The last bureaucrat

“History doesn’t repeat, but it rhymes”, said Mark Twain. I like when it salsa dances. When, for a moment, two paradigms overlap and coexist. Sail ships shared the sea with steam engines, and the Polish army famously faced German panzers with cavalry in 1939.

I think we are living through one of those seams now. In a hundred years, children may find it strange that intelligence had become widely available, and yet most of the important decisions about how people lived together were still mediated by bureaucracies.

Not because bureaucrats were useless. For a long time they were useful. Bureaucracy was one of the best answers to a world in which intelligence, literacy, coordination, and administrative capacity were scarce. That world has ended, and we get to build what comes next.

So who will the last bureaucrat be?

1/ why bureaucracy was born

Bureaucracy is not a modern invention. The Egyptians had scribes, the Romans had administrators. Any society that grows beyond the scale of personal relationships needs some way to remember, decide, and enforce.

It is the institutional form of centralization, and centralization almost always produces it. Once decisions are made away from where their consequences are felt, you need records, rules, offices, procedures, reports, approvals, and people whose job is to keep the machine coherent.

The modern version grew with the nation-state and the industrial age, and its logic was industrial: design one process, then scale it. This is how factories work, and over time it became how states worked. A country needed to collect taxes, educate children, build infrastructure, raise armies, issue documents, administer justice. The innovation was the scalable process, standardized and deployed everywhere. Napoleon’s France and Bismarck’s Germany show how deeply administration, centralization, military organization, and state-building became intertwined. The modern state needed trained officials the way an army needed trained soldiers.

It is worth being fair about what this bought. Before the Code civil, France ran on hundreds of local legal customs. After it, on one. A uniform and clumsy rule beat a thousand finely adjusted ones, because a farmer in Provence and a merchant in Lille could finally expect the same answer to the same question.

For a world of scarce intelligence, this made sense. If only a small number of people could reliably read, calculate, plan, or administer complex systems, the rational thing was to gather them near the center, have them design the rules, and send bureaucrats throughout the territory to apply them.

It was practical. Centralization let societies coordinate at a scale that would otherwise have been impossible. Bureaucracy is easy to mock now, but it was a miracle of human coordination. It answered a specific constraint: intelligence was scarce, information was slow, and coordination was expensive.

2/ what it cost us

Even before AI, the weaknesses were obvious to anyone who had to live inside one.

Start with distance. The person designing a rule is rarely the person living with its consequences. The person enforcing the process has little skin in the game. The person on the receiving end arrives as a case, a form, a number, a status, a missing document.

This is what makes bureaucracy dehumanizing on both sides of the counter. The citizen becomes a file, the official becomes a function, and the interaction is absurd even when everybody means well. A teacher in the public system follows guidelines written in a ministry, when the whole magic of teaching is catching one particular child’s attention on one particular morning. Nobody has ever managed to write that down.

Kafka understood this better than almost anyone. In The Castle, the cruelty is vague, procedural, impossible to locate. K. never manages to find anyone to argue with. There is no tyrant in the castle, and that is the horror of it rather than a detail of the plot. Bureaucracy is the one form of power that can crush a person without anybody deciding to.

Responsibility is the deepest flaw. Route a decision through procedures, committees, departments and rules, and moral agency dilutes until “I was following procedure” becomes an acceptable sentence. At its worst, this has let bureaucracies take part in the worst acts in human history, precisely because the system separates action from conscience.

Most bureaucracy is not evil, of course. Usually it is just slow, and slowness became a much bigger problem in the software age. A startup changes its product after watching ten users struggle with it. A bureaucracy can watch millions struggle with a process and preserve the process, because the process has become the point.

So the temptation is obvious. If the issue is speed and cost, use AI to process the forms, answer the questions, allocate the resources, detect the fraud, enforce the rules, and decide.

3/ the worst bureaucracy is a perfect one

The simplest AI reform is also the most dangerous one. Take the existing centralized bureaucracy and make it faster. Keep the hierarchy, the procedures, the command structure, the assumption that judgment flows from the center outward. Then insert AI into the machine. That is the premise behind most attempts to modernize government, and, in a different key, the instinct behind DOGE.

There is something true in the diagnosis. Many bureaucracies are too slow and too expensive, and an AI bureaucracy would be faster: more cases processed, more inconsistencies caught, less money wasted.

Bureaucracy remains a centralizing force, and automation does nothing about the problems that come with centralization. It makes the distance smoother, less visible, and harder to contest. A human bureaucrat can be tired, inconsistent, merciful, inattentive, or simply reachable. A machine applies the system perfectly. A broken bureaucracy is frustrating. A perfect bureaucracy would be frightening.

The twentieth century already showed what happens when a center tries to plan a society from above. Its weakness was informational as much as moral: it could not see enough, process enough, or adapt fast enough.

They tried anyway. The Gosplan balanced a few thousand categories of goods for an economy that ran on millions. Kantorovich won a Nobel in 1975 for the mathematics of allocating them. Glushkov proposed OGAS, a national computer network built to compute the plan, and it was never built. They failed for want of computing power, and for want of politics. Only one of those two constraints has gone away.

So imagine a command economy with AGI. Imagine a surveillance state with unlimited administrative capacity. Imagine every rule enforced instantly, every exception modeled, every deviation flagged, every appeal routed through one more automated layer. Imagine the USSR with AGI.

Imagine, China.

This is why “replace bureaucrats with AI” is the wrong ambition. AI lifts exactly the constraint that used to save us: it gives a center more sight, more memory, more prediction, more reach. That makes it more competent, and far more dangerous when it is wrong, captured, or merely too confident in its own model of the world.

The West should answer AI with something better than a more efficient administrative machine. The United States did not beat the Soviet Union by becoming a better planned economy. It won by letting more intelligence, initiative, experimentation, and failure live at the edges.

the other road

If intelligence is becoming abundant, we should build systems that use it wherever it appears instead of gathering it in the capital. That argument deserves its own essay, and I made it in Intelligence at every node.

What matters here is that this is buildable now, and that it will be built at the edges, by founders, mayors, teachers, engineers, and civil servants who stop asking permission from a center that no longer knows more than they do.

The end of scarce intelligence gives judgment back to the people who spent a century being asked to suspend it. The teacher gets to teach the child in front of her. The clerk gets to be someone who decides something, and who can be argued with.

Which brings us back to the question. There are two ways to have a last bureaucrat, and both of them end the story.

In the first, the last bureaucrat is a machine. It never tires, never doubts, never goes home, and there is nobody behind it to argue with. The office finally works, and it never closes again.

In the second, the last bureaucrat is a person who walks out one evening because there is nothing left for the office to do.

The second one is what I want to help build.

Vic