I. The Quiet Fantasy That Crosses Every Aisle
It usually starts with a joke — or something close enough to one that the person saying it can walk it back if the room goes cold.
A commuter train arrives forty minutes late because of a signalling dispute that has been unresolved for three years. A coastal highway project, fully funded and technically straightforward, sits paralysed inside an environmental review process that has now outlasted two administrations. A public health agency spends the better part of a decade drafting nutritional guidelines that, by the time they are published, have been watered down by eighteen rounds of lobbying, four changes of minister, and one ill-timed election cycle.
Someone in the room sighs.
“Why can’t we just get one smart person in charge and let them fix it?”
The comment rarely comes loaded with ideology. That is precisely what makes it interesting. It surfaces in progressive circles when climate legislation dies in committee for the third consecutive session, strangled by procedural manoeuvre rather than substantive objection. It surfaces in conservative ones when regulatory agencies accumulate mandates across decades without ever shedding them, growing like institutional coral. It surfaces most reliably among the self-described pragmatists — the centrists, the technocrats, the people who describe themselves as “policy people, not political people” — when both sides of a legislature would rather perform disagreement than resolve it.
The fantasy, in its purest form, runs something like this: imagine a leader uncorrupted by the incentives that deform conventional politicians. Someone with no donors to reward, no party apparatus to appease, no electoral horizon forcing every decision into a four-year window. A figure of genuine competence — the kind that gets things built, budgets balanced, institutions staffed with experts rather than loyalists. Call this person the Sovereign CEO.
The model borrows consciously from the corporate world. In a well-run organisation, a chief executive sets direction, appoints specialists, and holds them accountable for outcomes. Strategy is data-informed. Implementation is rapid. When a department underperforms, it is restructured rather than subsidised. The comparison to government seems almost to make itself: ministries become divisions, citizens become stakeholders, policy becomes product roadmap. The nation, in this vision, is simply a very large organisation that has been catastrophically mismanaged for want of the right executive.
What gives the fantasy its peculiar staying power is that it does not require the person entertaining it to be authoritarian in temperament. Most people who have quietly entertained this idea are, in their daily lives, thoroughly committed to democratic norms. They vote. They value civil liberties. They would recoil from a genuine coup. The fantasy they are indulging is softer and more respectable than all that: not a strongman, but a philosopher-administrator. Not a tyrant, but a very talented, very honest, very patient manager of national complexity.
And yet the fantasy, however gently held, is making a specific and consequential argument: that the mechanisms of democratic governance — elections, legislatures, independent courts, a free press, the whole scaffolding of institutional checks — are, at their core, inefficiencies. Obstacles. The price we pay for living in a system designed to accommodate disagreement rather than transcend it.
That argument deserves to be taken seriously — which means it deserves to be examined honestly.
Because the frustration that produces it is entirely legitimate. Modern democratic governments are frequently, genuinely, maddening in their slowness. The legislative gridlock in the United States Congress has become a structural feature rather than a periodic crisis. The European Union’s regulatory architecture — designed to harmonise twenty-seven distinct legal and political traditions — produces processes of such baroque complexity that entire industries employ specialists simply to navigate them. At the municipal level, planning permissions for affordable housing can take longer to clear than it takes to build the houses. The gap between what governments know needs doing and what they can actually accomplish, in the time available, is vast and frequently grotesque.
But the question worth asking is not whether democratic governance is slow. It is. The question is whether its slowness is a design flaw or a design feature — and whether the thing the Sovereign CEO fantasy promises to fix is actually the problem, or merely its most visible symptom.
II. When a Nation Looks Like a Broken Startup
There is a particular kind of frustration that Silicon Valley learned to monetise.
It goes by different names at different moments in history — disruption, innovation, moving fast — but its essential structure is always the same: identify a system that is doing something inefficiently, strip away the institutional friction around it, and replace the whole apparatus with something cleaner, faster, and more responsive to the people it is supposed to serve. Uber did it to urban transport. Airbnb did it to hospitality. A generation of venture-backed founders looked at every incumbered industry and asked the same foundational question: what would this look like if we built it from scratch today?
It was perhaps inevitable that someone would eventually point the question at government itself.
The intellectual case for treating the nation-state as a startup has been made with varying degrees of seriousness over the past two decades, but its core logic is consistent. Governments, the argument runs, are legacy institutions — built for conditions that no longer exist, staffed by incentive structures that reward process over outcome, and governed by accountability mechanisms so diffuse that no single person is ever clearly responsible for anything. They accumulate functions the way old buildings accumulate extensions: not by design, but by accretion, each addition made sense at the time, and the resulting structure is now labyrinthine, expensive to heat, and structurally dubious.
A well-run organisation, by contrast, has a clear command hierarchy. The chief executive sets direction. A cabinet of genuine specialists — not political appointees cycling through ministries they have no expertise in, but domain authorities who have spent careers mastering their fields — translates that direction into operational decisions. A Nobel Prize-winning economist runs the treasury. A career epidemiologist runs public health. A civil engineer of international standing runs infrastructure. Each of them is accountable to the executive for outcomes, not for the political palatability of their methods. Policy proposals are stress-tested against data and simulation before implementation. When something fails, it fails fast, and the system learns.
The appeal of this model is not merely aesthetic. There are measurable domains in which it appears to work.
Consider the transformation of Shenzhen. In 1980, it was a fishing town of roughly 30,000 people on the southern coast of China, separated from Hong Kong by a narrow strip of water and, at the time, several decades of divergent economic development. The Chinese government designated it a Special Economic Zone, gave it a mandate for market-oriented experimentation, and moved with a decisiveness that democratic systems structurally cannot replicate. Within four decades, Shenzhen had become a city of over seventeen million people, home to some of the world’s most significant technology companies — Huawei, Tencent, DJI — and a manufacturing ecosystem of extraordinary sophistication. Infrastructure that would require twenty years of environmental review, planning appeals, and legislative appropriation in most Western democracies was built in years. Entire districts materialised. A subway system of continental scale was constructed at a pace that transit planners in London or Los Angeles can only regard with a mixture of admiration and existential despair.
For observers already inclined toward the Sovereign CEO fantasy, Shenzhen is exhibit A. It is the physical evidence that centralised authority, competently wielded, can compress decades of development into a single generation. And it is difficult to dismiss entirely, because the buildings are real, the companies are real, and the prosperity — however unevenly distributed and however compromised by the surveillance architecture that accompanies it — is measurably real.
But Shenzhen is a carefully bounded example, and the boundaries matter enormously.
The Special Economic Zone model works partly because it is exceptional — a deliberate deviation from the surrounding system, given latitude precisely because it is not the whole country. It is also a model of economic development in a high-growth phase, which is an operationally different problem from governing a mature, complex, pluralistic society with competing constituencies, entrenched interests, and deep disagreements about fundamental values. Building a city when the primary variable is speed is a different problem from deciding how a society should balance privacy against security, or growth against environmental preservation, or the interests of this generation against the next. One is an engineering problem. The other is a political one. And the distinction, as we will see, is not trivial.
The startup theory of government also tends to reach, at some point, for the example of Singapore — and with more justification, because Singapore is not an economic zone but an actual state, governing actual citizens across the full range of what governments are required to do.
After independence in 1965, Singapore faced a constellation of challenges that would have tested any government: a tiny land mass with no natural resources, significant ethnic and religious heterogeneity, and neighbours whose intentions were not always benign. Under Lee Kuan Yew, it responded with a model of governance that was self-consciously technocratic. The civil service was built on rigorous meritocratic selection, with salaries deliberately set to compete with the private sector and corruption treated not as an unfortunate reality to be managed but as an existential threat to be eliminated. Long-term planning — in housing, education, water security, port infrastructure — was insulated from short-term electoral pressure. The results, measured by almost any conventional metric, were remarkable. Life expectancy, literacy, per capita income, institutional integrity: Singapore consistently ranks near the top of global indices in all of them.
The temptation, looking at Singapore, is to conclude that the startup theory of government has been vindicated. That a small, smart team with clear authority and long time horizons can simply outperform the lumbering machinery of adversarial democracy.
But there is a detail that the startup theory consistently underweights, and it is the most important detail of all.
Lee Kuan Yew was an exceptional human being in a specific historical moment. Singapore’s model did not produce its outcomes through the genius of its institutional design alone — it produced them through the intersection of strong institutions and an unusually disciplined, unusually far-sighted founding leader who happened to be in the right place at a moment of genuine national emergency, when the population was willing to accept constraints that populations in more settled circumstances would not.
This is not a replicable formula. It is not a template. It is, in the language of statistics, an outlier — and building a theory of governance on outliers is how civilisations make catastrophic category errors.
The startup analogy, pressed hard enough, eventually reveals its own fatal flaw. Most startups fail. The ones that succeed do so in conditions of relative simplicity — a defined market, a quantifiable product, a measurable outcome. And even the most celebrated of them, given enough time and enough power, tend to develop precisely the pathologies they were created to disrupt: rigidity, self-interest, insularity, and an institutional resistance to the kind of honest feedback that once made them fast.
Nations are not startups. They are not even companies. They are the most complex adaptive systems that human beings have ever attempted to govern — and the history of treating them otherwise is not, on balance, an encouraging one.
III. Plato’s Dangerous Gift to Modern Politics
The fantasy did not begin in Silicon Valley. It did not begin with management consultants rebranding statecraft as organisational strategy, or with venture capitalists discovering that disruption had geopolitical applications. It began, with characteristic directness, in Athens — in the work of a philosopher who had watched democracy execute his mentor and concluded, with cold precision, that the governed had no business selecting their governors.
Plato’s Republic, written in approximately 380 BC, is one of the most influential and most dangerous books in the history of political thought — dangerous not because it argues for cruelty, but because it argues for wisdom, and makes the argument so elegantly that readers have been seduced by it for two and a half millennia. The society Plato envisioned was not a tyranny in the conventional sense. It was a meritocracy of the enlightened — a state governed by philosopher-kings, individuals who had completed a decades-long education in mathematics, logic, ethics, and metaphysics, and who had arrived, through genuine intellectual effort, at an understanding of the Good that ordinary citizens could not be expected to share.
The Platonic critique of democracy was not merely rhetorical. It was structural. Democracy, Plato argued, treats all opinions as equally valid and all citizens as equally qualified to make collective decisions — which is, he observed, precisely what we do not assume in any other domain of expertise. We do not ask passengers to vote on the optimal flight path. We do not put bridge engineering to a public referendum. We select the most capable person, give them the relevant authority, and hold them accountable for the outcome. Why, Plato asked, should the governance of an entire society be any different?
It is a question that has never been satisfactorily answered — which is part of why it keeps being asked.
The philosopher-king concept did not survive antiquity merely as an academic curiosity. It has threaded itself, in various forms, through the entire subsequent history of political thought. The Roman emperor Marcus Aurelius — himself a committed Stoic philosopher who appears to have regarded power primarily as an obligation — was read by later generations as a living demonstration that the ideal was not entirely impossible. His Meditations, a private journal never intended for publication, reveal a ruler engaged in constant moral self-examination, reminding himself of the duties of office with a rigour that no election could have imposed and no electorate could have demanded. The Stoic tradition from which he drew had its own theory of natural hierarchy — not of birth or wealth, but of rational capacity, of the ability to govern one’s own passions as a prerequisite for governing others.
In the medieval Islamic world, the concept found expression in the tradition of the mirror for princes — a genre of political literature offering guidance to rulers on the cultivation of justice, wisdom, and restraint. Al-Farabi, writing in the tenth century, developed an explicitly Platonic theory of the ideal Islamic state, governed by a philosopher-prophet whose authority derived simultaneously from rational excellence and divine illumination. The assumption underlying all of this literature was consistent: that good governance was primarily a function of the quality of the ruler, and that the central political problem was therefore not institutional design but the identification and cultivation of exceptional individuals.
The Enlightenment complicated the picture without resolving it. Voltaire’s enthusiasm for Frederick the Great of Prussia — a monarch who read philosophy, corresponded with intellectuals, and described himself as the first servant of the state — reflected a recurring Enlightenment hope that reason might find its political expression not in popular sovereignty but in enlightened despotism. Frederick modernised Prussian administration, introduced legal reforms, and practised religious tolerance at a time when such tolerance was genuinely exceptional. He was also an absolutist who conducted wars of aggression and regarded his subjects as instruments of state power rather than bearers of rights. The enlightened despot, it turned out, was difficult to separate from the ordinary kind.
This is the recurring difficulty with the Platonic inheritance. The argument for philosopher-kings depends on two premises that are each, independently, highly contestable. The first is that wisdom of the kind required for governance can be reliably identified in advance — that we have, or could develop, the diagnostic tools to distinguish genuine understanding from its convincing simulation. The second is that individuals possessing such wisdom would reliably use power in the interests of the governed rather than themselves. Plato was aware of the second problem and tried to solve it institutionally: philosopher-kings, in his design, would own no property, have no family, and be deliberately insulated from the material incentives that corrupt ordinary rulers. He was, in effect, trying to engineer away the human condition.
History has not been kind to this project.
What history has produced, repeatedly, is a different figure: the leader who arrives with genuine competence and genuine good intentions, who accumulates authority on the strength of early successes, and who then — gradually, then suddenly — loses the capacity to distinguish between the interests of the state and their own judgment of what those interests require. The philosopher-king becomes, not through malice but through the ordinary mechanics of unchecked power, something considerably less philosophical.
Singapore’s Lee Kuan Yew, for all his genuine achievements, presided over a legal and political system that used defamation suits to suppress political opposition, controlled the press with a directness that would be unacceptable in any liberal democracy, and structured electoral boundaries in ways that reliably advantaged the ruling party. Whether these constraints were a reasonable price for stability and prosperity is a legitimate debate. What is not debatable is that they represent exactly the kind of power that the benevolent autocrat always eventually claims: the authority to determine, unilaterally, which freedoms the population needs and which it can safely be denied.
Plato’s gift to modern politics, then, is double-edged. He gave us the most rigorous and intellectually serious version of the technocratic fantasy — and he gave us no reliable way to ensure it remains a fantasy about wisdom rather than a rationalisation for control. The philosopher-king, stripped of Plato’s idealism and placed in actual political conditions, tends to become something that Plato himself would have recognised: not the guardian of the Good, but a very intelligent person who has become convinced that their intelligence exempts them from the normal constraints of accountability.
That conviction, as we will see, has a specific and well-documented effect on the information available to people who hold it.
IV. The Man at the Top Always Hears Good News
There is a scene that repeats itself, with minor variations, across the entire recorded history of centralised power. A leader — competent, well-intentioned, surrounded by capable people — makes a decision based on the best available information. The decision proves to be wrong. Not catastrophically wrong at first, but measurably, undeniably wrong in ways that the people closest to the consequences can see immediately. And then something strange happens.
The information about how wrong the decision was does not travel upward.
It stops somewhere in the middle of the hierarchy, softened, contextualised, and reframed into something that sounds less like failure and more like a temporary deviation from an otherwise sound trajectory. By the time it reaches the leader, it has been so thoroughly processed by the institutional anxiety of everyone who handled it along the way that it bears only a genealogical relationship to what actually happened. The leader, receiving this laundered version of reality, makes another decision. The cycle continues.
This phenomenon — the systematic distortion of information as it moves toward centralised authority — is arguably the most structurally important problem in autocratic governance, and it receives far less attention than the more dramatic pathologies of dictatorship. We talk extensively about corruption, about violence, about the suppression of political opposition. These are real and serious failures. But they are, in a sense, visible failures — they produce evidence, they generate resistance, they eventually become impossible to ignore. Information distortion is different. It is invisible by design. It operates not through the commission of obvious wrongs but through the quiet omission of inconvenient truths, and it does its damage long before anyone recognises that damage is occurring.
The mechanisms that produce it are not mysterious. They are the entirely predictable consequences of rational behaviour under conditions of concentrated power.
When your professional survival depends on the approval of a single authority, the calculus around honesty changes in ways that do not require bad character to understand. A mid-level official who delivers bad news risks being associated with the bad news. One who delivers good news — or who frames bad news in ways that imply it is being competently managed — is demonstrating reassuring qualities: composure, capability, optimism. In an organisation with normal checks and balances, this incentive is partially counteracted by the existence of independent scrutiny — auditors, journalists, opposition politicians, oversight committees — that can surface what the internal hierarchy prefers to conceal. In a system where those counterweights have been removed or subordinated, the incentive operates without friction.
The results, historically, have been remarkable in their consistency.
During the collectivisation campaigns of the Soviet Union in the late 1920s and early 1930s, local officials responsible for meeting grain procurement quotas faced a choice between reporting accurate shortfalls — which would imply personal failure and invite punishment — and reporting inflated figures that suggested the campaign was succeeding. The incentive to falsify was overwhelming, and the falsification was systematic. Stalin and the central planning apparatus in Moscow received production data suggesting agricultural output was meeting targets at precisely the moment when millions of people were starving. The famine that killed between five and eight million people in Ukraine and Kazakhstan between 1932 and 1933 was, in significant part, a famine that the government did not officially know was happening — because the information architecture of the Soviet state had been structured in a way that made knowing impossible.
This is an extreme example, but the extreme version illuminates the general principle. The problem is not unique to regimes willing to use violence to enforce conformity. It appears, in milder but structurally identical forms, in any organisation where power is sufficiently concentrated and the costs of delivering unwelcome assessments are sufficiently high. Research on organisational behaviour has documented the phenomenon extensively in corporate settings: the more hierarchical and punitive a management culture, the less accurate the information that reaches senior leadership, and the more likely the organisation is to pursue failing strategies long past the point where a well-informed executive would have changed course.
The Sovereign CEO fantasy tends to assume that a genuinely wise and self-aware leader could solve this problem through personal qualities — by actively cultivating a culture of honesty, by explicitly rewarding bearers of bad news, by demonstrating through consistent behaviour that accurate information is valued above comfortable information. And there are leaders who have made serious efforts in this direction. The difficulty is structural rather than characterological: it is not enough for the leader to want honest information if the entire system below them has been shaped, over time, by the knowledge that honest information carries risk.
Consider what it actually requires to deliver a genuinely unwelcome assessment to someone who holds absolute authority over your career, your reputation, and in some systems your liberty. It requires not merely personal courage — though it requires that — but also a calculation that the leader’s commitment to honest counsel is robust enough to survive the specific content of what you are about to say. And that calculation must be made fresh every time, against the background of every previous instance in which someone in a similar position made the same calculation differently and suffered for it. The institutional memory of what happens to people who tell the truth accumulates faster than any individual leader’s efforts to counteract it.
The philosopher-king, in other words, does not simply need to be wise. They need to be wise, and visibly, consistently, and credibly committed to hearing things they do not want to hear — and they need to maintain that commitment across the full duration of their tenure, including the years when they are tired, when they are under pressure, when the evidence contradicts decisions they have publicly staked their authority on, and when the people delivering the bad news are not doing so with perfect tact.
That is an extraordinarily demanding standard. It is a standard that almost no human being, in any context, consistently meets.
And even if the leader meets it, there remains a deeper problem — one that cannot be solved by any amount of personal virtue or institutional commitment to candour. It is not merely that centralised power distorts the flow of information. It is that the information required to govern a modern society well is so vast, so distributed, and so irreducibly local in character that no central authority, however well-intentioned and however well-informed, can ever fully possess it.
That is the subject of the next section — and it is, in many ways, the most fundamental challenge the Sovereign CEO fantasy has yet to answer.
V. The Unbearable Weight of Knowing Everything
In 1945, a Austrian-British economist named Friedrich Hayek published an essay that was, on its surface, a technical contribution to a debate about economic planning. It was titled, with academic modesty, The Use of Knowledge in Society. It ran to fewer than twenty pages. And it contained an argument that remains, eight decades later, one of the most penetrating analytical challenges ever mounted against the idea that a small group of sufficiently intelligent people can successfully direct the affairs of a large and complex one.
Hayek’s argument was not primarily about incentives, or corruption, or the abuse of power. It was about something more fundamental and, in some ways, more uncomfortable: the nature of knowledge itself.
The central planning debates of the mid-twentieth century tended to assume that the core problem with centralised economic management was computational — that the challenge of coordinating an entire economy was simply a matter of processing enough data quickly enough, and that with sufficiently advanced technology, that processing could in principle be achieved. Hayek’s intervention was to argue that this framing missed the point entirely. The problem was not computational. It was epistemological. And it could not be solved by any conceivable increase in processing power, because the information required for effective coordination did not exist in any form that could be gathered and processed centrally.
The knowledge that actually governs economic life — and, by extension, social and political life — is not the kind of knowledge that appears in reports, statistics, or expert analyses. It is what Hayek called local knowledge: the knowledge of particular circumstances, of specific times and places, of individual capabilities and preferences and relationships that exists, irreducibly and exclusively, in the minds of the people who possess it. A farmer knows things about their specific soil, their specific microclimate, and their specific market relationships that no agricultural ministry, however expert and however well-staffed, can know. A small manufacturer knows things about their supply chain, their workforce, and their customers that cannot be fully captured in any dataset, because significant portions of it are tacit — known in practice, embedded in relationships, expressed in behaviour rather than articulable in language.
This distributed, granular, partially tacit quality of real-world knowledge is not a temporary limitation awaiting a technological solution. It is a structural feature of complex social systems. And it means that the Sovereign CEO — however intelligent, however well-advised, however genuinely committed to making good decisions — is permanently operating with an impoverished version of the information that the system they are governing actually contains.
Prices, in Hayek’s analysis, are the mechanism through which decentralised systems transmit this otherwise inaccessible knowledge. When the price of a material rises, it signals — without anyone needing to understand why, or to communicate the underlying reasons — that something has changed in the balance of supply and demand, and that actors throughout the system should adjust their behaviour accordingly. The signal travels instantly, it carries no bureaucratic overhead, and it aggregates the distributed knowledge of millions of individual actors into a single number that everyone can act on. No central authority could replicate this function, because replicating it would require knowing what the price already encodes — which is precisely the knowledge that is not centrally available.
The implications extend well beyond economics.
Modern societies are not merely economic systems. They are dense, overlapping networks of social relationships, cultural practices, institutional arrangements, and individual life projects — each with their own local logic, their own embedded knowledge, their own criteria for what counts as a good outcome. A health policy that works brilliantly in an urban, highly educated, well-connected population may be entirely unsuited to a rural community with different social structures, different relationships with medical authority, and different material conditions. An education reform that improves measurable outcomes in one cultural context may systematically undermine the informal learning networks that actually carry knowledge in another. A housing programme that solves a quantifiable shortage may, in doing so, dissolve the neighbourhood relationships that gave the community its capacity for collective action.
None of this means that centralised policy is always wrong or that expertise is irrelevant. It means that the gap between what a central authority knows and what it would need to know to govern well is not a gap that can be closed by hiring better advisers or processing more data. It is a gap that is constitutive of the relationship between any governing authority and the society it governs — and the appropriate institutional response to it is not to try to close it, but to design systems that do not require it to be closed.
This is precisely what decentralised political systems attempt, however imperfectly, to do. Federalism distributes decision-making authority to the level at which local knowledge is most concentrated. Markets allow individual actors to make decisions on the basis of information that only they possess. Democratic deliberation — slow, frustrating, and frequently irrational in its surface manifestations — is, among other things, a mechanism for surfacing the distributed preferences and local knowledge of a population in ways that no technocratic assessment process can replicate.
The Sovereign CEO fantasy implicitly assumes that governing a modern society is primarily a problem of intelligence — that if you could find someone smart enough, informed enough, and honest enough, they could take all the relevant inputs, run them through a sufficiently sophisticated analytical framework, and produce decisions that are better than the messy, compromised, agonisingly slow outputs of democratic deliberation. Hayek’s insight is that this assumption is not merely optimistic. It is a category error.
It confuses the kind of problem that governance actually is — a problem of coordinating the distributed knowledge and competing legitimate interests of millions of individuals — with the kind of problem that intelligence can solve: a problem with a knowable answer that a sufficiently capable mind can, in principle, compute.
The difference between these two types of problems is not academic. It has produced, in the twentieth century alone, policy catastrophes of extraordinary scale — not because the people making the decisions were unintelligent or malicious, but because they were attempting to solve a problem that was not, in the relevant sense, solvable from the position they occupied.
And yet the fantasy persists. Partly because Hayek’s argument is abstract and the frustrations of democratic governance are visceral. Partly because the cases where centralised authority appears to have worked — Shenzhen, Singapore, the Marshall Plan’s administrative machinery — are vivid and concrete while the counterfactual costs are invisible. And partly because the twentieth century’s most dramatic centralised failures can always be attributed, by those inclined to do so, to the specific pathologies of specific regimes rather than to anything structural about the concentration of power itself.
But there is another reason the fantasy persists — one that is harder to argue with because it is, in a limited and carefully bounded sense, true. There are conditions under which centralised systems genuinely do outperform decentralised ones. There are problems for which speed and coordination matter more than local knowledge and distributed preference. There are moments when the off-road truck is simply too slow, and what you need is something faster.
Understanding when those conditions apply, and when they do not, requires a different kind of analysis — one that begins not with the quality of the leader, but with the design of the system.
VI. Formula 1 Governments and the Vehicles We Actually Need
Begin with a thought experiment that has nothing to do with politics.
You are given a transportation problem. You need to move people and goods across a large, diverse, and climatically varied territory — some of it well-paved urban infrastructure, some of it unmaintained rural track, some of it no road at all. The terrain changes unpredictably. The weather is unreliable. The destinations are numerous and constantly shifting. You have a budget, a timeline, and a mandate to keep things moving under all conditions, including the bad ones.
Someone walks into the room and offers you a Formula 1 race car.
It is, by any objective engineering standard, a magnificent machine. Its power-to-weight ratio is extraordinary. Under ideal conditions — a smooth, dry, purpose-built circuit, operated by a trained specialist, supported by a pit crew of forty engineers monitoring every variable in real time — it performs at levels that no other vehicle can approach. It is the product of billions of dollars of accumulated engineering intelligence, refined across decades of competitive pressure into something very close to the physical limit of what is achievable. As a demonstration of what optimisation can produce, it is almost without equal.
It will be completely useless for your actual problem.
Not because it is poorly designed, but because it is exquisitely, precisely, irreversibly designed for a specific set of conditions that your problem does not provide. Take it off the circuit and onto a gravel road and it will be damaged within minutes. Remove the pit crew and the telemetry and the specialist driver and its performance collapses. Subject it to rain, to mud, to an unexpected obstacle, to any of the routine unpredictabilities of real-world operation, and its extraordinary optimisation becomes an extraordinary liability. The same properties that make it fast under ideal conditions make it fragile under real ones.
Now someone offers you a truck.
It is not beautiful. Its power-to-weight ratio is unimpressive. On a smooth circuit it would be lapped repeatedly by the race car and would not particularly care. It is loud, expensive to run, and it handles with the elegance of a building. But it will start in cold weather. It will navigate unpaved surfaces. It will carry substantial loads over long distances without requiring a specialist crew to keep it operational. When something goes wrong — and things will go wrong — it can usually be repaired with tools that are available in the field by people who learned to use them without a degree in aeronautical engineering. It is not optimised for any single condition. It is adequate across a very wide range of them.
This, rendered in mechanical terms, is the central trade-off in political system design.
Autocracies are Formula 1 cars. They are systems optimised for a specific operating condition: the presence of a competent, well-informed, public-spirited leader operating in a stable environment with manageable complexity and a population willing to accept centralised direction. Under those conditions — which do occasionally obtain, for limited periods, in specific historical circumstances — they can achieve results that democratic systems structurally cannot. The pace of infrastructure development in China’s high-speed rail network, which now covers more than 45,000 kilometres and took roughly fifteen years to build at scale, is the race car on a good circuit. It is genuinely impressive. It represents a coordination capacity that no democratic system, with its environmental reviews and legislative appropriations and judicial challenges and local referenda, could have replicated in the same timeframe.
But the operating conditions for autocratic systems are rarely stable for long, and the consequences when they degrade are not proportional to the degree of degradation. They are catastrophic in a specific way that the mechanical metaphor captures precisely: when a Formula 1 car loses a component at speed, it does not slow down gracefully. It disintegrates. The same engineering properties that concentrate performance under ideal conditions concentrate failure under adverse ones. There is no graceful degradation. There is function, and then there is wreckage.
Democratic systems fail differently, and the difference matters enormously.
When a democratic government makes a serious policy error — and they make them regularly, consequentially, and sometimes with considerable human cost — the system contains, by design, multiple mechanisms that can identify the error and force a correction. An independent judiciary can strike down legislation that violates constitutional principles. A free press can surface the evidence of failure that the government would prefer to remain buried. An opposition party, motivated by nothing more elevated than the desire to win the next election, will reliably publicise the government’s mistakes with a thoroughness and enthusiasm that no internal audit process could match. And elections — slow, expensive, frequently distorted by money and media, producing outcomes that satisfy nobody completely — nevertheless provide a regular, legitimate, and peaceful mechanism for replacing governments that have lost the confidence of the governed.
None of these mechanisms are fast. None of them are elegant. The process by which the United Kingdom’s National Health Service has been reformed, restructured, and reformed again across successive governments of different political orientations is not, by any reasonable aesthetic standard, an efficient process. It is slow, contentious, expensive in administrative overhead, and productive of considerable institutional confusion. But it is also a process that has, over seventy-five years, repeatedly corrected course in response to evidence of failure — adjusting priorities, reallocating resources, responding to the distributed knowledge of clinicians and patients and administrators in ways that no single centralised plan could have anticipated.
The truck keeps moving. Slowly, noisily, with constant argument about the route. But it keeps moving.
There is a further dimension to the mechanical metaphor that is worth developing, because it concerns not just how these systems perform but how they are maintained — and maintenance, in political systems as in mechanical ones, turns out to be the variable that most reliably predicts long-term survival.
Formula 1 cars require constant, expert, resource-intensive maintenance to remain operational. Remove the pit crew and the car degrades rapidly. The maintenance infrastructure is not a peripheral support function — it is a core component of the system’s ability to perform. And crucially, the maintenance must be performed by specialists who understand the machine deeply, who have the authority to take it off the circuit when it needs attention, and whose professional judgment is not overridden by the driver’s desire to keep racing.
Autocratic systems have an analogous maintenance requirement, and they consistently fail to meet it. The institutions that could perform the maintenance function — independent courts, a free press, an autonomous civil service, political opposition with genuine power — are precisely the institutions that centralised authority finds most threatening and most systematically dismantles. An autocrat who tolerates a genuinely independent judiciary is tolerating an institution that might rule against them. One who permits a genuinely free press is permitting an institution that will publish things they would prefer unpublished. The maintenance infrastructure and the performance architecture are in direct tension, and under conditions of concentrated power, performance architecture wins — until the day it catastrophically doesn’t.
Democratic backsliding research conducted by the V-Dem Institute at the University of Gothenburg, which tracks political systems across the world, consistently finds that the erosion of democratic institutions follows a recognisable pattern: independent oversight mechanisms are weakened first, often under the justification of efficiency or national interest, and the full consequences of that weakening become apparent only years later, when the error-correction mechanisms are no longer functional and the errors have compounded beyond manageable scale. By the time the race car hits the gravel, the pit crew has already been dismissed.
The truck, by contrast, is almost impossible to total. Its redundancy — the very quality that makes it inefficient under ideal conditions — makes it resilient under adverse ones. When one correction mechanism fails, others compensate. When a government pursues a disastrous policy, the opposition pursues power by promising to reverse it. When a court makes a wrong decision, the legislature can respond with new law. When a legislature passes bad law, the courts can review it. The system is not optimised. It is, in the engineering sense, over-specified — carrying more capacity for self-correction than it needs in good times, precisely so that it has enough in bad ones.
The question, then, is not which system produces the better outcomes under ideal conditions. The answer to that question is straightforward, and it favours centralised authority, at least in the domains where speed and coordination are the primary variables. The question is which system you want to be living in when the leader is not exceptional, when the conditions are not stable, and when the decisions being made are consequentially, damagingly wrong.
And there is a follow-on question that the Formula 1 analogy makes unavoidable: even if you are fortunate enough to have the right driver, what happens when they are no longer in the car?
VII. The Throne Always Outlasts the Person Sitting On It
There is a particular kind of tragedy that recurs so consistently across the history of centralised power that it has acquired, in the study of political systems, almost the status of a natural law. It goes like this: a capable leader builds a system around their own capabilities. The system works, for a while, extraordinarily well. The leader ages, or dies, or is removed, and power passes to someone else. And the system, which was never really a system at all but a single person with institutional architecture arranged around them, begins to fail in ways that are rapid, severe, and very difficult to arrest.
Call it the succession trap. It is the point at which the benevolent autocracy’s foundational assumption — that good governance is primarily a function of having the right person in charge — collides with the one variable that no political system has ever successfully engineered away: the biological mortality of human beings, and the statistical near-impossibility of producing exceptional leaders on demand.
The Roman example is so frequently cited that it risks becoming a cliché, but it retains its force precisely because it is so clean and so well-documented. Marcus Aurelius, who ruled from 161 to 180 AD, is as close to the philosopher-king ideal as history offers — a man who read Epictetus in his youth, governed with conspicuous conscientiousness, and left behind in his Meditations a record of moral self-interrogation that remains genuinely moving seventeen centuries later. His reign was not without war or difficulty, but it was marked by a quality of institutional seriousness that his contemporaries recognised and his successors failed to replicate.
His successor was his son Commodus.
Commodus was not a monster in the operatic sense — not, at least, initially. He was something more instructive: a thoroughly ordinary man placed in a position that required an extraordinary one. He delegated to favourites, neglected administration, developed an enthusiasm for appearing in the arena as a gladiator that Romans found simultaneously undignified and alarming, and presided over a steady deterioration of the governmental competence his father had built. Within fifteen years of Marcus Aurelius’s death, Rome had entered what historians call the Year of the Five Emperors — a period of violent political instability that marked the beginning of the empire’s long institutional decline.
The system had not changed. The laws were the same. The administrative structures were the same. The legions were the same. Only the person at the top had changed — and that, it turned out, was everything.
This is not an isolated historical anecdote. It is a pattern of such consistency that it constitutes something close to an empirical regularity. The Ottoman Empire under Suleiman the Magnificent reached the peak of its territorial extent and administrative sophistication, then spent the following centuries in a slow institutional dissolution that accelerated with each mediocre successor. Frederick the Great of Prussia — Voltaire’s enlightened despot, the philosopher-king in military uniform — built a state of remarkable efficiency that his nephew Frederick William II promptly began to unravel through a combination of fiscal irresponsibility and strategic misjudgement. The first generation builds. The second maintains, with difficulty. The third inherits a system designed for a genius and operates it as an ordinary person, which is to say, badly.
Singapore, again, is instructive — not as a failure, because it has not failed, but as a case study in how thoughtfully designed the succession problem is even when thoughtful people are actively trying to solve it. Lee Kuan Yew was sufficiently self-aware to recognise that the system he built was partly a function of his own authority and credibility, and he invested considerable effort in constructing institutional arrangements — the People’s Action Party’s internal meritocracy, the civil service’s professionalism, the elected presidency as a constitutional safeguard — that might survive his departure. His chosen successor, Goh Chok Tong, was a capable administrator who managed the transition with considerable skill. His son, Lee Hsien Loong, governed for two decades with a competence that most democratic leaders would envy.
And yet the question of whether Singapore’s model is transferable — whether it represents a durable institutional achievement or an extended exercise in founder-dependent governance that has not yet faced its defining test — remains genuinely open. The institutions are strong. But they were built by, and in important ways still reflect the authority of, an exceptional founding figure whose like may not appear again. What happens in the third generation, or the fourth, when the institutional memory of the founding moment has faded and the system must run on its own internal logic rather than the borrowed credibility of its creator, is a question that Singapore has not yet had to answer.
The deeper problem with succession is not merely that exceptional leaders are rare — though they are, statistically, vanishingly rare, and any political theory that depends on their regular appearance is a political theory built on wishful thinking. The deeper problem is that the process by which autocratic successors are selected is almost perfectly designed to produce the wrong people.
In competitive democratic systems, leaders are subjected — imperfectly, noisily, often in ways that reward the wrong qualities — to a prolonged process of adversarial testing. They must build coalitions, survive public scrutiny, respond to criticism, adapt to changing conditions, and demonstrate the capacity to function under pressure across a range of circumstances. The process selects for charisma and media fluency more than it should, and for genuine governance competence less than it should. But it does select against certain catastrophic failure modes: the candidate who cannot function under public scrutiny, who collapses when challenged, who refuses all accountability, tends to be filtered out before reaching the highest office — or, having reached it, tends to be removed before doing terminal damage.
Autocratic succession processes select for entirely different qualities. In systems where the outgoing leader chooses their successor, selection pressure favours loyalty over competence, agreeableness over independent judgment, and personal closeness to the departing authority over the capacity to govern independently. The person most likely to be chosen is the person who poses least threat to the incumbent — which is frequently, and not coincidentally, the person least equipped to replace them. In systems where succession is contested among elites, the selection pressure shifts toward political manoeuvring, factional alliance-building, and the capacity to neutralise rivals — none of which correlates reliably with the ability to run a country well once the contest is over.
The result is a systematic regression toward the mean, and the mean, in the distribution of human capability and character, is not philosopher-king territory. The mean is ordinary — ordinary ambition, ordinary judgment, ordinary susceptibility to flattery and self-interest and the comfortable certainty that one’s own views are correct. In a democratic system, ordinary is manageable, because the system is designed to function with ordinary people and to correct for their ordinary failures. In an autocratic system, ordinary is dangerous, because the system was designed for someone exceptional and has no correction mechanism when that exception is not present.
There is a final, grimly ironic dimension to the succession trap that deserves acknowledgement. The qualities that make a leader most effective as a benevolent autocrat — the willingness to take responsibility, the confidence to make decisions without consensus, the capacity to maintain authority in the face of opposition — are precisely the qualities that make them most likely to cling to power beyond the point of their own effectiveness, and most resistant to the honest assessment that it is time to leave. The benevolent dictator who builds a succession plan and adheres to it, who identifies their own limitations and transfers authority gracefully to someone better equipped, who subordinates their identity to the institutional continuity of the system they built — this figure exists in the literature of political philosophy. In the historical record, they are considerably harder to find.
Lee Kuan Yew stepped back, gradually and with evident difficulty, over many years. He remained a Senior Minister, then a Minister Mentor, then an unofficial but unmistakable presence in the government he had built until his death in 2015 at the age of ninety-one. Whether this represented a successful succession or a successful extension of influence by other means is a question that reasonable observers have answered differently. What is not in question is that genuine, clean, complete transfers of authority from a founder-autocrat to a successor have been, across the full sweep of history, extraordinarily rare.
The throne always outlasts the person sitting on it. And what comes after is always, in some essential sense, a gamble.
VIII. The Algorithm Will See You Now
Every generation restates the Sovereign CEO fantasy in the technical vocabulary available to it. The Enlightenment used the language of rational administration. The twentieth century used the language of scientific management and central planning. The early twenty-first century used the language of data-driven decision-making and evidence-based policy. Each iteration carried the same essential promise: that the right tools, applied by the right people, could finally resolve the tension between the complexity of governance and the limitations of human judgment.
The current iteration uses the language of artificial intelligence. And it is, by some margin, the most seductive version yet.
The case for algorithmic governance is not made, in its serious form, by people who want to replace democratic deliberation with a chatbot. It is made by people who observe, correctly, that modern governments are drowning in complexity that human cognitive architecture was not designed to manage. The interactions between tax policy and labour markets, between infrastructure investment and urban development patterns, between public health interventions and economic behaviour, between energy policy and geopolitical risk — these are systems of such density and nonlinearity that even the most expert human analysts can only model small portions of them at a time, and the models are always, to some degree, wrong in ways that the modellers cannot fully anticipate.
An AI system, the argument runs, could hold all of these interactions simultaneously. It could process the full available dataset on policy outcomes across jurisdictions and time periods. It could run simulations at a scale and speed that no human team could approach, identifying second and third-order consequences that political intuition routinely misses. It could do all of this without ego, without the desire for re-election, without the cognitive biases that cause human decision-makers to overweight vivid recent events and underweight statistical base rates, and without the factional loyalties that cause human politicians to advocate for policies they know are suboptimal because the alternative would hand advantage to their opponents.
The vision is not entirely fanciful. There are genuine and significant ways in which computational tools have already improved the quality of governance in specific domains.
Estonia’s digital governance infrastructure — built systematically since the late 1990s and now encompassing everything from tax filing to medical records to voting — has demonstrably reduced administrative friction, increased transparency, and created a data architecture that allows policy analysis of a quality that paper-based systems cannot approach. Estonian citizens can see, in real time, who has accessed their government records and why. Tax returns are largely pre-populated by systems that already hold the relevant data, reducing compliance costs to near zero for most citizens. The digital infrastructure has not replaced democratic deliberation — Estonia holds regular elections and has a functioning multiparty parliament — but it has removed significant layers of bureaucratic overhead that previously consumed resources without adding value.
Chile, under Salvador Allende in the early 1970s, attempted something more ambitious with Project Cybersyn — a real-time economic management system designed by the British cybernetician Stafford Beer that would allow central planners to monitor economic indicators across the country and respond to disruptions within hours rather than weeks. It was, by the standards of its era, a remarkable piece of systems thinking. It was also incomplete, underpowered by available technology, and rendered moot by the 1973 coup that ended the Allende government. But it represents an early and serious attempt to use computational tools to address exactly the knowledge problem that Hayek identified — to create an information architecture that could give central planners something closer to the distributed knowledge that markets transmit through prices.
The more recent concept of liquid democracy takes a different approach — not replacing human judgment with algorithmic analysis, but restructuring the mechanisms of democratic participation to allow more flexible and continuous delegation of political authority. In a liquid democracy, citizens do not simply vote for representatives every four or five years. They can delegate their vote on specific issues to individuals they trust to be more informed — a climate scientist for energy policy, a doctor for health legislation, an economist for fiscal decisions — and can revoke that delegation at any time. The result, in theory, is a system that combines the legitimacy of democratic participation with something closer to the epistemic quality of expert judgment, without permanently transferring authority to any fixed elite.
Taiwan’s digital democracy experiments under vTaiwan and the Pol.is platform have demonstrated that large-scale deliberative processes, facilitated by computational tools, can produce genuine consensus on contested policy questions in ways that traditional legislative processes cannot. By mapping the opinion space of thousands of participants and identifying the proposals that command the broadest agreement rather than the most passionate advocacy, the system has resolved regulatory disputes — over ride-sharing, over online alcohol sales, over telecommunications policy — that conventional political processes had left deadlocked. It is not a replacement for democratic institutions. It is a set of tools that can make those institutions work better, by surfacing the distributed preferences of the population with a granularity that representative systems routinely miss.
These examples are genuinely encouraging. They suggest that the choice between efficient autocracy and inefficient democracy is not as binary as the Sovereign CEO fantasy implies — that there are institutional innovations that can recover some of the speed and coherence of centralised decision-making without surrendering the accountability and error-correction mechanisms of distributed governance.
But the algorithmic temptation has its own pathologies, and they are worth examining with the same rigour we apply to its promises.
The first is the problem of objective functions. An AI system optimises for whatever it is instructed to optimise for — and the choice of what to optimise for is not a technical decision. It is a political one, of the deepest kind. If the system is instructed to maximise GDP growth, it will recommend policies that maximise GDP growth, potentially at severe cost to equality, environmental sustainability, community cohesion, and the hundreds of other things that people value and that do not appear cleanly in economic statistics. If it is instructed to optimise for a composite wellbeing index, the composition of that index — the relative weight given to material prosperity versus leisure time versus social connection versus political freedom — is itself a value judgment that cannot be made algorithmically. Someone has to make it. And whoever makes it is, functionally, exercising the kind of sovereign authority that the system was supposed to distribute.
The second problem is more subtle and, in some respects, more dangerous. AI systems trained on historical data will, unless very carefully designed, reproduce the distributional patterns embedded in that data. A system trained on historical policing data will recommend policing strategies that reflect historical policing priorities — which are not neutral. A system trained on historical economic data will model the economy as it has been, with all of its structural inequalities intact, rather than as it might be reorganised. Algorithmic bias is not merely a technical glitch awaiting a technical fix. It is the expression, in computational form, of the political choices embedded in the data — and it is particularly dangerous because it presents those choices as objective outputs rather than contestable decisions.
The third problem is accountability. When a human minister makes a bad decision, there is a person to hold responsible, a decision-making process to scrutinise, and a political mechanism for imposing consequences. When an algorithmic system produces a bad outcome, the accountability chain dissolves into a network of design decisions, training choices, and data selection processes that is genuinely difficult to interrogate even for experts, and essentially opaque to the public. The Sovereign CEO, at least, is a person who can be questioned, criticised, and removed. The Sovereign Algorithm is considerably harder to cross-examine.
None of this means that computational tools have no role in improving governance. They clearly do, and the examples of Estonia and Taiwan suggest that role can be substantial. But the techno-optimist version of the Sovereign CEO fantasy — the idea that sufficiently advanced AI could finally resolve the tension between efficiency and accountability by making governance simultaneously faster and more informed — is making an error that Hayek would have recognised immediately. It is assuming that the problem of governance is primarily a problem of information processing, when the deeper problem is, and has always been, a problem of legitimacy: of who has the authority to make binding decisions about how a society is organised, on what basis that authority rests, and what mechanisms exist to contest and revise those decisions when they prove to be wrong.
Technology can make the processing faster. It cannot resolve the political question of who does the processing, in whose interests, and accountable to whom. That question does not have a technical answer. It has only a political one — which means it requires exactly the kind of messy, slow, frustrating, and irreplaceable deliberative machinery that the Sovereign CEO fantasy has always been trying to transcend.
IX. Efficiency Is the Wrong God to Worship
There is a word that appears, with striking regularity, in every serious argument for the Sovereign CEO model. It appears in the defences of Singapore’s technocracy, in the admiring accounts of China’s infrastructure programmes, in the venture capitalist manifestos about disrupting government, and in the quiet sighs of frustrated citizens watching their legislature fail to pass a budget on time for the sixth consecutive year.
The word is efficiency.
It is worth pausing on this word — not to dismiss it, because efficiency is a genuine and important value in many human contexts, but to examine what it actually means when applied to political systems, and whether the thing it describes is in fact the thing we should be optimising for.
In its original domain, efficiency is a precise engineering concept. A machine is efficient to the degree that it converts inputs into desired outputs without waste. An efficient engine extracts the maximum possible work from a given quantity of fuel. An efficient algorithm produces the correct output in the minimum number of computational steps. The concept has a clear reference point — the theoretical maximum performance achievable under given physical constraints — and a clear measurement methodology. Efficiency, in engineering, is not a matter of opinion. It is a ratio.
When the concept migrates into economics, it becomes somewhat less precise but retains a quantifiable core. An efficient market, in the technical sense, is one in which prices reflect all available information, so that resources flow toward their most productive uses without artificial distortion. An efficient firm produces its output at minimum cost. These definitions are contestable at the margins, and the conditions required for true market efficiency are never fully met in practice, but the concept retains enough precision to be analytically useful.
When the concept migrates further into politics, something important happens to it. It loses its reference point. Efficiency, applied to governance, implicitly assumes that we know what output we are trying to maximise — that there is a clear equivalent of work extracted from fuel, or correct output from algorithm, against which political performance can be measured. But political systems do not have a single output. They have multiple, competing, and frequently incommensurable outputs, weighted differently by different citizens according to values that are not themselves subject to optimisation.
Consider a concrete example. A government faces a decision about how to allocate a fixed infrastructure budget between high-speed rail connecting major urban centres and road maintenance in rural regions. The high-speed rail investment produces higher measurable returns by most standard economic metrics: it moves more people, generates more economic activity, reduces carbon emissions per passenger mile, and connects the nodes of the economy that produce the most output. By almost any efficiency criterion, it is the better investment.
But the rural road maintenance serves communities that are already economically marginalised, that have fewer alternative transport options, that bear a disproportionate share of the costs when infrastructure is neglected, and that are, in a democratic system, equal bearers of political rights whose preferences carry the same formal weight as those of urban commuters. The choice between these investments is not an efficiency calculation. It is a values negotiation — about what kind of society this is, about the obligations of the prosperous to the less prosperous, about whether economic productivity is the primary purpose of public infrastructure or merely one consideration among several.
No Sovereign CEO, however intelligent and however honest, can resolve this negotiation by analysis alone. They can model the economic consequences of each choice with considerable sophistication. They can present the trade-offs with admirable clarity. But the decision of which trade-off is acceptable — of how much rural access is worth how much urban productivity, or vice versa — is not a technical question with a correct answer. It is a political question, and its answer is legitimate only to the degree that it reflects the considered preferences of the people who will live with its consequences.
This is what democratic deliberation, at its best, is actually doing — even when it looks, from the outside, like nothing more than organised confusion. The lengthy and often agonising process by which modern democracies make major collective decisions is not simply inefficiency waiting to be optimised away. It is the mechanism through which competing values are surfaced, contested, partially reconciled, and translated into policy that the losing side can accept as legitimate even though they lost. The legitimacy of the outcome — the fact that people who disagreed with the decision can nonetheless recognise it as the product of a process they had standing in, is not a luxury feature of democratic governance. It is the core product.
Autocratic systems can produce outcomes. Democratic systems produce outcomes and the legitimacy that makes those outcomes sustainable over time. The difference matters enormously, because unsustainable outcomes are not actually efficient, they are costs deferred. A policy that is optimal by every economic metric but was imposed without democratic consent will generate resistance, non-compliance, legal challenge, and political instability that eventually consume the efficiency gains many times over. The history of technocratic reform programmes imposed on populations without adequate consultation is largely a history of technically sound ideas that failed in implementation because the people required to implement them had not been part of the decision to adopt them.
There is a further dimension to the efficiency argument that deserves direct confrontation: the claim, implicit in much of the Sovereign CEO literature, that democratic systems are inefficient because they spend too much time and resources on process rather than outcomes. This claim contains a kernel of truth wrapped in a significant misunderstanding.
Democratic processes are expensive. Elections cost money. Legislatures require staffing, infrastructure, and the considerable time of elected representatives who might otherwise be doing something more directly productive. Judicial review, freedom of information requests, independent oversight bodies, public consultations, all of these consume resources that a purely output-focused accounting would classify as overhead. And they are overhead, in a narrow sense.
But they are overhead that performs a function that is not visible in output metrics until it is absent. Independent courts are not efficient in the sense that they slow down the implementation of government decisions. They are efficient in the deeper sense that they prevent a category of catastrophic governance failures, the arbitrary seizure of property, the selective prosecution of political opponents, the unilateral rewriting of rules to favour incumbent power that, when they occur in systems without judicial independence, impose costs that dwarf any savings from faster decision-making. A free press is not efficient in the sense that it produces friction between government and governed. It is efficient in the sense that it surfaces the information failures and policy errors that, in its absence, compound quietly until they become crises.
James Madison, drafting the constitutional architecture of the United States in the late eighteenth century, understood this with remarkable clarity. The system he helped design was not intended to be fast. It was intended to be survivable — to distribute power in ways that prevented any single faction from doing irreversible damage, at the cost of making it difficult for any single faction to do very much very quickly. The inefficiency was the point. Or rather, the inefficiency was the price of a different and more important kind of efficiency: the efficiency of a system that could absorb serious mistakes without catastrophic failure, and that could sustain legitimate governance across generations and through crises that no founder could have anticipated.
The Sovereign CEO fantasy, at its core, is a fantasy about transcending this trade-off. It imagines a form of governance that is simultaneously fast and wise, simultaneously decisive and accountable, simultaneously powerful and benevolent, that has all the performance advantages of concentrated authority without any of its structural failure modes.
This is not a political programme. It is a wish.
And wishes, however understandable, are not a basis for institutional design, because institutions must be built not for the best case, when the leader is wise and the conditions are favourable, but for the full distribution of cases, including the ones where neither of those things is true.
Which brings us, finally, to the paradox at the heart of the whole enterprise.
X. The Raft Was Never the Inferior Vehicle
The dream does not persist because people are foolish.
That point is worth stating plainly before anything else, because the easiest and least honest response to the Sovereign CEO fantasy is to treat it as a symptom of authoritarian sympathy, or political immaturity, or simple ignorance of history. Most of the people who have entertained it are none of those things. They are people of reasonable intelligence and genuine civic concern who have looked at the performance of democratic institutions across recent decades and arrived at a conclusion that the evidence does not straightforwardly contradict: that these institutions are struggling, that the gap between the problems they face and the solutions they produce is widening, and that something fundamental may need to change.
That conclusion is not wrong. It is incomplete.
The fantasy of the Sovereign CEO contains, at its core, a genuine insight — one that its critics are too quick to dismiss and its proponents too slow to examine. The insight is this: that governance requires judgment, and judgment requires the freedom to act on it, and that political systems which fragment authority so thoroughly that no one can act decisively have their own failure modes, their own costs, their own victims. The person who cannot get housing built because forty separate approval processes stand between the planning decision and the first foundation pour is not suffering from an abstraction. The patient who dies on a waiting list while health administrators argue about restructuring proposals is not a theoretical casualty of institutional inefficiency. The child growing up in a failing school district while legislators debate education reform across a decade of inconclusive sessions is not a philosophical thought experiment.
Democratic inefficiency has real human costs. Acknowledging this is not a concession to authoritarianism. It is a precondition for taking the reform of democratic institutions seriously.
But here is what the fantasy gets wrong — not at the margins, but at its structural core.
It assumes that the primary variable in governance quality is the quality of the governor. That if you could just find, or engineer, or select for the right person — wise enough, honest enough, capable enough, selfless enough — the institutional architecture around them could be simplified, the checks could be loosened, the friction could be reduced, and the results would be better for everyone. It is a theory of governance as a talent problem. Find the talent, and the system follows.
Every section of this piece has been, in one way or another, an argument against this assumption.
The information distortion problem is not a talent problem. It is a structural problem that degrades the quality of information reaching even the most talented leader, because the people below them are rational actors responding to rational incentives, and the rational incentive in a system of concentrated power is to tell the leader what the leader wants to hear. No amount of personal commitment to honest counsel fully counteracts an institutional architecture that makes honesty risky.
The knowledge problem is not a talent problem. It is an epistemological problem that places hard limits on what any central authority can know about the distributed, local, partially tacit reality of the society they govern — limits that do not yield to intelligence, or diligence, or access to better data, because the knowledge in question does not exist in any form that centralised analysis can capture.
The succession problem is not a talent problem. It is a probability problem. Exceptional leaders are, by definition, statistical outliers. A political theory that requires their regular appearance is a political theory that will fail, with high probability, in any given generation — and will fail catastrophically, because the system it has built is calibrated for the outlier and has no graceful response to the mean.
The efficiency problem is not a talent problem. It is a conceptual problem. Efficiency is the wrong criterion for evaluating political systems, because political systems are not optimisation machines. They are legitimacy machines — mechanisms for converting the irreducible disagreements of a plural society into decisions that enough people accept as binding to make collective life possible. A system that produces optimal decisions without legitimate process is not more efficient than democracy. It is solving a different, and easier, problem, and calling the solution governance.
What democracy offers, in place of these things, is something that sounds considerably less exciting and is considerably more important: the institutional equivalent of a guarantee.
Not a guarantee of good outcomes — democracy produces bad outcomes regularly, and anyone who denies this is not paying attention. Not a guarantee of wise leadership — democratic electorates select leaders of wildly varying quality, and the selection mechanisms are influenced by factors that have nothing to do with governance competence. Not even a guarantee of stability, in the short term — democratic transitions can be turbulent, democratic institutions can be stressed, democratic societies can make collective decisions of remarkable foolishness.
The guarantee that democracy offers is narrower and more fundamental than any of these. It is a guarantee of recoverability. Of the structural capacity to identify failure, attribute responsibility, and change course without requiring a revolution, a coup, or the biological mortality of an incumbent. It is a guarantee that the cost of being wrong is bounded — that when the system makes a mistake, as it will, the mistake does not automatically compound into a catastrophe, because the error-correction mechanisms are still in place, still functional, and still accessible to the people who need them.
This is what Winston Churchill was gesturing at, with characteristic imprecision, when he described democracy as the worst form of government except all the others. The joke lands because it captures something true: that the case for democratic governance is not that it is good, but that it is recoverable. That its failures are the kind you can survive and learn from, rather than the kind that hollow out a society’s capacity for self-correction and leave it dependent on the next exceptional leader to emerge from whatever political process the previous one left behind.
The ship and the raft.
A centralised system of government — competently led, coherently organised, insulated from the friction of democratic contestation — is, in the right conditions, a powerful vessel. It is fast. It is impressive. It crosses the distances that matter in the time that matters. In the literature of political philosophy and the data of developmental economics, there are real examples of real ships that made real crossings that slower vessels could not have made. The argument for the ship is not entirely without merit.
But the ship has a property that the raft does not. When it hits a storm — a leadership failure, a succession crisis, an information environment so distorted by the architecture of power that the navigator is steering by a map that no longer corresponds to the water — it does not slow down. It does not list and recover. It does not give its passengers time to reach the lifeboats. It founders, and it takes everything aboard with it.
The raft is not romantic. It is uncomfortable, it is slow, it is operated by committee and moves by argument and frequently travels in the wrong direction for extended periods before the argument is resolved. It has no captain in the relevant sense — only a constantly renegotiated consensus about which way is approximately forward. It is wet. It requires constant maintenance by people who would often rather be doing something else. It is, by almost every aesthetic standard, inferior to the ship.
But the raft has a property that the ship does not. It cannot sink. Its distributed structure, its redundancy, its lack of any single point of catastrophic failure — these are not design limitations. They are design achievements, arrived at through the oldest and most expensive form of engineering: the accumulated lesson of everything that has gone wrong before.
The people who built democratic institutions were not naive about human nature. The authors of the Federalist Papers were not optimists about the wisdom of crowds or the virtue of politicians. They were, if anything, thoroughgoing pessimists about both — which is precisely why they designed a system that did not require either. They assumed that leaders would be self-interested, that majorities would be tyrannical, that any concentration of power would eventually be abused, because the historical record gave them no reason to assume otherwise. And from those assumptions they built something that was not designed to perform well, in the short term, under ideal conditions.
It was designed to survive.
In an era of resurgent authoritarianism, of democratic backsliding across four continents, of algorithmically-amplified frustration with institutions that were never designed to be loved — only to endure — this design principle deserves more credit than it receives. The frustration with democratic governance is real and legitimate and should be taken seriously as a signal that reform is necessary. What it should not be taken as is evidence that the underlying architecture is wrong.
The Sovereign CEO fantasy will persist. It will be restated, in each generation, in whatever vocabulary that generation finds most compelling — as philosopher-kingship, as enlightened despotism, as scientific management, as data-driven governance, as artificial intelligence. Each restatement will contain a genuine insight and a fatal flaw, and the fatal flaw will always be the same one: the assumption that the right person, given enough authority, can transcend the structural problems that are not problems of persons at all.
The raft was never the inferior vehicle.
It was always the only one designed for the actual journey — the one that does not end when conditions are ideal, but continues through every storm that history, with its boundless creativity for disruption, has not yet thought to send.
