A friend of mine started a company in a single weekend last spring. That is not true, but it is close. He built a website, a payment system, an application, and a series of automated emails that together created the entirely convincing impression that a business existed somewhere. There was a logo. There was a customer portal. There was a help desk that answered, politely, questions no one had yet asked. On Sunday evening he sent me a link. On Monday morning he applied for a business bank account.
That was where the waiting began.
A row of empty chairs in a roomOpen the photograph →
The bank wanted to know who his customers were. He would have liked to know that himself. Then came a form about projected revenue, the origin of his capital, and the countries he intended to operate in. Some of the questions made sense. Others had the majestic pointlessness of questions added after an incident and never removed since. Because who ever removes a question?
The attachment was too large. The signature was on the wrong page. Someone would be in touch. When she was, three weeks later, the software was four versions further along. She spoke slowly and carefully, as though haste were a form of recklessness. Was it a webshop, she wanted to know. Not exactly, he said. A platform, then? Not exactly either. There was a pause in which you could almost hear an administrative category trying to close itself around a new phenomenon.
It took five weeks in the end. By then my friend had abandoned the original idea. While he waited, he had used the same tools to build something else he liked better. The account was opened in the name of a company that, in a sense, no longer existed.
His neighbour lives in a different century. She works at a mid-sized institution that last year forbade its staff to use generative language models, with the exception of a handful of participants in a pilot. The pilot runs for nine months and has a steering group, a risk assessment, and a communications plan. Participants may use the system for summaries, but not for analysis, decisions, client contact, or any text that leaves the building.
Row of traditional terraced housesOpen the photograph →
She and my friend live on the same street. Their bins go out on the same morning. Their children attend the same school. Professionally, they are an industrial revolution apart.
It is tempting to explain the difference as a matter of temperament. One of them is curious, the other is careful. But something larger sits underneath. One of them has been granted a new speed of action. The institutions the other depends on were designed for the old one.
Productivity is too small a word for that shift. It suggests more boxes an hour and shorter meetings. What is actually changing is the distance between a thought and a deed. Until recently that distance was a strip of no man's land, filled with specialists, budgets, schedules and handovers. An idea could be excellent and sit unbuilt for years, simply because the world pushed back.
That resistance had a name. We called it technology.
Anyone who has tried to change something inside a large organization knows the wall. The technology could not do it yet. The team had no capacity. The architecture was too complex. The solution was scheduled for next year. Years ago I worked at a financial institution where the waiting list for a simple change or addition to the software ran between eighteen and thirty months. That list was not a list. It was a geological record of organizational longing. At the top lay the projects some forceful manager or executive had declared urgent. Below them, the improvements everyone found useful but nobody found useful enough. At the bottom lay the fossils: ideas whose author had long since changed jobs or left the building.
Developers were recruited without pause, many of them from abroad. Teams were added, suppliers contracted, acceleration programmes launched. And still the waiting time did not move. Every increase in capacity summoned new demand, the way a wider motorway attracts new traffic.
Close-up of striated rock layersOpen the photograph →
When something unexpectedly had to happen quickly, there was no procedure for it. There was an important person for it. An executive could pull a project forward, but the decision did not resolve the scarcity. It relocated it. Somewhere else a team was dismantled, a promise quietly revised. Urgency was not a property of the work. It was a ranking of influence.
At the time this seemed inevitable, and whole organizations were built around the assumption. Strategy was about choosing what you would not do. Portfolio management was, in essence, the art of disappointing people with spreadsheets.
And as long as building a system took a year, a month spent waiting for approval vanished effortlessly into the shadow of it. The technical delay lay over the landscape like a thick fog, and in that fog the forms, the committees, the deferred decisions and the unclear responsibilities could carry on undisturbed. Nobody knew exactly how much time they cost, and nobody had any interest in finding out.
Technology, then, was not merely a problem. It was the problem behind which all the other problems could shelter. It was the last excuse available to an organization that did not know where its own slowness came from.
Now the fog is lifting.
Building software has become so fast and so cheap in so many places that the delay around it no longer disappears into the whole. Work that used to take months now yields a surprisingly solid first version within days. This is not because everyone has suddenly become a programmer. It is because a growing share of technical craftsmanship has been handed a toolbox ten times the size.
Aerial view of a winding road through forest with fog rolling across itOpen the photograph →
That does not mean technology has stopped being a problem. Large systems remain complicated, and a prototype is not yet a reliable product. But in more and more cases the technical obstacle is no longer the heaviest one, and sometimes it can be cleared with a few instructions. And then something else comes into view.
Anyone building something today discovers that the computer is rarely the one keeping them waiting. The waiting starts afterwards. For a supplier who has to grant access. For a contract. For a meeting three weeks out. For an executive who formally owns a decision but stands too far from the substance to dare to take it.
A considerable part of what we used to call technical slowness turns out to have been organizational slowness, dressed up as technology.
That is a less comfortable discovery. A technical constraint can be outsourced, funded, or eventually solved. A human system cannot. It consists of habits, interests, fear, experience, status and carefully distributed responsibility. It has no error message. It never admits that it has stalled. It only asks for another meeting.
The history of the electric motor offers the sharpest comparison. When electric drive became available at the end of the nineteenth century, factory owners did the obvious thing. They took out the steam engine and put an electric motor in its place. The rest of the factory stayed where it was. The machines were arranged around a central drive shaft and drew their motion from it through leather belts and pulleys. The power source had been replaced, and modernization was considered complete.
The real gain arrived decades later, when a new generation of manufacturers understood that electricity was not merely a better power source but something else entirely. A motor did not have to stand in the middle of the factory. Every machine could have one of its own. The floor could be arranged around the sequence of the work instead of the demands of the power supply. In 1900, steam still supplied eighty percent of the mechanical drive capacity in American manufacturing. By 1929, seventy-eight percent came from electricity. And only in the nineteen-twenties did industrial labour productivity truly turn.
Technology had not simply replaced a machine. It had made visible that the factory itself had been designed around a constraint that no longer existed.
Vintage industrial machinery with pipes and gearsOpen the photograph →
That is where our organizations stand now. They add artificial intelligence to existing processes and leave the shape of those processes untouched. Reports are written faster, analyses produced faster, software shipped faster. And decisions travel exactly the road they always travelled. The new tooling runs in a factory still built around a central shaft.
At first this produces gains. Then congestion sets in. Executives who used to complain about a lack of information now drown in impeccably worded memos. Lawyers are handed products that already exist technically, before anyone has established which category they fall into. Security teams face months of work on discovered vulnerabilities and are expected to close them within days. The queue for development disappears and returns as a queue for decisions.
My friend's neighbour told me about a proposal that had taken months. Not the implementation. The proposal itself. Conversations were held, risks inventoried, stakeholders consulted, presentations made. By the time the document reached the agenda, a colleague had already put together a working version of the idea at home. He showed it during the meeting.
That was not the intention. The meeting had been convened to decide whether an exploration should be undertaken, and the existence of a working example made half the questions redundant and the other half urgent. The response was not enthusiasm but procedural concern. Who had given permission? Which data had been used? Where was it running? Who owned the code?
Four reasonable questions. Together, though, they exposed something uncomfortable. The institution had built an entire apparatus to determine which ideas were allowed to become real, and technology had enabled someone to reverse the order. Reality first, permission afterwards.
Aerial view of open land crossed by made roads and worn tracksOpen the photograph →
The Dutch have a good word for what that colleague did. An olifantenpaadje, an elephant path, is the track that walkers and cyclists wear into the grass over time because the route that was laid out turns out to be a detour. English calls it a desire path. It is not an act of defiance but an act of efficiency, and it shows precisely where reality has overtaken the design. Such a path is not a problem in need of asphalt. It is information.
And information is exactly what organizations are now receiving about themselves, in quantity. Those procedures are not badly designed. They are barely designed at all. Behind most of them sits no considered decision model but a collection of historical compromises. A committee born of distrust between two departments. A manager who keeps the right to decide because nobody wants the fight it would take to move it. A form that survives because the person who once needed it has left, and nobody knows what happens if it goes. This is how institutions come about. Not as machines but as cities, growing century after century around their own streets.
As long as technology caused the greatest delay, that city could function. Now that the traffic moves faster, the junctions turn out to be in the wrong places.
It would be too easy to conclude that the fastest organizations simply win. This new speed has an ignorance of its own. My friend uses code he does not fully understand, infrastructure he cannot oversee, and legal texts of which he mainly knows that they sound convincing. His pace is real. Part of the certainty it rests on is borrowed.
And the questions his neighbour's organization asks are not invented. Who is liable? What happens to the data? What if something that works beautifully for one employee is used tomorrow by ten thousand people? The old world is not slow out of sheer incompetence. It carries the scars of earlier mistakes. Almost every irritating rule began as an answer to something that went wrong. The structure we now feel locked inside is one we put up ourselves, plate by plate, each time after a new lesson.
The aim, then, cannot be to abolish care. It has to be to design a form of care that no longer depends on slowness. That calls for decisions taken closer to the work. For small, reversible experiments that do not have to follow the same route as large, irreversible ones. For controls built into the tooling instead of queued at the end. And for people who know in advance the boundaries within which they may act on their own.
Trust becomes a form of infrastructure. That sounds softer than computing power, data or capital, but it is probably the scarcer commodity. When employees can build far more on their own, an organization has to decide how much freedom it actually grants them. Not in glossy posters about entrepreneurship, but in access rights, budgets, liability, and the right to make mistakes.
That forces an unusual kind of self-examination. The question is not which new AI tool to buy. The question is why a decision here takes an average of twenty-two days. What surfaces are questions of behaviour and of power, and those are rarely answered by a nine-month pilot.
The neighbour, meanwhile, received permission to use the system for rewriting internal emails, provided she checked every text for confidential information. Her organization called this an important next step.
It would be easy to make a hero of my friend and a laggard of his neighbour, but they are less each other's opposites than two subjects in the same experiment. He is finding out what happens when technology moves faster than institutions. She is finding out what happens when institutions try to force a new technology into old categories.
A strip of photographic negativeOpen the photograph →
The promise of artificial intelligence, then, may not be that it solves our problems. It may be that it removes the problem we were able to hide behind. It works like a photograph in a developing bath. What was present in the negative all along slowly appears. What we get to see is not an orderly landscape but the roads we laid ourselves. Some of them are necessary, others are redundant, most are more winding than they need to be.
That is a disenchanting picture, and potentially a very usable treasure map. If the past century was about building better and better technology, the coming decades could be about building better institutions. Not faster for the sake of speed, but faster where it matters. Not less careful, but more careful in the places where care actually counts. Not because technology forces us to, but because for the first time it leaves us the room.
The excuse of hiding behind technology is gone. That is bad news, and probably the best thing that could have happened to us.
◆