what cannot be accelerated.

Where scarcity went

Scarce material is already a weak foundation. De Beers launched Lightbox in 2018 to sell laboratory-grown diamonds and announced in May 2025 that it would close the business. By then, the company said, wholesale prices for lab-grown stones in jewellery had fallen by 90%. A diamond can now be a manufactured good, priced like one.

AI does to competence what the laboratory did to carbon. I founded a lab in Zürich that works on language models, so I say this without nostalgia: the price of a decent draft or a passable design is heading towards zero. Nobody pays a premium for what a machine produces in a second.

What keeps its price is anything with a clock inside it that nobody can wind faster. The iris of fine perfumery spends three years in the ground, and its root then dries for about three more before a perfumer can use it. At A. Lange & Söhne in Glashütte every movement is assembled twice: built and adjusted, taken apart, cleaned and finished, then built again. No quarterly target shortens either process.

I have a stake in this view. I co-founded Collectif Noir, an independent watch house in Geneva whose small series of mechanical watches run on its own movement, and I am a seed investor in Thomas De Monaco Parfums, an independent perfume house in Zürich. With one foot in a model lab and the other in a watch house, I see two industries misread luxury and AI from opposite sides.

Two industries, both wrong

The luxury industry believes it must speed up to look modern. It has borrowed the drop from streetwear and the gadget from Silicon Valley. In August 2022 Tiffany sold 250 digital passes called NFTiffs, each redeemable for a pendant, to owners of CryptoPunk images. They cost 30 ether apiece, about $50,000 at the time, and sold out in roughly 20 minutes.

Selling out in 20 minutes is an achievement for a ticket agency. A house whose value lies in patience teaches its customers to expect speed, and speed is the one contest it will lose to anyone with a server. The other reflex is the price tag. Bain’s 2024 luxury study counted about 50 million customers lost in two years and pointed to brands’ continued price rises. A higher price with no more time in the product is a tax on loyalty, and those customers declined to pay it.

The technology industry makes the opposite mistake. It thinks luxury is a logo on a premium tier. Hence the gold Apple Watch: a product on an annual release cycle, dressed as an heirloom. Anything that improves every year is out of date every year, which is a virtue in software and fatal in an object meant to outlive its buyer.

The future of luxury

A sceptic will call all of this romance. Luxury, he says, is marketing and status, and has been since Thorstein Veblen described conspicuous consumption in 1899. A mechanical watch keeps worse time than a phone. Its buyer pays to be seen.

I grant the premise and reject the conclusion. The sceptic is right about much of what is sold as luxury: a factory product at a workshop price. But status attaches to whatever is hard to get, and what is hard to get changes.

When machines make material and competence abundant, status moves to the one input that cannot be manufactured or copied. Marketing can claim time, but it cannot fake it for long, because a campaign made in weeks is imitated in weeks. Six years of iris take six years to imitate.

The sales figures agree. For 2024 Bain forecast the first fall in personal luxury goods since the financial crisis, the pandemic aside, and the fashion and leather division of LVMH shrank by 1% on an organic basis. Hermès grew its sales by 15% at constant exchange rates. A Kelly bag there can take 20 hours to make, four of them for the handle. "You cannot compress time, at one point, without compromising on quality," its artistic director, Pierre-Alexis Dumas, told CBS that December.

An iris planted this year will reach a perfume bottle in 2032 at the earliest. No software update will bring that date forward.

Tesla has paid people to walk. The job, advertised in Palo Alto in August 2024, was called Data Collection Operator. The work was to put on a motion-capture suit and a virtual-reality headset and walk for more than seven hours a day, carrying up to 30 pounds of equipment. The pay ran from $25.25 to $48 an hour. Applicants had to be roughly the height of the robot they were walking for.

That advertisement is the most honest thing the robot industry has published. Tesla builds the robot itself, and the one input it could not manufacture was a person moving. The race is told as a contest of better hardware and bigger models. The real shortage is robot training data, and the scarcest kind is tacit knowledge: the skill of ordinary working people, which nobody ever wrote down.

Tacit knowledge has no archive

Language models had it easy. People wrote everything down and then put it online. Ken Goldberg, a roboticist at Berkeley, reckons that the text used to train large language models would take one person about 100,000 years to read. The largest robot dataset reported so far, he wrote in Science Robotics in August 2025, is on the order of one year.

Money will not close the gap. The internet is a record of what people could say. Nobody wrote down how a cook holds a knife or how a watchmaker seats a wheel, because nobody could. Ask the cook and she will show you. It takes years.

Michael Polanyi, a physical chemist who turned philosopher, called this kind of knowledge tacit. In The Tacit Dimension, published in 1966, he set out from what he called the fact that "we can know more than we can tell". His example was the human face, which a person can pick out among a million without being able to say how. In 2014 the economist David Autor called the consequence Polanyi’s paradox: a machine cannot be handed the rules for a task whose rules nobody can state. He also named the way round it, which is to build machines that learn from human examples.

I know the limits of telling from the inside, because the user manual was once my trade. I worked with Electrosuisse and the VDE to have it certified along with the product, as something its safety depends on. A manual is the most serious attempt industry makes to put into words what a hand must do. It can say which button to press.

The Nespresso machines I worked on took the other route and removed the skill: a capsule and a button stand in for the barista. A robot in a kitchen or a workshop has no such exit. It has to acquire the skill, and a skill that cannot be told can only be shown.

What simulators cannot feel

The standard reply is that the showing can be had for nothing. A simulator produces practice faster than any workshop, and has taught robots to run and to do backflips. YouTube holds about 35,000 years of video, by Goldberg’s count. On this view nobody need pay a cook, since a computer can invent a million kitchens and the internet has already filmed the rest.

Neither reaches the hand. Goldberg, who is no enemy of his own field, calls simulation "notoriously unreliable" for manipulation. Recovering the motion of fingers and objects from flat video he calls a grand challenge for computer vision, one not expected to be solved in the foreseeable future.

The reason is simple. Skill lives in contact: in the pressure of the fingers and the give of an onion under a blade. A simulator has to guess those forces and a camera across the room never saw them. A video of a cook proves that the onion got chopped.

The industry’s own conduct settles the matter. Tesla did not advertise for more simulation. It handed its walkers a suit, and nobody asked them to explain how they walk. Many companies, Goldberg notes, are assembling fleets of robots and people to collect data by teleoperation: a person steers the machine through a task such as folding a towel, again and again. An industry that can buy any amount of computing power is hiring walkers and towel folders.

The past of work

The cook and the machinist are usually filed under the past of work, as the people automation is coming for. They are its training set. Every robot that does useful work with its hands will do it because a person did it first and somebody thought to record how.

That is the reason I co-founded Archaí Physics, which records how people really work so that robots can learn from it. Its first device, 360°, is worn like headphones and is in development. The record has to be made at the bench and at the stove, by the people who work there.

It also changes what counts as scarce. Chips can be ordered and models can be copied. A watchmaker with decades at the bench can be neither. Collectif Noir, the watch house I co-founded in Geneva, builds its own movement. The drawing of a movement specifies every part and says nothing about the pressure that seats a wheel. That is kept in a pair of hands, and so far nowhere else.

Alberto Cantino, an agent of the Duke of Ferrara, was sent to Portugal officially to trade horses. In November 1502 he reported to his master on a different purchase: he had paid twelve gold ducats for a map. Portugal guarded its charts as state secrets, and this one showed what the secret was worth. It had the newly found coast of Brazil on it, and the line that divided the world beyond Europe between Portugal and Spain.

Ships went where the charts said, so whoever drew the chart decided what counted as a coast and whose it was. That contest has started again, and this time the territory is work. Physical AI, the kind that acts in the world instead of describing it, needs a record of how things are actually handled: how a hand seats a screw or turns a patient in a bed. Whoever makes that record sets the terms for everyone who builds on it. Europe is rich in exactly this material and is looking the other way.

What physical AI lacks

The text that language models learned from was already written and lay on the open web for the taking. No such library exists for manual work, and none can be typed up now, because very little of a skill survives being written down. World models, and the world action models that turn them into movement, have to be fed with recordings of the work itself.

America owns the computers. The research group Epoch AI credits the United States with about 75% of the AI supercomputing power it could trace, and China with 15%. What America lacks is the footage.

In October 2026 NPR reported from Karur, in southern India, where textile workers pack aprons with cameras on their heads and others film themselves doing housework. Those paid for it get about $2.60 an hour. One textile worker said nobody had told her what the footage was for. Most clients for the household recordings are in America. The country with three-quarters of the machines is learning housework from Tamil Nadu.

China has the factories and the robots. In 2024 it installed 295,000 industrial robots, 54% of the world’s total, by the count of the International Federation of Robotics. America installed 34,200.

China also knows what it is missing. In Beijing a centre of 10,000 square metres has opened where humanoid robots practise in 16 staged settings, a care home among them, and produce more than six million data points a year. A staged care home has no patients in it. Mass production works by taking judgment out of the task, which is its genius and makes it a poor teacher of judgment.

A law and a lament

Europe’s AI strategy so far amounts to a law and a lament. The law is the AI Act, in force since 1 August 2024. The lament is about chips and cloud, and the numbers support it: European providers hold about 15% of their own continent’s cloud market, down from 29% in 2017. The law governs models mostly built elsewhere, and the lament envies the machines they run on. Neither looks at what Europe has.

Germany’s craft register lists more than a million businesses. In Switzerland two-thirds of young people start out in vocational training. Add the hospital wards, the restaurant kitchens, the dental laboratories and the watch benches of Geneva, where I co-founded an independent watch house, and few places can match that stock of trained hands. Skill does not appear on a balance sheet, so nobody counts it. Each of those workplaces holds what physical AI cannot get from a data centre or a production line.

Recorded on Europe’s terms, that stock is an asset: the worker agrees and is paid, and the workshop licenses the record instead of handing it over. Recorded on someone else’s terms it is a giveaway, and Karur shows the going rate. Making such a record is the business of Archaí Physics, which I co-founded. Its first device sits on the head like a pair of headphones and is still in development. At OptAI in Zürich we work on the world models that will need it.

Data has no passport

Data, the objection runs, has no nationality. Record what you like and it will flow to the best model, which will be American or Chinese. That was true of text, because the text already existed and nobody had to ask. A skill has to be recorded in the room, with the person present, by someone who chose the sensors and wrote the licence. A file can be copied without limit, which is exactly why the first contract is the one that counts.

Standards follow the record. When delegates met in Washington in 1884 to pick a prime meridian, they chose Greenwich by 22 votes to one. Nearly two-thirds of the world’s ships already sailed by charts drawn from it, and the majority saw no reason to redraw them. Models will come and go. The format in which a continent’s skill was first written down will stay, and so will the name on the licence.

Portugal drew the map and Ferrara got it for twelve ducats. It has never gone home. It is in Modena.

Philippe Starck’s lemon squeezer has three thin legs and neither strainer nor bowl. Alessi, the Italian firm, has made it since 1990, and the Museum of Modern Art in New York keeps one in its collection. For the tenth anniversary Alessi plated 10,000 of them in gold. Lemon juice spoils gold plate, so the anniversary squeezer was never meant to meet a lemon.

I hold nothing against it. It is a good small sculpture, and the gold edition admits as much. My quarrel is with the lesson the trade took from it: that the question of what is good design can be settled by looking. It cannot. A thing proves itself the thousandth time a hand reaches for it, and use is the only critic that cannot be charmed.

Judged on day one

In 2025 the iF Design Award drew almost 11,000 entries from 66 countries. Half went through to a final jury of 102 people, who met in Hamburg for three days. I do not doubt their eye. A jury can pick an object up and switch it on. It cannot own it for ten years.

The same holds for the launch photograph and for the review written in the first week. Each judges an object on its first day, when it is clean and has never been dropped. A first-day verdict favours whatever is new to the eye, and novelty is the one quality sure to be gone by the second year.

I worked on the Citiz and Pixie machines and the Aeroccino milk frother for Nespresso, and my work has won several design awards. I am not handing them back. Prizes go to machines that are new, though, and limescale keeps its own calendar.

Screens have made the bias worse. In 2014 IKEA said that about three-quarters of its product images were computer-generated, which means an object can now sit for its portrait before it exists. Most products are first met as a picture the size of a thumb and bought on that evidence. I trained in photography, so I know what a lens forgives. Design culture has become a branch of photography, and it rewards the object that photographs well over the object that lives well.

The thousandth morning

Anyone who wants to know what is good design should ask a coffee machine. It is judged every morning for ten years, some 3,650 sittings, by a person who is half asleep and has no interest in design. That person reaches for the machine without looking at it. This critic has never seen the press kit.

Good product design is whatever survives that reach. The lever sits where the hand already is. The drip tray comes out without spilling what it caught. Neither shows in a photograph, and both are decided at the drawing board, long before any jury meets.

The honest critics have plain names: wear, habit, cleaning, repair. Wear shows where hands really went, which is rarely where the designer drew them. Habit is the severest of the four: once a movement is learned the object disappears from attention, and that disappearance is the highest mark it can earn. Cleaning finds every seam that was put there for the camera. Repair asks whether anyone thought about the day after the warranty ends.

Then comes the manual nobody reads. In Japan, on behalf of Electrosuisse and the VDE, I worked to have the user manual counted as part of a product’s safety and of its certification. European machinery law ranks the remedies: a danger must first be designed out or guarded against, and only what remains may be handed to the user as a warning. Read that way, a manual is a book of confessions, each caution marking a spot where the object could not speak for itself. A thick manual is a bad review written by the maker.

Beauty earns its keep

The case against all this is that beauty matters for its own sake, and that a world of obedient appliances would be a dull one. Beauty does matter. The mistake is to hand it to the eye alone. The word aesthetic comes from the Greek for perception by the senses, and the hand is one of them: the weight of a lid and the sound of a lever closing are aesthetic facts as much as any silhouette.

Dieter Rams, whose ten principles of good design are quoted more often than they are obeyed, put the matter flatly: "A product is bought to be used." He also held that the aesthetic quality of a product is integral to its usefulness, because the things a person uses every day affect their well-being. Beauty of that kind is confirmed by use, and it earns its keep: a thing that is a pleasure to hold gets mended and kept.

Beauty that cannot bear to be touched belongs on a plinth, and there is no shame in a plinth. Starck’s squeezer has its museum. A coffee machine gets no such mercy. It gets Monday.

On the control panel at Three Mile Island a light told the operators that a relief valve had closed. The valve was stuck open. All the light reported was that the power which opens the valve had been switched off. It was about four o’clock on the morning of 28 March 1979, and coolant went on escaping for more than two hours. Within minutes a hundred alarms were sounding around a crew trying to understand its own machine.

The radiation that got out did negligible harm to anyone’s health, by the regulator’s own account, but the harm to the name of nuclear power has lasted to this day. I count that light as a failure of design before a failure of engineering: a control that says something other than what the machine is doing is a lie built in at the factory. It also shows how design and trust are joined. People trust what is well made, because the care they can see is the only evidence they get of the care they cannot. Beauty is that evidence.

What the eye audits

Almost nobody can inspect a pressure vessel or a neural network. Anybody can inspect a weld seam, or hear whether a door closes with one sound or with three. From such things people infer the rest, and the inference is sounder than it looks. A firm that sweats over the part you see rarely skimps on the part you do not, because both come from the same habit.

In 1995 Masaaki Kurosu and Kaori Kashimura, two researchers at Hitachi’s Design Center in Tokyo, showed 26 layouts of a cash machine to 252 people. How easy a layout looked to use depended more on how beautiful it was than on how easy it really was. The trade calls this the aesthetic-usability effect and files it under bias. I call it the way people have always judged things they cannot open.

My own trade taught me the same from the other side. I designed machines for Nespresso and devices for Sony and Panasonic, where the buyer reads a seam slightly out of line as a verdict on the whole company. At Electrosuisse I built a department on one premise: the instructions belong to the safety of a machine, and its certificate should cover them. An instruction nobody can follow is a defect, like a frayed cable.

Why design matters most

Nuclear power and artificial intelligence need public trust more than any other technology, and both present themselves badly. The face of nuclear power is the cooling tower. It emits water vapour and nothing else, and coal plants have them too, yet it became the icon of dread. The face of AI is the chat window: an empty box with a cursor, which gives no sign of what the machine knows and what it merely guesses.

The makers of both answer fear with numbers. Our World in Data puts coal at about 25 deaths for every terawatt-hour of electricity it produces and nuclear at 99.8% fewer, Chernobyl and Fukushima included. The figure is true and has calmed few of the frightened, and the benchmark scores that AI labs publish do no better. A statistic asks you to trust whoever did the counting, which is the very thing in doubt.

Britain understood this nearly a century ago. When Londoners objected to a coal power station going up at Battersea, its builders commissioned the architect Sir Giles Gilbert Scott in 1929, with construction already under way, to refashion the exterior. He wrapped it in fine brickwork and shaped the chimneys as fluted columns. The state now protects a coal plant as a listed building.

One of the boards I serve on is Deep Atomic’s. The company is developing the MK60, a compact reactor built to give an AI data centre 60 MW of electric power and 60 MW of cooling. A machine like that will stand close to where people work, and they will judge it by its gate and its control room long before they read a page of its safety case. The Piëch Mark Zero, whose design I shaped, made the same argument at lower stakes. A new carmaker has no record, so the surfaces of its first car must do the work that decades of reputation do for the old names.

The beautiful lie

Beauty also deceives, and Theranos is the standing example. The studio of the Swiss-born industrial designer Yves Béhar designed the casing of its Edison blood-testing machine. Béhar later told Fast Company that his part had been the look of the thing, and that he had seen the cables and tubes inside with no means of telling whether they worked. They did not. Elizabeth Holmes, the founder, was convicted of fraud in January 2022, and the handsome box turned out to be wrapped around an empty promise.

The objection proves the premise. Nobody forges a worthless currency, and swindlers borrow beauty because beauty is what people trust. The Theranos casing was styling, a shell drawn around a mechanism its designer could not vouch for. Design worth the name starts inside, with what the machine does, and makes the surface report it truthfully. By that standard the box in California was ugly, and so was the control panel in Pennsylvania.

The cure for beautiful lies is to let designers into the engine room and hold them to what they find there. After Three Mile Island the American regulator required plants to give their operators, in the control room, a direct indication of whether the relief valves were open or shut. It took a partial meltdown to buy one honest light.

The moulders of the Watertown Arsenal in Massachusetts went on strike over a stopwatch. One of them refused to be timed at his work and was dismissed, and on 11 August 1911 his fellow moulders walked out. It was the year Frederick Taylor published The Principles of Scientific Management. The men in the foundry had a clear idea of whom the watch was working for.

A century later people buy the stopwatch themselves and wear it to bed. Self-optimisation has turned the person into a device with a dashboard, scored each night and reviewed each quarter. It is a category error. I have designed products for Sony, Nespresso and others, and a product has three things a person lacks: a specification, a user and an end of life.

Three things missing

A specification is a target set by somebody else. The machines I helped to design did not choose their own heat-up time. People with a market and a price in mind chose it, and the machine was judged against the result. Optimisation only means something against such a target. A life does not come with one, so the self-optimiser borrows a target from whoever is offering: the default setting of an app, or the morning routine of a stranger with a podcast.

A user is the person the product is for. Ask who the user of an optimised person is, and the candidates are an employer who gets more hours and an audience that gets the progress report. The person doing the optimising is the one party who never takes delivery.

The habit got a flattering name in 2007, when two Wired editors, Gary Wolf and Kevin Kelly, proposed calling it the quantified self. The moulders would have recognised the instrument. At Watertown the employer bought the watch and the worker objected. Now the worker buys it and thanks the app.

End of life is the third. A product is finished on one date and withdrawn on another, and between the two nobody asks it to get better. A person is never released, which suits the upgrade trade very well: a customer who can never be finished is a customer for good.

What self-optimisation borrows

Self-improvement culture talks like an engineering department. It speaks of metrics and iterations, of stacks and protocols. It took the vocabulary and skipped the practice. In 1974 the computer scientist Donald Knuth told programmers to forget about small efficiencies "say about 97% of the time", because "premature optimization is the root of all evil". He was writing for people whose job is speed.

Good engineers optimise very little. They find the one part that limits the whole and improve it. Then they stop, because the point of the exercise is that everything else can be left alone.

They stop for a second reason. Every gain in one figure is paid for in another: make a machine smaller and something inside it has to give. The specification says which trade to make. A person optimising a whole life at once has no such document, so every figure is supposed to rise together. No machine has ever been built that way.

The best case against me is that discipline and measurement plainly do improve lives. They do. A pianist who practises with a metronome plays better, and a founder who reads the accounts every month runs a better firm.

In both cases the measurement is aimed at something outside the person, a sonata or a company, which the person chose. It also stops when the work is done. The metronome stays at home on the night of the concert. What I object to is measurement that has no concert.

Ask less of people

There is a better use for the same energy, and I have done it for a living. For Electrosuisse and the VDE I worked in Japan on a rule that sounds dull: the manual is safety equipment, and has to pass certification with the rest of the product. The idea underneath is blunt. When a person gets a device wrong, the device has failed, and that includes the paper that came with it. The burden of improvement sits with the thing.

Self-improvement culture runs this backwards. If the software is confusing and the office is loud, the person is sold a course in focus. I would rather fix the software and the office. An hour spent tuning oneself helps one person, briefly, while the same hour spent on a tool or a room helps everyone who meets it afterwards.

A good object has absorbed the effort so that its owner is spared it. The work went in once, at the factory, and it does not have to be repeated every morning by everyone who owns one.

Aviation learned this in public. On 30 October 1935 Boeing’s Model 299, the prototype of the B-17 bomber, crashed on take-off from Wright Field in Ohio because a control lock had been left engaged. Some argued that the aircraft was too big to handle. The Air Corps did not go looking for better pilots. It wrote a checklist.

John Maynard Keynes wrote an essay about his grandchildren in 1930, while the world economy collapsed around him. Within a hundred years, he predicted, living standards in the advanced countries would be four to eight times higher, and three-hour shifts or a fifteen-hour week would do. He got the hours wrong and the next part right: once the economic problem was solved, man would face "his real, his permanent problem", which is what to do with the freedom.

The debate about AI and work has not caught up with him. It is stuck on jobs and income: who loses a salary and who pays for it. These are real problems, but they are problems of distribution, and a state that can run a pension system can solve them. The harder question is the meaning of work when production costs almost nothing. My answer is that meaning moves from making to choosing, and that taste becomes the last scarce skill.

Making was never it

I am a designer by training and have made products for Sony, LG, Panasonic, Samsung, Canon and Nespresso. People assume the job is drawing. Drawing is the part a student picks up in a year and a machine now does in seconds. The job is deciding which sketch goes forward and which feature is refused.

Refusing is the harder half. Every refusal is a judgment that something should not exist, made by a person who then has to defend the empty space where it would have been. The finished object shows none of this, which is why the work is so easily mistaken for drawing. A coffee machine with two buttons is the record of an argument somebody won.

Now the making is on tap. A model will produce a logo, a contract, a film score or a page of working code on request, and then a thousand variants of each before lunch. When anything can be had by asking, ability stops being the bottleneck. What matters is whether the thing should exist at all, and which of the thousand is right. A thousand is a worse position than three, as anyone who has stood in front of a wall of paint samples knows.

The meaning of work

This is where the income debate, for all its urgency, stops short. It is the comfortable half of the subject, because income can be modelled and purpose cannot. Pay people to do nothing and you have solved the rent. You have not told them what they are for. Keynes expected trouble: no country, he wrote, could look forward to the age of leisure and abundance "without a dread".

Work made people necessary. Being necessary for output is ending, one trade after another, and no transfer payment replaces it. What remains is being necessary for judgment. Someone has to say this one and put a name to it. A model can draw fifty floor plans for a hospital ward, but it cannot be the person the nurses hold to account.

That is a heavier job than the one it replaces. Labour could be measured in hours and excused by effort. A choice is right or wrong, and it belongs to whoever made it. The designer’s old burden, saying no and answering for it, now falls on everyone else.

Taste on trial

The charge against taste is that it is either snobbery or statistics. If it is snobbery, the preference of people who went to the right schools, it deserves to lose. If it is statistics, a machine can learn it, and one already has. The first public versions of Stable Diffusion were trained in part on images picked by an aesthetic score, and the score came from a predictor taught on people’s ratings of pictures from one to ten.

Both halves mistake what taste is. A score like that is the average of what people have liked so far, and it is as useful as a market survey. It tells you what will please. It says nothing about what should exist next, because the next thing has no ratings yet.

Ask a model which version is best and it will answer. Ask again and it may pick another, and it loses nothing either way. Taste is a choice with a cost attached: a name on the product and money in the tooling.

Nor is it a privilege of class. The baker who will not sell yesterday’s bread has it, and so does the carpenter who sends back warped timber. They learned it by deciding often and living with the results. Free production opens this to more people than ever, since until now you needed a factory to find out whether your judgment was any good.

Watchmaking went through all of this more than half a century ago. In December 1969 Seiko put the first quartz wristwatch on sale in Tokyo, and from that day a mechanical movement was the worse way to tell the time. The mechanical watch survived as an act of pure choice. I helped to found a house in Geneva, Collectif Noir, that makes them in small series with a movement of its own. Quartz is still more accurate, and we are still building the other kind.

Sixty-three millimetres. That is about the distance between the pupils of an adult. Because of that gap each eye receives a slightly different picture, and the brain turns the difference into depth perception. A photograph keeps one of the two pictures and throws the other away.

Artificial intelligence learned to see from photographs. LAION-5B, an open dataset released in 2022 for training image models, holds 5.85 billion captioned pictures, and every one of them is flat. The industry has since moved on to world models, systems meant to grasp physical space well enough to act in it, and it feeds them the same diet. It is teaching machines about a world they have only ever seen with one eye shut. Then it wonders why a robot fumbles a drawer.

What a photograph discards

A camera is an adding machine. Each pixel sums the light that reaches it through the lens and stores a brightness. The direction each ray came from is lost at that moment, and the geometry of the scene goes with it. A toy car close to the lens and a real one across the street can land on the same pixels.

A light field keeps what the camera adds away. In 1996 Marc Levoy and Pat Hanrahan of Stanford gave the term a working definition: the radiance at a point in a given direction. Record how bright the light is and where it comes from, and the picture keeps its depth.

My own research on depth perception and binocular vision led me to light field technology, with the aim of a true record of the world for future generations. The consumer market had other ideas. Lytro, founded in 2006, sold a light field camera whose party trick was refocusing a snapshot after it was taken, and shut down in 2018. The technology was fine. It had the wrong customer: people with two eyes do not need a camera to tell them where things are, and a machine does.

Depth perception needs a hand

Google DeepMind presented Genie 3 in August 2025 as a general purpose world model and said it could give robots a vast space to train in. It renders its worlds at 720p and 24 frames a second. Whatever the model knows inside, what it shows is a flat picture that moves. Video does hold a second viewpoint whenever the camera travels, but nobody wrote down where the camera went, so that has to be guessed as well.

Psychology ran the decisive experiment in 1963. Richard Held and Alan Hein put ten pairs of kittens in a carousel. One kitten of each pair walked, and the other rode in a gondola that the first one pulled, so both saw the same striped walls for the same three hours a day. The walkers learned to place their paws by sight and chose the shallow side of a visual cliff every time. The riders saw exactly what the walkers saw, and it did them little good.

Seeing, on that evidence, is something an animal does, and watching is a poor substitute. A child gets the full kit for free: two viewpoints and a hand. The eyes tell the hand where to go, and the hand tells the eyes whether they were right. Robot makers know this and bolt depth sensors onto their machines. The models that are supposed to supply the understanding were still raised on flat pictures.

One picture is enough

The reply from the software side is that the missing dimension can be computed back. Monocular depth estimation has become very good. Depth Anything V2, published in 2024, draws a convincing depth map from a single image, and Apple’s Depth Pro returns one with absolute distances in 0.3 seconds. If one eye is enough for a model, the flat archive is no handicap.

Look at how those models were taught. Depth Anything V2 learned from 595,000 synthetic images, rendered from three-dimensional scenes in which the depth of every pixel is known, and its teacher model then labelled 62 million real photographs for the students. The third dimension was in the room all along. Somebody had to supply it.

What the model adds on its own is memory. A single image is ambiguous as a matter of geometry, so the model fills the gap with what doors and chairs usually measure. That works for things it has met before and for pictures meant to be looked at. A photograph of a corridor pinned to a wall sends one fixed eye the same picture as the corridor. A person with one eye settles the matter by moving the head and reaching out, which is my argument in two gestures.

A robot’s work happens in the last few centimetres, where a handle stands proud of a drawer front and fingers have to close on it. A guess that makes a handsome depth map is a poor guide there. Held and Hein’s riders saw every stripe the walkers saw. On the visual cliff they could not tell the deep side from the shallow.

Miles Davis took his band into a former church on East 30th Street in New York with almost no rehearsal behind it. By the account of the pianist Bill Evans, who wrote the sleeve notes, Davis had conceived the pieces only hours earlier and arrived with sketches: some scales, some melody lines. So What, the first track, is built on two scales, and John Coltrane did what he liked with them. That session, on 2 March 1959, and a second one in April produced Kind of Blue, since certified five times platinum in America.

Nobody in that room wondered who was in charge. Davis had picked the players and brought the sketches, and inside those few lines each man played as he saw fit. That is my picture of modern leadership: one person decides, everyone knows who, and inside that clarity each person runs their own work without asking. The fashionable alternative is the flat organisation, where nobody is in charge and everything is a meeting. What it delivers is a hidden hierarchy and endless process.

The flat organisation myth

The flat organisation is sold as the cure for bad bosses. Remove the boss, the theory runs, and people will be free. What happens instead was described in 1970 by Jo Freeman, an American feminist and political scientist, in a talk on the leaderless groups of the women’s movement, later published as The Tyranny of Structurelessness. Her finding was that "there is no such thing as a structureless group". Power stays where it was and moves out of sight, and "the rules of how decisions are made are known only to a few".

Business has since repeated the experiment. In 2012 the staff handbook of Valve, an American games company, became public, declaring that the firm had no management and that nobody "reports to" anybody else. A year later Jeri Ellsworth, a hardware developer whom Valve had let go, gave her side in a podcast interview. By her account there was "a hidden layer of powerful management structure" and the place "felt a lot like High School". That is one former employee’s account, and it is what Freeman would have predicted.

A flat company still has a hierarchy and declines to print it. Because nobody may decide, everything must be agreed, which takes a meeting, and then a second meeting to establish what the first one agreed. The people who thrive in this are the ones who enjoy meetings. They are seldom the ones doing the work.

Small teams, short lines

I know both sizes from the inside. As an industrial designer I worked for some of the largest manufacturers on earth, Sony and Nespresso among them. As a founder I build small companies, on purpose.

Smaller and better beats bigger, and the reason is arithmetic. Fred Brooks set it out in The Mythical Man-Month in 1975: the lines of communication in a group of n people number n(n-1)/2. Five people have ten lines between them. Fifty have 1,225.

Every one of those lines is a chance for a message to arrive late or altered. So large organisations hire people to carry messages, and then write processes to check the carriers. The person who knows should talk to the person who decides, with nobody in between to tidy things up. A small team can do that across a table, which is why small teams ship while large ones are still booking the room.

Bureaucracy is what an organisation builds once it has stopped trusting its people. Each form and each approval step records some old mistake, usually one person’s, now charged to everybody. The careless shrug it off. The able, who never needed the rule, pay for it every day in hours and patience, and they are the first to leave.

What modern leadership owes

People want responsibility and the freedom to decide how they meet it, and they want their work to be valued. Fulfilment is what those add up to. A clear hierarchy provides them, because it settles the one question that otherwise eats the day: who decides. With that answered, leadership and autonomy stop being rivals. The leader says what the job is and why it matters, then gets out of the way.

The leader’s own devotion belongs to the cause. The organisation is a tool for serving it, and so is the leader. This also settles what appreciation means. A perk programme is what companies offer in its place. Appreciation is being handed something that matters and trusted to get it right.

A clear hierarchy invites abuse of power. That is the serious case against it, and it is true. It is equally true of every arrangement that contains power, and the flat one hides the abuser as well. A named boss can be contradicted in the open and can be removed. A clique cannot, because officially it does not exist.

Freeman herself asked for specific authority to be given to named people, and for those people to be held to account. A boss who serves himself instead of the cause has given up the only claim to command he had. Smallness does the rest. In a company of twelve, a tyrant runs out of staff quickly.

In 1980 an engineer at ETA, the Swiss maker of watch movements, ordered an injection moulding machine that cost 500,000 Swiss francs. He was 26 and had no authority to order it. The firm, by his account, had shed 4,000 jobs in a year. Summoned to the general manager’s office for one o’clock, he had two hours to invent a reason for the purchase, so he went to a colleague, Jacques Müller, who put a plastic watch on paper.

The engineer was Elmar Mock, and the drawing became the Swatch: 51 parts where a conventional watch had 91, launched in Zürich on 1 March 1983, more than 700m sold. I have had the pleasure of working with Mock on several projects, and I know no better account of how innovation works than his. His image comes from chemistry: innovation passes through three states and starts as a gas. The conclusions are mine, and he is not to blame for them: most firms that ask for innovation are solids that intend to stay solid, and now that machines produce gas for nothing, the skill worth paying for is condensation.

Gas, liquid, crystal

The image is set out in the book Mock wrote with Gilles Garel, published in English as The Innovation Factory. There the three states are gas, liquid and crystal, and they are states of mind. The title of this piece is my own shorthand for the first state.

Creaholic, the invention firm Mock founded in Biel in 1986, says gas, liquid, solid. Marcel Aeschlimann, a co-founder, describes the gas state as one where the molecules are free and ideas can be thought up from scratch. His example is the ejection seat, where the chair molecule meets the dynamite molecule. In the liquid the idea condenses into a concept that could work, and in the solid the thing gets made. I use the firm’s word from here on, since nobody has mistaken a corporation for a crystal.

Mock also separates innovation from renovation. A company must listen to its clients and give them what they ask for, he told a Paris seminar in 2016, but that is renovation. He added that he had no wish to criticise it, since it is what keeps everyone fed.

One section of the book’s chapter on the Swatch is headed, in part, "Perfect Peace for Two Dissidents". Mock, by his own account, needed eleven years for nine years of compulsory school.

Renovation with better lighting

Mock is polite about renovation. I am less so about the firms that sell it under another name. Most companies that announce an innovation programme want version 1.3 in a new colour, presented on a stage. Renovation is a respectable business. Calling it innovation is theatre, and nobody books a stage for something that might fail.

The processes such firms install to manage innovation are the solid state defending itself. A stage gate asks a gas for a business case, and an idea funnel asks it for a sales forecast. Gas has neither, so it fails the review, and the firm concludes that nobody had a good idea. Mock’s own verdict on planning is three words long: "Planning kills innovation."

Large companies do innovate, their defenders say, and the Swatch came out of one. It did, and the manner matters. It began with a purchase order nobody had approved and a sketch that Mock calls childlike. Ernst Thomke, the head of ETA, kept asking him whether he was crazy, and then backed the watch. Mock told the Paris seminar that it was developed separately and under protection, with outside help, and met enormous opposition.

Nestlé behaved the same way with coffee. Its management doubted Eric Favre’s capsule system for about a decade, then in 1986 set up Nespresso as a company of its own. The book counts the protection of a mentor among the lessons of the Swatch. I put it more bluntly: large firms innovate when someone senior hides the gas from the solid, and no process chart contains that person.

Gas is now free

Here the argument becomes mine alone: Mock’s model of how innovation works dates from a time when a good idea was hard to come by, and that time is over. Ask a language model for a hundred concepts for a wristwatch and it delivers them before the coffee is cold, every one plausible and none tested. I founded a lab that works on such models, and ideas are the cheapest thing they make. A company with unlimited gas has not bought breakthrough innovation. It has bought fog.

What stays scarce is condensation, and the Swatch shows what it costs. Thomke gave Mock and Müller six months. The prototype took fifteen. An engineering mistake made the hands turn backwards. By late 1981 there were five finished watches, and they stopped after five days.

The design that survived built the movement straight into the plastic case by ultrasonic welding. Mock has told the Guardian why: the team decided that every screw was a danger, because whatever can be screwed in can come unscrewed. Nothing in a list of a hundred concepts tells you that.

A designer’s working life is spent at that phase change. The machines I worked on for Nespresso began with an idea a child can state: coffee from a capsule. My work on user manuals for Electrosuisse and the VDE began with another: people should understand the device in front of them. The value came after the sentence, in the wall thickness a mould will fill and the wording a certifier will sign.

Müller’s sketch was ready by one o’clock. The launch in Zürich was three years away.

For most of the history of consumer products, the manual was written for a person who would never read it. It went into a drawer. Then something changed. The manual became part of product safety. If the instructions are wrong, the product is not safe, even when the machine itself is fine.

I spent years on that change. The lesson was simple. Information is part of the product. A dishwasher is the steel, the pump and the software, and also what you need to know to use it without breaking it or hurting yourself.

A new reader

Now the manual is getting a new reader, the robot. A robot that loads a dishwasher, opens an oven or closes a drawer has the same problem a person has on day one. It has to understand the object before it touches it. Where is the handle. How far does the door open. How much force is too much.

Today robots mostly find this out by trying. They explore, fail, try again. That is slow and expensive, and in a home it can break things. Every robot repeats the same discovery for the same object.

Objects that explain themselves

The better way is for objects to explain themselves, as data a machine can use. Dimensions, joints, forces, limits, what happens when you push here. Measured once, checked, and shared.

That is the work I care about now. It sits between design, standards and AI. Designers know how products are meant to be used. Standards bodies know how to make information trustworthy. AI needs both.

What changes for designers

If robots become users, design changes too. A handle is no longer only for a hand. A button needs to be readable by a camera. An appliance that a robot can operate safely will sell better than one it cannot.

We went through this once with people. The manual became part of the product. It is about to happen again, with a different reader.