The Roman poet Juvenal asked the question two thousand years ago: Quis custodiet ipsos custodes? Who will guard the guards themselves? He was writing about the impossibility of ensuring marital fidelity, but the phrase has become the defining question of governance. Every state needs administrators. Every set of administrators eventually begins to serve its own interests rather than those of the people it was created to serve. Every civilisation that has failed to solve this problem has declined. None has solved it permanently.

In January 2025, the United States created the Department of Government Efficiency — DOGE — under the direction of Elon Musk. Its stated goal was to identify $2 trillion in federal spending cuts. By early 2026, it had achieved approximately $215 billion in estimated savings through contract cancellations, fraud detection, workforce reductions, and the elimination of grants and programmes deemed wasteful. Musk later acknowledged that $2 trillion was a “best-case outcome” and that $1 trillion was more realistic. Even the lower figure would represent the largest peacetime reduction in government spending in American history.

The reaction was predictable. Federal employee unions fought every cut. Judges issued injunctions. Journalists wrote stories about the human cost of layoffs. Politicians who had spent decades expanding government programmes accused DOGE of recklessness. The bureaucracy, in short, defended itself — with every tool at its disposal.

This is not new. It is the oldest story in the history of governance. And understanding it requires going back much further than 2025.

Part 1: Rome — From Lean Republic to Administrative Empire

The Roman Republic was, by ancient standards, remarkably lean. The Senate and the magistrates governed an expanding territory with a minimal permanent bureaucracy. Provincial governors were appointed for fixed terms. Tax collection was outsourced to private contractors (publicani). The army was a citizen militia, called up for specific campaigns and disbanded afterwards. The entire system was designed to prevent any individual or institution from accumulating permanent power.

This changed under the Empire. Augustus, the first emperor, created a professional civil service to administer the vast territories that the Republic’s ad hoc system could no longer manage. This was a rational response to a real problem: you cannot govern an empire stretching from Britain to Mesopotamia with part-time magistrates and private tax farmers. You need permanent officials, standardised procedures, and institutional continuity.

But the bureaucracy grew. And grew. By the time of Diocletian (284–305 AD), the Roman administrative apparatus had become enormous. Diocletian divided the empire into over 100 provinces (up from about 50 under Augustus), each with its own governor, staff, and administrative hierarchy. He created a vast network of state factories (fabricae) to supply the army, a system of price controls (the Edict on Maximum Prices), and a tax system so complex that it required an army of assessors, collectors, and auditors to operate.

The historian A.H.M. Jones estimated that the late Roman bureaucracy employed tens of thousands of officials, supported by an even larger number of clerks, scribes, and enforcers. The cost of maintaining this apparatus was staggering. Lactantius, writing in the early fourth century, complained that “the number of recipients began to exceed the number of contributors” — that there were more people living off the state than paying into it.

The bureaucracy did not just consume resources. It actively resisted reform. Officials who owed their positions, their income, and their social status to the existing system had every incentive to preserve it and no incentive to streamline it. When emperors attempted to reduce the administrative burden, they found that the bureaucracy simply absorbed the reforms and continued expanding. New positions were created to oversee the elimination of old positions. Commissions to reduce complexity produced additional layers of complexity.

The Roman bureaucracy did not cause the fall of the Western Empire. But it contributed to the sclerosis that made the empire unable to respond effectively to the crises of the fifth century. When the Visigoths, Vandals, and other Germanic peoples pressed against the frontiers, the empire’s response was slow, uncoordinated, and often counterproductive — not because it lacked resources, but because the administrative machinery that was supposed to deploy those resources had become an end in itself.

Part 2: The Ottomans and the Janissary Problem

The Ottoman Empire offers an even more dramatic illustration of the same pattern.

In the fourteenth century, the Ottomans created the devshirme system — a practice of taking Christian boys from the Balkans, converting them to Islam, and training them as elite soldiers (Janissaries) and administrators. The system was designed to create a loyal, meritocratic governing class with no ties to the existing Ottoman aristocracy. The boys owed everything to the Sultan and nothing to any family, tribe, or faction. It was, in its way, a brilliant solution to the problem of bureaucratic capture.

For two centuries, it worked extraordinarily well. The Janissaries were the most formidable military force in Europe and the Near East. Ottoman administrators were competent, loyal, and efficient. The empire expanded from a small Anatolian principality to a superpower controlling southeastern Europe, the Middle East, and North Africa.

But the Janissaries, like all bureaucratic classes, eventually began to serve their own interests. By the seventeenth century, they had secured the right to marry, to pass their positions to their sons, and to engage in private commerce. They had become, in effect, a hereditary caste — exactly what the devshirme system had been designed to prevent. They used their military power to veto reforms, depose sultans who threatened their privileges, and block the adoption of new military technologies (particularly firearms and European-style drill) that would have reduced their importance.

By the eighteenth century, the Janissaries were a parasitic institution. They consumed a large share of the state’s revenue, provided increasingly poor military performance, and violently resisted every attempt at modernisation. Sultan Selim III attempted to create a new, modern army in the 1790s. The Janissaries murdered him. His successor, Mahmud II, spent sixteen years carefully preparing before he finally disbanded the Janissaries by force in 1826, in what became known as the “Auspicious Incident.” Thousands of Janissaries were killed. Their barracks were destroyed. Their very name was erased from official records.

The parallel with modern bureaucracies is uncomfortable but instructive. The Janissaries were created to serve the state. They ended up holding the state hostage. The only way to remove them was through an act of deliberate, planned, and violent destruction by the sovereign power. No reform from within was possible, because the Janissaries controlled the mechanisms of reform.

Part 3: China’s Examination System — Meritocracy Ossified

The Chinese imperial examination system, established in its mature form during the Sui Dynasty (605 AD) and lasting until 1905, was perhaps the most sophisticated meritocratic bureaucracy in human history. Any man, regardless of birth, could in theory rise to the highest offices of state by passing a series of increasingly difficult examinations in Confucian classics, poetry, and administrative theory.

For centuries, this system produced a governing class of remarkable competence. Chinese administration was, by the standards of its time, extraordinarily efficient. The empire maintained roads, canals, granaries, and a postal system across a territory larger than Europe. It conducted regular censuses, collected taxes with reasonable efficiency, and provided a degree of public order that European visitors found astonishing.

But the examination system became ossified. The curriculum was fixed on the Confucian classics and did not evolve to incorporate new knowledge — science, technology, economics, or military strategy. The system selected for men who could write elegant eight-legged essays on ancient texts, not for men who could build railways, design weapons, or negotiate with foreign powers. By the nineteenth century, China’s bureaucratic class was superbly educated in a body of knowledge that was largely irrelevant to the challenges the empire faced.

When Western powers arrived with steamships, modern artillery, and industrial economies, China’s bureaucracy was unable to respond effectively. The examination system had created a class of administrators who were brilliant within their domain but incapable of thinking outside it. Reformers who proposed adopting Western technology and methods were blocked by officials whose entire careers, identities, and social status depended on the existing system. The result was China’s “century of humiliation” — a period of military defeats, unequal treaties, and territorial losses that only ended with the Communist revolution in 1949.

Part 4: The Modern Western Bureaucracy

The growth of Western bureaucracies since 1945 follows the same pattern, albeit in a less dramatic form.

After the Second World War, Western governments expanded rapidly to provide the welfare state, regulate the economy, and manage the Cold War. This expansion was, like Diocletian’s reforms and the Ottoman devshirme, a rational response to real problems. Citizens needed healthcare, pensions, education, and social insurance. Economies needed regulation to prevent the kind of catastrophic failures that had produced the Great Depression. National defence required permanent military establishments and intelligence agencies.

But C. Northcote Parkinson, writing in 1955, identified the dynamic that would define the next seventy years: “Work expands so as to fill the time available for its completion.” Parkinson demonstrated, with characteristically dry British humour, that the number of officials in the British Colonial Office continued to grow even as the number of colonies shrank. The Admiralty employed more civil servants as the Royal Navy got smaller. Bureaucracies, Parkinson observed, expand not because there is more work to do, but because officials create work for each other.

The numbers bear this out. In the United States, the Code of Federal Regulations has grown from approximately 10,000 pages in 1950 to over 180,000 pages today. The European Union produces an estimated 2,500 new regulations per year. The UK’s tax code is now longer than the complete works of Shakespeare — several times over. Each regulation requires officials to draft it, implement it, enforce it, and adjudicate disputes about it. Each official requires managers, support staff, office space, and pension contributions. The system feeds itself.

Supra-national bureaucracies have compounded the problem. The European Union employs approximately 60,000 officials in its institutions — a number that does not include the hundreds of thousands of national officials required to implement EU directives. The United Nations system employs over 75,000 people. The World Health Organisation, the World Trade Organisation, the International Monetary Fund, and dozens of other international bodies each maintain their own permanent bureaucracies, with their own budgets, their own career structures, and their own institutional imperatives.

These organisations were created to solve specific problems. Many of those problems have been solved, or have changed beyond recognition, or have proved insoluble. But the organisations persist, because the people who work in them have mortgages, careers, pensions, and identities that depend on the organisation’s continued existence. As Thomas Sowell has observed, it is almost impossible to name a government programme that has been terminated because it achieved its objectives. Success and failure produce the same result: the programme continues.

Part 5: DOGE and the AI Revolution

This is the context in which DOGE must be understood. Elon Musk’s initiative is not the first attempt to cut government waste. The United States has a long history of such efforts: the Hoover Commissions (1947 and 1953), the Grace Commission (1982), Al Gore’s “Reinventing Government” initiative (1993), and numerous others. All produced reports. All identified billions in potential savings. All were largely absorbed by the bureaucracy they were designed to reform.

What makes DOGE potentially different is technology. Artificial intelligence can do much of what bureaucrats do — process applications, check compliance, review contracts, detect fraud, answer queries — faster, cheaper, and more accurately. AI does not need a pension. It does not form a union. It does not create work for itself. It does not resist its own obsolescence.

The implications are profound. If AI can replace a significant portion of government administrative functions, then the economic and political case for maintaining large permanent bureaucracies collapses. Governments could become dramatically leaner — not by cutting services, but by delivering the same services with a fraction of the human staff. Tax collection, benefits administration, regulatory compliance, procurement, and dozens of other functions could be largely automated.

This is not speculation. Estonia has already demonstrated what a digitally native government looks like. Estonian citizens can file taxes in minutes, vote online, start a business in hours, and access virtually all government services through a single digital identity. Estonia achieves this with a fraction of the per-capita government employment of larger European nations.

Part 6: What Comes Next

History suggests two possible futures.

In the first, AI enables a genuine reduction in the size and cost of government. Nations that successfully slim their states gain a competitive advantage: lower taxes, faster decision-making, less regulatory burden on businesses and individuals, and more resources available for productive investment. The 2030s see a divergence between lean, AI-enabled states and bloated, bureaucracy-captured states. The lean states attract talent, investment, and innovation. The bloated states stagnate.

In the second, the bureaucracy absorbs AI just as it has absorbed every previous reform. AI systems are deployed not to replace officials but to assist them — creating new roles for “AI oversight officers,” “algorithmic accountability managers,” and “digital ethics compliance coordinators.” The technology that should have reduced the bureaucracy instead adds another layer to it. Parkinson’s Law prevails, as it always has.

The supra-national dimension adds further complexity. The European Union’s approach to AI regulation — the AI Act, the Digital Services Act, and the growing body of algorithmic governance rules — risks creating an entirely new bureaucratic apparatus dedicated to regulating the technology that was supposed to reduce bureaucracy. If the EU requires thousands of officials to oversee AI systems that were designed to replace thousands of officials, the net effect is zero — or worse.

The historical pattern is clear. Bureaucracies that cannot be reformed from within must eventually be dismantled from without. The Roman bureaucracy was swept away by the Germanic kingdoms. The Janissaries were destroyed by the Sultan’s cannon. China’s examination system was abolished by revolution. In each case, the institution had to be broken because it could not be bent.

The question for the 2030s is whether democratic societies can achieve what Mahmud II achieved in 1826 — the deliberate, planned dismantling of a parasitic institutional class — without the violence. AI provides, for the first time in history, a tool that could make this possible. Whether it will be used for that purpose, or whether it will be captured by the very institutions it threatens, is the defining governance question of our time.

Juvenal’s question remains unanswered. But for the first time, we may have a guard that does not need guarding.