On November 19, 1955, The Economist ran an essay by a correspondent with a table built from Admiralty statistics. It compared the Royal Navy of 1914 with the Royal Navy of 1928.
Capital ships in commission: 62, then 20. Officers and men: 146,000, then 100,000. Admiralty officials in London: 2,000, then 3,569.
The fleet had lost two-thirds of its big ships. The office that ran it had grown by 78.45 percent. Critics at the time had mocked “a magnificent Navy on land.” The author, the historian C. Northcote Parkinson, went further, in a deadpan scientific voice: “The officials would have multiplied at the same rate had there been no actual seamen at all.” He offered two forces to explain it. “An official wants to multiply subordinates, not rivals,” and “officials make work for each other.”
The essay was satire, written in a mock-scientific voice. But it put numbers to a familiar complaint about bureaucracy. The case has the rest, and Parkinson’s Law has its own page.
What Weber meant
The German sociologist Max Weber had described bureaucracy a generation earlier, in Economy and Society, first published in 1922. He meant something more specific, and he thought it was the most effective way yet found to run large affairs.
A bureaucracy, in his description, has fixed areas of authority set by rules. It has a hierarchy of offices, with a set route for appealing a lower office’s decision to a higher one. Its work runs on written documents, the files. And it separates the office from the home: official money from private money, official business from private life. Decisions are made “without regard to persons,” according to “calculable rules.” (The translations here are mine.)
Why did it spread? “The decisive reason for the advance of bureaucratic organization has always been its purely technical superiority over every other form,” Weber wrote. A fully developed bureaucracy compared with older ways of running things “exactly as a machine does with non-mechanical modes of production.” He listed what it raised to the optimum: “precision, speed, unambiguity, knowledge of the files, continuity, confidentiality, unity, strict subordination, reduction of friction and of material and personal costs.”
That is the strength in the plain version above. When the rules are written down, the work doesn’t depend on who is on duty, or on whether they like you.
The price Weber saw
Weber was not naive about it. In a parenthesis he noted that the bureaucratic apparatus “can, and in fact does,” get in the way of handling an individual case on its merits, then added that this was not a matter to go into there. The coldness and the strength came from the same place. Bureaucracy’s particular character, which Weber called welcome to capitalism, develops “the more perfectly, the more it is ‘dehumanized,’” he wrote: the more it succeeds in removing “love, hatred and all purely personal” feeling from official business. That, he added, is exactly the quality praised as its virtue. He also called the “official secret” bureaucracy’s “specific invention,” defended “with fanaticism.”
Parkinson’s Admiralty, in his satirical account, shows another cost: an office can keep growing for its own reasons, whatever happens to the work. Zhang Ruimin, who led Haier, named Weber’s bureaucracy, the pyramid, as one of the two foundations of traditional management that the internet would overturn, because customers now knew more than the company. Haier’s answer was to break itself into thousands of small units, at a cost of its own. The case describes it.
The dental practice
Imagine a small dental practice with three receptionists. Each handles late cancellations differently: one always charges the fee, one never does, one decides by mood. Patients compare notes, and the arguments start. The practice writes a rule: under 24 hours’ notice, the fee applies, unless the patient has a medical emergency. Now any receptionist gives the same answer, and anyone new can learn it in a minute. Then a patient calls an hour before, from a hospital waiting room with a sick child. The rule covers the patient’s own emergency, not a child’s, so it says charge. A receptionist with no room to judge might charge, and the practice might lose a patient of ten years.
Rules for an AI
Anyone writing instructions for an AI model faces Weber’s trade in a new form. Anthropic laid it out in the constitution it wrote for Claude. Clear rules, it said, “offer more up-front transparency and predictability, they make violations easier to identify, they don’t rely on trusting the good sense of the person following them, and they make it harder to manipulate the model into behaving badly.” But “rules often fail to anticipate every situation and can lead to poor outcomes when followed rigidly in circumstances where they don’t actually serve their goal.”
Anthropic said it generally favors judgment over strict rules, while keeping a small set of firm limits and explaining the rules it does set. Rules make the most sense, it wrote, when errors would be costly enough that predictability becomes critical, when individual judgment may not be robust enough, or when firm commitments are needed to resist manipulation.
It also gave an example of a rule backfiring. Teach a model always to recommend professional help on emotional topics, even when that doesn’t serve the person, and it risks learning that it is “the kind of entity that cares more about covering myself than meeting the needs of the person in front of me.” Weber might have recognized that official.
He built his account around a machine. Parkinson’s joke was that the Admiralty’s machine kept running, and growing, after much of the navy it served had gone.