The Reimagined State
Leadership in the age of AI requires a state designed for a new era. What would government look like if we started again today?
CHAPTER 01
The state has the data but not the whole picture
It can count the population, record births and deaths, measure economic activity and monitor the spread of disease.
But knowledge and data are divided across departments, agencies and regulators. Each institution sees the part of the world for which it is responsible. Few can see how those parts interact. Fewer still can see, in real time, how the actions of government are changing them.
A prime minister or president can decide to build a railway, reduce hospital waiting times or make the country a leader in artificial intelligence. Money can be allocated, legislation passed and responsibility assigned. But months later, it may be difficult to answer some basic questions. Is the priority on track? What is preventing progress? Which institutions need to act? Have the original assumptions changed? What needs to happen next?
The answers usually exist somewhere, but government lacks the ability to assemble them into a shared picture of what is happening.
This is the central paradox of the modern administrative state.
The question is:
Why?
CHAPTER 02
The modern administrative state was built for industrial-era constraints
As populations grew, economies expanded, and governments assumed responsibility for education, infrastructure, welfare and public health, older forms of administration became inadequate. States responded by building specialised departments, professional civil services, formal procedures and clear hierarchies of authority.
Three constraints shaped what institutions could know and do.
01
Intelligence was scarce
Understanding a problem required people to find information, read and compare it, draw on institutional memory and advise those with the authority to act.
02
Coordination was expensive
People had to transmit information, reconcile institutions, assign actions, chase progress and supervise execution.
03
Change was slower
Governments could only operate through annual budgets, periodic reviews and strategies intended to last for years.
Information moved upwards. Authority moved downwards. Government divided complexity among specialised institutions and attempted to reconstruct the whole through reports, committees and central coordination.
It was a rational, and enormously successful, response to the conditions in which it developed – but the result was an operating model built around the limits of human intelligence, human coordination and institutional time.
CHAPTER 03
AI changes the constraints government must operate within
Government is one of the world’s largest consumers of cognitive and coordination labour. Much of its work consists of finding information, reading it, comparing it, remembering it, transmitting it, reconciling institutions and ensuring that agreed actions take place. AI does more than make this work faster.
01
Generative AI changes the economics of knowing
Systems can search, synthesise and analyse information at a scale no human institution could previously achieve. They can compare evidence, model options, retrieve institutional memory and monitor developments continuously.
02
Agentic AI changes the economics of coordination and action
Agents can pursue bounded objectives across multiple steps: investigating problems, monitoring risks, reconciling information, coordinating actions, following up and escalating when intervention is required.
03
More capable AI could change the economics and pace of change itself
As AI systems become more capable, they may accelerate scientific discovery, technological development and economic adaptation. Government will not merely gain faster tools. The world it governs may begin to change faster too.
The argument does not depend on a particular prediction about how AI will develop, or in particular on the development of artificial general intelligence (AGI). Under modest progress, AI makes institutional reform possible. Under rapid progress, it makes reform urgent.
CHAPTER 04
What becomes scarce when intelligence becomes abundant?
AI can produce a thousand analyses and model a thousand policy options, but a government must still choose between them, secure consent and accept responsibility. AI can identify where a power station should go. It cannot create the land, grid connection, engineers, capital or political authority required to build it.
As intelligence and the capacity to coordinate become more abundant, what remains scarce becomes more important: Attention. Judgement. Authority. Political capital. Legitimacy and trust. Physical capacity. Time.
AI therefore makes clearer what political leadership is for: choosing between legitimate priorities, resolving trade-offs and turning intelligence into action. AI can increase that human agency without removing democratic constraint. The same systems that give leaders greater capacity to act can make their decisions more transparent, auditable and accountable.
Abundant intelligence expands what is possible. Political leadership decides what becomes real.
CHAPTER 05
The Reimagined State uses machine intelligence to increase human agency
Adding AI to the existing machinery of government is not enough. Without redesign, AI can make old processes faster and existing silos more capable, but it will not transform the state.
The Reimagined State should be designed around six principles that will make government more capable:
01. See the whole
Government needs a live operating picture of its priorities: the trajectory, the money committed, the institutions involved, the dependencies between them, emerging risks and the uncertainty surrounding the evidence. Where the administrative state largely reports what has happened, the Reimagined State should be able to see the trajectory while there is still time to alter it. Better information should allow the centre to intervene less often and more effectively.
02. Organise around outcomes
The world contains problems, but government encounters each one through multiple departments holding different information, incentives and responsibilities. The Reimagined State should connect knowledge based on the outcomes it is trying to achieve. This does not require every institution to be reorganised or controlled from the centre. Expertise can remain distributed, but government should centralise the foundations that allow it to work together – digital identity, interoperability, shared standards, institutional memory and a common operating picture – while enabling experimentation, action and the flow of intelligence across the system.
03. Make government agentic
Agentic AI makes it possible to identify, act on, track and escalate risks without repeated manual intervention at every stage. The design of these agents is important. Its permissions, escalation rules, available actions and audit requirements determine how public power operates. Digital architecture becomes a form of constitutional infrastructure: administrative mandates must be bounded, explicit, auditable and revocable, while political mandates remain human.
04. Learn continuously
The administrative state operates periodically: predict → approve → implement → review. The Reimagined State adapts through feedback: decide → act → observe → learn → adapt. Spending, regulation, delivery and policy can incorporate evidence from implementation as it emerges, with agentic systems then identifying problems, tracking agreed interventions and measuring whether they worked. As technological change accelerates, government needs adaptable systems and continuous awareness of the technological frontier to learn fast enough to remain relevant to the world it governs.
05. Elevate human judgement
As intelligence and coordination become more abundant, government should organise itself around what remains scarce: human attention, judgement, authority and responsibility. Machines will increasingly search, analyse, monitor, model, remember, coordinate and execute bounded routine actions. Human beings will choose, judge, negotiate, invent, persuade, empathise and take responsibility. This implies a different civil service and leader’s office: fewer people moving information through hierarchies; more operators, experts, engineers, negotiators and empowered frontline professionals. Scarce political attention can be concentrated on the decisions where judgement and authority matter most.
06. Make capability accountable
Greater capability requires stronger accountability. AI systems should be auditable, their mandates visible and revocable, and their actions open to challenge, appeal and redress. Properly designed systems can also make responsibility clearer: recording what government knew, what advice was given, who made the decision and what happened afterwards. Machine authority may be delegated, but human beings must remain responsible for how it is exercised.
Taken together, these principles describe a state with greater agency. A state able to understand what is happening, organise around outcomes, follow decisions through, learn from their effects and concentrate human attention where judgement matters most.
CHAPTER 06
Capability becomes democratic authority
State capacity is often treated as a technocratic concern. This is wrong - it is fundamentally political. Citizens elect governments to make choices real. When governments repeatedly promise change but fail to deliver it, the damage extends beyond the incumbent. Confidence in the ability of politics to achieve anything begins to weaken.
Failure cycle:
Promise → failure → distrust → less room for reform → caution → further failure
Capability cycle:
Action → delivery → trust → authority → greater ambition → greater capability
AI matters politically because it can increase the agency of government and demonstrate its capacity to deliver.
Intelligence is not power until it becomes action
The global competition over AI is currently measured in models, chips, compute, energy, capital and talent. But possessing advanced intelligence will not by itself make a country powerful.
AI-generated discoveries must become medicines. Designs must become factories. Energy plans must become power stations. Military intelligence must become deployable capability. Better policy ideas must become better public services.
As intelligence becomes more widely available, the ability to translate it into action will become increasingly decisive.
That depends on energy, infrastructure, industry, people and land. It also depends on institutions capable of bringing them together.
The geopolitical competition of the AI age is, at its core, a race to convert intelligence into power. Governments are part of that conversion mechanism.
The race for power in the AI era is also a race to reform the state.
CHAPTER 07
Build for what remains human
The administrative state was designed for a world in which intelligence was scarce, coordination was expensive and change was comparatively slow.
AI challenges all three conditions. But it cannot decide what a society should value, which trade-offs it should accept or on whose authority power should be exercised.
The purpose of the Reimagined State is not to remove human beings from government. It is to reorganise government around the things for which human beings remain indispensable.
We should not spend the next 20 years placing increasingly powerful intelligence inside institutions designed around its scarcity. We should build the institutions that abundant intelligence makes possible – and that human judgement makes legitimate.
The countries that do so will not merely possess the most intelligent machines.
They will have the most capable states.
The work starts now
The opportunity to reimagine the state is here. Leaders should not wait for AI to transform the world around government; they should start transforming government itself – redesigning the centre, building new institutional capabilities and putting abundant intelligence at the service of human judgement and political leadership. Governments can start that work today.
Explore