The question leaders should be asking themselves is not whether AI is a good thing or a bad thing. Rather, it is simply the thing: the defining technological reality of our age. Their task is to understand it, shape it and ensure that their countries have the mechanisms to both harness and control it.
AI is often compared with the Industrial Revolution, but even that comparison risks understating the scale and speed of what lies ahead. The previous revolution transformed our world over generations; AI will do so within a decade. The speed is without historical precedent. Its impact will be more cross-cutting than any technology before it. New capabilities are emerging at an extraordinary pace, with profound implications for economies, education, science, labour markets, public services and national security.
The opportunities are equally immense. For developed countries, AI productivity gains can help break the cycle of rising taxes and spending that too often fails to improve public-service outcomes. For developing countries, AI can dramatically expand access to scarce expertise, allowing them to accelerate development in ways that were previously impossible. AI will transform health care and improve quality of life; revolutionise science and accelerate the pace of discovery; and democratise access to high-quality education around the world.
But the risks are equally profound and growing, and so too is public concern. AI will disrupt labour markets. It is likely to concentrate extraordinary amounts of technological and economic power. AI systems are already dramatically lowering the costs and increasing the capabilities of malicious actors to conduct cyber-attacks autonomously. There is growing concern about how AI may enable new biological threats too. Countries will be forced to confront serious questions about how they can best oversee, control, or influence autonomous and highly advanced AI systems.
But there is another risk that leaders must take just as seriously. Countries also suffer profound and compounding disadvantage if they fail to absorb major technological revolutions.
History shows what is at stake. During the Industrial Revolution, the countries that prospered were those that embraced technological change and adapted to the new world. Countries such as Japan and Germany actively sought out new technologies, learned from those at the frontier, and built the institutions and capabilities needed to put them to use. Others were slower to adapt and fell behind, suffering not just an economic setback, but the compounding consequences for living standards, health care, education and opportunity for improvement. When the world changes, government must change with it.
Today’s political leaders must do the same. They must make hard choices that will significantly alter their countries’ trajectories. However, the institutions of the modern state meant to support political leaders in making these choices are simply not equipped for what is coming.
That is why, when leaders ask me for advice on how to navigate the age of AI, I talk about the need to reimagine the state. Countries cannot protect themselves from AI by ignoring it. The only meaningful path is to engage with it, building institutions capable of operating at the speed and complexity of the world now emerging, and harness AI to transform how the state itself works.
AI can give leaders a far clearer picture of what is happening across government, bring together information and expertise that today sit in silos, and allow the centre to identify problems, track delivery and intervene far more effectively. But it cannot decide what a country should prioritise or make the hard political choices that will shape its future. Political leaders remain responsible for that.
Getting this right means a state that can see more clearly, act more effectively and adapt more rapidly; that can solve increasingly complex problems, deliver public services that match the innovation and responsiveness people experience elsewhere in their lives, and give citizens renewed confidence in democratic governments’ ability to get things done.
This paper shows how leaders can seize this opportunity to deliver a better future for their country if they meet this moment and make the bold transformations it requires.
Tony Blair, Executive Chairman
Strong political leadership has always played a decisive role in determining which states succeed and which do not. This is especially true during times of disruption. Decisions made during these periods are often urgent and based on limited information, and their outcomes can have long-lasting effects on a country’s future trajectory.
Today, every leader is facing this scenario. Artificial intelligence will reshape every aspect of our world, from how wars are fought and labour markets operate to how science and research are conducted. Capabilities at the frontier are advancing almost weekly. Leaders are therefore confronting a technology whose consequences will extend across almost every area of government.
Yet while AI’s potential is widely understood, how it will evolve is not. Leaders know they need to act, but what to do and how much to do are often less clear. They must decide how much energy and compute to make available, how to prepare their economies and workforces, how to secure increasingly digital states, and how far to intervene in the development and deployment of the technology. In the West, they must make these choices amid a growing “techlash” against AI – but while doing less might help politically in the short term, it may carry significant long-term impacts on a country’s stability and security.
Political leaders cannot navigate this alone and are highly dependent on the institutions that support them, both to provide the evidence and expertise needed to understand the issues and potential decisions, and to implement those decisions. Yet the challenges that leaders face today often require multiple institutions, departments or regulators to work together, drawing on different strands of evidence and operating under different regulations. This increases the complexity of coordination and delivery. Leaders’ ability to navigate the age of AI will therefore depend on both the quality of the choices they make and on the capacity of the institutions around them to inform and implement those choices.
Technology, and especially AI, will be key to transforming how a state’s institutions work and improving their capacity to deliver. AI can help by enabling government to synthesise information across institutional boundaries, retrieve and apply institutional knowledge, and sustain coordination around shared outcomes. But technology alone is not enough. A new operating model is also needed that can combine political decision-making and authority with institutional expertise and is supported by abundant machine intelligence. Done well, this would give leaders greater capacity to understand the choices before them – and their potential consequences – while giving institutions greater capacity to translate those decisions into action.
We call this new model the Reimagined State. It uses AI and other technologies to transform how knowledge, coordination and decisions flow through government, enabling both leaders and institutions to operate with greater agency and effectiveness. By breaking down informational and institutional silos, government can connect knowledge and data across institutional boundaries, improving its understanding of problems and its ability to coordinate action.
The centre of government – the institutions that directly support the head of government and coordinate priorities across the state – sits at the heart of this transition. It connects political decision-making to the state’s distributed knowledge, expertise and capabilities, and helps translate those decisions into sustained delivery. To perform this role effectively in the age of AI, the centre itself must be transformed. Today’s centres were designed for a world in which information was scarce, coordination was labour-intensive and leaders relied on periodic reporting to understand what was happening across government. AI and other technologies create the opportunity to fundamentally change this, allowing the centre to maintain a more continuous view across government, draw on distributed institutional knowledge and sustain coordination around political priorities.
However, the Reimagined State is not a model for concentrating unconstrained power at the centre or weakening the institutions of democratic government. Greater state capability must be matched by greater transparency and accountability. As government becomes better able to connect information, coordinate institutions and follow decisions through, it should also become easier to understand how decisions were reached, what evidence informed them, where authority was exercised and who is responsible for the outcome. This new model therefore strengthens democratic governance by using technology to make government simultaneously more capable, more transparent and more accountable.
The states that succeed in the age of AI will be those that transform their institutions, build the digital foundations on which AI-enabled government depends and use these capabilities to fundamentally change how the state operates. This paper sets out how these reforms provide a pathway towards the Reimagined State.
Recommendations
The Tony Blair Institute for Global Change (TBI) has developed a three-part framework for building the institutions, digital foundations and capabilities governments need to navigate the age of AI.
1. Prepare institutions to deliver in the age of AI
Effective leaders need effective institutions capable of translating political direction into action. In the age of AI, governments need to become more flexible and adaptive without sacrificing the stable institutions that preserve expertise and long-term capability. The objective is not to reorganise the whole state around temporary missions, but to make it easier to bring together and reconfigure capabilities around changing priorities.
Build outcome-oriented institutions capable of learning and adapting. Governments should combine stable institutions that preserve expertise and continuity with flexible structures that can form and adapt around changing priorities. Leaders should concentrate political attention on a limited number of outcomes, align authority and resources behind them, and use evidence from implementation to adapt strategies and institutional arrangements as circumstances change.
2. Build the state’s digital foundations
The best AI-enabled states will be those with the strongest digital foundations. Governments must invest in building and maintaining the shared digital foundations that enable data to be gathered and shared, institutional memory to be digitalised, and the state to operate securely and resiliently.
Invest in shared digital and data foundations. Governments should build common infrastructure that lets people securely prove who they are, enables different systems to work together and allows information to be shared safely across institutions. Departments should remain responsible for ensuring their data are accurate and properly understood.
Create shared institutional memory across government. Governments should preserve significant decisions, evidence and lessons as governed institutional knowledge that persists as people, teams and structures change. This memory should be versioned and traceable to authoritative records, preserving the context and decision lineage needed for officials and AI systems to understand what was known, decided and learned over time.
Embed security and operational resilience into the state’s foundations. Governments should build modular and adaptable systems that remain secure and operational when models, suppliers or individual components fail or change. Common interfaces, continuous security testing and fallback arrangements should ensure that greater technological dependence does not create new points of state failure.
3. Deliver the AI-enabled state
Strong institutions, supported by robust digital foundations, must work to deliver the Reimagined State, moving from piecemeal pilots to whole-of-government transformation. As the transformation unfolds, it is essential that governments continue to transform the state in ways that maintain accountability and transparency.
Create a common route from experimentation to system-wide adoption. Governments should establish common pathways for AI applications to move from experimentation through evaluation and deployment to scale. Shared procurement, evaluation and technical requirements should allow successful capabilities to spread across government while duplicated, unsafe or ineffective systems are stopped.
Build AI-enabled strategic intelligence and simulation capabilities at the centre. Governments should use AI to integrate information across institutions, identify emerging risks and dependencies, and test how major choices could interact across the state. Combining AI with forecasting, modelling, operational data and expert judgement can expand the range of possibilities leaders are able to consider while keeping political decisions in human hands.
Introduce bounded agents to sustain government action. As the technology matures, governments should use AI systems that can carry out specific tasks within clearly defined limits, such as tracking commitments, requesting updates and initiating approved workflows. Their remit should remain explicit, bounded, auditable and revocable.
Ensure AI-enabled government remains aligned and democratically accountable. More powerful institutions require stronger mechanisms for scrutiny and control. Governments should preserve clear lines of authority, ensure decisions are traceable and human responsibility is maintained, and ensure that greater use of machine intelligence strengthens rather than obscures democratic accountability.
The aim is a wholly different operating system of government. One that can move at the same pace as the world around it and, crucially, at the pace its citizens expect. The countries that respond to this moment of change with courage and ambition have the opportunity to restore confidence that government can get things done, build public services that are more responsive to people’s needs, and replace a politics of constant announcements with one of sustained delivery and continuous improvement.
Chapter 1
Good governance depends on strong and effective political leadership, as well as on strong, agile and effective institutions. This is especially true during times of disruption, when leaders must navigate significant change with limited options, incomplete data and long-lasting consequences. While political leaders must own the decisions they make and chart a clear path forward, it is ultimately institutions that deliver.
The age of AI will demand both leaders who understand the importance of this moment and institutions prepared to deliver initiatives that may require societal-level mobilisation and significant investment. Countries are already navigating a complex and rapid transformation of their economic, industrial, energy, geopolitical and security strategies. The increasing interconnection between these domains means that a small change in one area may have significant impacts elsewhere; the nonlinear nature of these problems can be hard for any single individual or institution to grasp on their own. The strategies that a country develops to address this challenge will therefore come to define its future strength and agency. Those that invest now in improving their governance, institutions and digital foundations will be best prepared to successfully navigate the age of AI.
However, many governments are entering this transition at a time when they are already struggling to deliver change at the pace and scale citizens expect. This growing gap between political ambition and visible results is contributing to a wider crisis of confidence in government. The Organisation for Economic Co-operation and Development’s latest cross-national survey found that only 40 per cent of respondents had high or moderately high trust in their national government, while 43 per cent reported low or no trust.[_] As governments repeatedly struggle to turn political commitments into tangible improvements, dissatisfaction risks becoming directed not simply at particular policies or incumbents, but at the capacity of the system itself to respond.[_]
What lies behind this problem is not necessarily a lack of state capability, but the way that capability is organised. Modern governments have accumulated substantial expertise across departments, agencies, regulators and delivery bodies, but that expertise sits within institutional boundaries. Building infrastructure, securing energy supply, updating a defence strategy or preparing an economy for AI increasingly requires many of these institutions – each holding different information, exercising different powers and operating under different rules – to move in the same direction. Even if every institution carries out its work well, the outcome can still fall short because no single institution is accountable for the outcome and, in many instances, institutional heads may have priorities that compete with those of the political leadership they serve. The challenge for leaders today is to better connect the distributed capabilities across government into a coherent direction.
History demonstrates that this is possible. Periods of national transformation have often involved leaders building and adapting organisations around the missions that they sought to achieve.[_]
In the early 1960s under Park Chung-hee, South Korea established the Economic Planning Board, concentrating responsibility for economic planning, budgeting, foreign economic cooperation and statistics in a powerful central agency.[_] The board connected economic strategy with resource allocation and implementation across government. Over subsequent decades, South Korea progressed from a low-income agrarian economy to a high-income economy with globally competitive industries in advanced manufacturing and technology.[_]
In the years around independence, Singapore had to contend with widespread slum housing, high unemployment and an economy constrained by limited industry.[_] Its leaders created institutions with clear mandates to address these constraints, including the Housing & Development Board (HDB) and the Economic Development Board (EDB).[_] The EDB became a central institution in Singapore’s drive to attract investment and develop export-oriented industries, while the HDB undertook a large-scale public-housing programme. These institutions gave strategic priorities organisational form, dedicated capability and continuity over time.
Estonia provides a more recent example. After restoring its independence in 1991, the country had to rebuild and modernise state institutions and infrastructure with limited resources. Estonia’s leaders treated digitalisation as a core element of state development, building common foundations including digital identity, interoperability infrastructure, connectivity, digital skills and cyber-security.[_] Sustained investment in these capabilities helped transform Estonia into one of Europe’s most advanced digital states.[_]
Each example demonstrates how states, supported by strong leadership with a clear vision for the future, can build new institutions that deliver, driving change and prosperity in the process. Navigating and succeeding in the AI era will require a comparable institutional shift in every country around the world.
Chapter 2
State power and influence have always depended on information, and the institutions and technologies used to convert it into action.[_] Max Weber, whose work laid the foundations for the modern study of bureaucracy and its role in the state, identified this as the source of bureaucratic authority: officials exercise power through technical expertise and through the official knowledge that a bureaucracy accumulates in the ordinary course of its business and preserves in its files.[_] That combination is what has allowed governments to administer increasingly complex societies. It also means that what a state can do has always been bounded by what its institutions can know, and by how quickly that knowledge can be assembled and acted upon.
Technology has repeatedly enhanced the capacity of the state by making it faster and cheaper to record information, communicate, coordinate activity and deliver services. Each advance has enabled governments to perform existing functions more effectively while creating opportunities to organise themselves and serve the public in new ways. Computerisation transformed record-keeping and administration; e-government put information and services online; and digital government began to redesign services around users and establish shared foundations such as digital identity, interoperable systems and common platforms.
New technologies have expanded what the state can do
Source: TBI analysis
None of these technologies made bureaucracy obsolete. Instead, each has expanded what bureaucratic institutions can know, coordinate and deliver, allowing governments to manage growing levels of complexity more effectively.[_]
AI represents the next, and most consequential, step in this progression. Yet, unlike other technological transformations, AI will require substantial institutional changes. The industrial model for government, prevalent across the world today, developed when processing information was slow and expensive, and when knowledge accumulated within institutional boundaries. This model has served governments well, but it is not able to deliver effectively in the AI era, where the pace of change is rapid, decisions require quick responses and the cost of intelligence is declining. AI makes it possible to access and synthesise information on an unprecedented scale, enabling a state that can see across institutional boundaries, organise around outcomes, learn continuously, and translate political direction into sustained action. We call this the Reimagined State.
From the industrial operating model to the Reimagined State
Source: TBI analysis
The Reimagined State is not an industrial state equipped with more advanced tools, nor is it a post-bureaucratic state run autonomously by machines. The functions and institutions of government remain, but the relationships between knowledge, authority and action are redesigned.
As machine intelligence becomes more abundant, the capabilities that remain scarce become more important: political attention, judgement, authority, legitimacy, trust, physical capacity and time. AI can identify a conflict between government objectives, but it cannot decide which should prevail. That requires political leadership. The objective of the Reimagined State is therefore to automate intelligence to elevate judgement, concentrating scarce human capacity on the choices where negotiation, persuasion and responsibility matter most.
This makes higher-agency leadership possible. But it cannot be achieved through technology or individual determination alone. It requires an institutional mechanism capable of connecting political direction to the distributed knowledge, authority and capabilities of the state – and of sustaining that connection as implementation produces new evidence and circumstances change. That mechanism is the centre of government.
Chapter 3
The Reimagined State represents a transformation of how the whole of government operates. No individual department can deliver it alone. A department can adopt AI and improve its own systems, but it cannot establish the common standards, shared foundations and cross-government coordination required for knowledge and capabilities to move across institutional boundaries. Without these, AI risks producing more capable individual institutions that remain disconnected from one another.
The transition must therefore be led from the centre. Whether organised around Downing Street, the White House, the Kantei (Japan’s prime minister’s office) or another executive institution, the centre of government sits between political leadership and the institutions through which leaders govern. Its purpose is to ensure that the priorities set by political leaders are translated into action and ultimately into results.
What is the centre of government?
The centre of government is the institutional embodiment of the political leader, whether a president, prime minister or equivalent. It serves as the connective tissue between the leader and the rest of the state, and is responsible for ensuring that political ambition is translated into whole-of-government action and delivery. Its authority derives not from its size – it is typically small relative to government departments – but from the mandate of the leader.
The centre takes different forms in different political systems. It may be concentrated in a single executive office or spread across several institutions, and its name, size and responsibilities vary between countries. What defines it is its proximity to the leader and its role in coordinating and driving the government as a whole.
To build the Reimagined State, the centre must drive two connected transformations.
First, it must transform how it performs its own role. As the pace of change accelerates and the demands on political leadership grow, the capacity of the centre becomes increasingly important. Centres of government have always connected political direction to the machinery of the state. Their role is to institutionalise political leadership: concentrating attention on priorities, coordinating action across institutions and maintaining pressure on delivery after the initial decision has been made.
AI can transform the centre’s capacity to perform this role. It can provide a more complete and current view of government, reduce the human effort required to assemble and interpret information, and help sustain coordination between formal interventions. In particular, it can strengthen each stage of the cycle through which political direction is translated into delivery:
Decide what matters most. AI can help leaders understand how priorities interact and where they depend on the same infrastructure, and expand the range of consequences they can examine under different assumptions. However, it is still up to the government to determine what it values and whose interests should prevail.
Act on decisions across government. Once a direction has been set, the centre must connect it to the money, legislation, institutions and capabilities required for delivery. AI can map dependencies, synthesise institutional positions and identify unresolved disagreements or gaps in responsibility. This makes it easier to coordinate action across institutional boundaries while departments and public bodies retain responsibility for delivery.
Observe implementation as it unfolds. Rather than relying principally on periodic reporting, AI can analyse operational and financial data alongside programme reports, evaluations, correspondence and frontline feedback. This can provide a more current view of trajectory, expenditure, dependencies and emerging risks, while directing human attention towards anomalies and questions requiring investigation.
Learn from implementation. AI can help government compare results with the assumptions underlying a decision, connect emerging problems to relevant institutional knowledge and preserve lessons that might otherwise remain dispersed across documents, systems and individuals. This allows the centre to understand what is actually happening, rather than relying on what was expected to happen.
Adapt while there is still time to affect the outcome. As evidence changes, the centre can help institutions adjust plans, redirect resources or escalate choices requiring renewed political authority. Over time, AI agents operating within clearly defined limits could monitor commitments, request updates and initiate approved workflows between formal interventions. Their authority must remain explicitly delegated, bounded, auditable and revocable.
Second, the centre must lead the wider transformation programme across government. This does not mean that the centre should determine how every institution uses AI or attempt to deliver the transformation itself. The centre’s role is to provide the common infrastructure, rules, capabilities and coordination that individual institutions cannot create alone, while identifying shared barriers, reducing duplication and enabling successful approaches to spread across government.
These foundations include secure digital identity, mechanisms for exchanging data lawfully, shared technical and governance standards, common services, accessible institutional memory, and a common operating picture of government priorities and performance. They should be established once and made available across the state. This is crucial to improving how citizens and businesses experience the state. Reliable digital identity and permissioned data exchange can allow information to be provided once rather than repeatedly requested by different institutions. Shared services can make government more proactive and joined up, while preserving access to human judgement when circumstances require it. The internal transformation of government and the outward transformation of public services therefore depend on the same underlying architecture.
Figure 3 brings these elements together. Political leaders define priorities; the centre sustains a continuous cycle of decision, action, observation, learning and adaptation; and institutions contribute specialist knowledge and undertake delivery. Beneath them sit the shared foundations that connect the system internally and support better services externally.
How the Reimagined State works
Source: TBI analysis
Chapter 4
Building a new, AI-enabled Reimagined State that can help leaders better navigate the age of AI will require action across three main domains:
Prepare institutions to deliver in the age of AI: Organise government around a limited number of strategic outcomes, with clear political direction, ownership, authority and resources, and enable institutions to learn and adapt as implementation unfolds.
Build the state’s digital foundations: Establish the common identity, data, interoperability, institutional-memory and security capabilities required for information and knowledge to move safely across government.
Deliver the AI-enabled state: Create a common route for AI applications to be evaluated, adopted and scaled across government, while building new intelligence and coordination capabilities at the centre and maintaining democratic accountability.
This effort should be primarily led and driven by the centre, while departments and public bodies remain responsible for applying it within their own mandates and delivering results.
Prepare Institutions to Deliver in the Age of AI
Government institutions are traditionally designed to perform enduring functions consistently, such as collecting taxes, administering benefits, regulating markets or providing public services. Stability, defined responsibilities and repeatable processes are essential to this work. But many of the outcomes that matter most to political leaders, particularly in the AI era, cut across established responsibilities, require multiple forms of intervention and unfold under conditions of uncertainty. As a result, there is growing interest in designing new innovation bureaucracies that can fulfil both functions simultaneously, by being flexible when needed yet backed by stable pillars of expertise.[_]
For example, across many priorities, successful delivery of an outcome is likely to require a new organisational unit that draws on capabilities across institutions. To keep up with the pace of change in the age of AI, governments must be able to create, adapt and wind down organisations with far greater flexibility than they can today. This is primarily an organisational problem rather than a technological one. Bureaucracies have historically been resistant to change and subject to competitive dynamics, with institutions often prioritising their own mandates, resources and ways of working over the broader priorities of government.[_] Where expertise and institutional knowledge remain highly concentrated within individual organisations, these dynamics can be difficult to disrupt.
Yet AI can also help to overcome some of these barriers by making it easier to access and exchange knowledge and information across institutional boundaries. This can reduce the cost of bringing together expertise from across government, allowing new teams to form around outcomes rather than being constrained by existing organisational structures.[_] However, this kind of adaptive and flexible approach is only possible if the state also maintains stable institutions capable of preserving expertise and learning over time. This is especially important where the state must make long-term decisions that will impact the country’s trajectory for generations, for example on significant infrastructure investments or national security.
The important challenge for leaders will be to understand which problems require a more flexible approach and where stability must remain. In both cases, it is essential that institutional structures reflect the nature of the task.[_] In some cases, an existing institution can be given a renewed mandate, a different operating model and the authority required to lead delivery. In others, government may need to create a purpose-built organisation or cross-government mission structure that brings together the relevant expertise, powers and resources. Where existing arrangements have become an obstacle to delivery, responsibilities may need to be transferred and obsolete processes, functions or organisations retired. Institutional reform should not always mean adding another body to the state. Governments therefore need a capacity for institutional renewal and, where necessary, creative destruction.[_]
The start of this institutional transformation must ultimately be led and enabled by the centre of government. Individual institutions can reform how they operate, but they cannot redesign the wider configuration of the state. Only the centre can set priorities across government, determine how authority and resources should be distributed between institutions, resolve competing mandates and, where necessary, create new structures or retire obsolete ones.
Recommendation: Build outcome-oriented institutions from the centre capable of learning and adapting.
This begins with political clarity. While AI will enable the centre to monitor activities across government more closely, political leaders’ attention is likely to remain scarce. It is therefore essential for the centre of government to distinguish between different levels of political priority. A small number of leader-level priorities should receive sustained oversight, while a wider portfolio of government outcomes can be monitored at a lower intensity. These priorities should then shape the institutional configuration around them: determining which capabilities need to be brought together, where authority should sit and where changes to the machinery of government are required.
Once a priority has been set, the institutions responsible for delivering it need the authority and resources to act. Funding should be considered across the institutions responsible for each priority, with multi-year allocations, where appropriate, to provide sufficient certainty for planning over the government’s term. Decision rights should also be defined in advance, specifying which institutions have authority to act, which matters require collective agreement and which should be escalated to the leader. Where delivery stalls, the centre should be able to intervene by deploying limited, flexible funding or specialist capability, convening the responsible institutions, resolving regulatory or institutional barriers, and recommending the reallocation of resources. Intervention should be targeted and temporary: its purpose is to overcome barriers and restore delivery, not to absorb permanent responsibility from departments.
As the state delivers, it learns, and those lessons need to flow back to the centre so both can adapt to what implementation reveals. Departments and frontline institutions will often be the first to identify where an intervention is working, where assumptions are proving incorrect or where new barriers have emerged. Regular stocktakes led by the head of government should link this information to decisions. Rather than simply reporting activity, they should assess whether outcomes are changing, identify unresolved choices and agree the actions required to restore progress.
This cycle is likely to keep accelerating. For lessons to be useful, however, institutions need a clear way to judge whether they are making progress towards the outcome they have been given. Each priority therefore needs a clear set of indicators against which progress can be assessed. But those indicators cannot become the mission itself. Institutions adapt to what they are measured against, creating the risk that they improve an indicator while moving further away from the underlying outcome. Quantitative measures should therefore be considered alongside evaluation, citizen experience and frontline professional judgement, and revised when they no longer provide useful evidence of progress.
This creates a continuous governing cycle at the heart of the Reimagined State, where the centre provides political direction, authority and resources, while institutions deliver, generate knowledge and feed evidence back. As that evidence changes, both must be able to adapt while ensuring government remains aligned with the public purpose it set out to achieve.
Build the State’s Digital Foundations
The flexible institutional model described requires strong digital foundations to function. To build a better digital and AI-enabled state, the government will need to invest in creating shared digital foundations through which people, institutions and AI systems can access information, establish identity and authority, and interact securely. Many of the foundations required are already well known. They include reliable data and digital records, digital identity, secure data exchange, common standards and reusable services. These are already important, but AI makes them even more so; without these foundations, even the most capable models will remain constrained if the state’s underlying digital infrastructure is decrepit.
As with institutional reform, no individual department can solve this problem alone. The centre must therefore lead the development of a shared architecture for the state. Its role is not to centralise all government information or dictate every departmental system, but to identify which capabilities must be common, set the rules and standards for interaction between institutions and AI systems, and ensure the resulting architecture is adopted across government.
Recommendation: Build shared digital and data foundations across government.
The centre should establish common requirements for data stewardship and quality; metadata and discoverability; interoperability and secure exchange; access and lawful reuse; lineage and lifecycle management; digital identity and access controls; purpose limitation and secure reuse; and auditability and records management. These requirements should cover people, organisations, digital services and AI systems.
Departments should remain responsible for the quality and management of the data they collect and maintain. Their subject expertise is essential to understanding what the data represent, how they were produced and where their limitations lie. But data should not be treated as an institutional possession where authorised use could improve outcomes elsewhere in government. Common infrastructure and interfaces should make appropriately governed exchange the default, rather than requiring institutions to negotiate bespoke arrangements whenever information is needed.
Reliable digital identity and access controls are also essential. The government must be able to establish who or what is requesting information, for what purpose and under whose authority. These foundations should support a shared operating picture at the centre. Information on outcomes, expenditure, delivery milestones, dependencies, risks and unresolved decisions should be available at the cadence required for action. But the operating picture should preserve the evidence beneath the synthesis, distinguish observed facts from generated interpretations and communicate uncertainty where relevant. Its purpose is to connect the information required for collective action, not to reduce complex public outcomes to a single central dashboard.
But a government’s information problem extends beyond data. Much of its institutional knowledge is held in submissions, emails, reports and the memories of individual officials, tying what the state knows to particular people and organisations. This makes organisational change costly. When ministers move, teams are reorganised or responsibilities shift between institutions, important information and context can be lost, forcing new teams to reconstruct knowledge that government once possessed. This constant shifting and loss of tacit knowledge weakens the state’s ability to learn continuously and to retain these lessons.
A more flexible and adaptive state could exacerbate these risks and must therefore be paired with the ability to create and maintain institutional memory that endures structural change. If teams are expected to form, evolve and dissolve around changing priorities, the knowledge they generate must be preserved and remain available to those who follow. AI is a key part of this solution, making institutional knowledge easier to retrieve and access, but it can only do so if that knowledge has been captured, stored and governed properly. Yet AI also creates new risks to the integrity of institutional memory. As generated summaries are reused in subsequent analysis, qualifications and context may disappear, and an interpretation may gradually come to be treated as an authoritative fact. Government therefore needs not only to make its knowledge searchable but also to preserve the provenance, version and historical context of the information that AI retrieves and interprets.
Recommendation: Create shared institutional memory across government with verifiable provenance and decision lineage.
To prepare for a new era of AI-enabled government, a new knowledge architecture is needed that preserves significant decisions, the authority under which they were made, and the evidence and assumptions that informed them. This should also record evaluations and delivery lessons, the alternatives considered and the trade-offs resolved. It should connect decisions to their subsequent implementation, allowing government to understand not only what changed and why, but what happened as a result and what was learned.
Institutional memory should be versioned and time-aware. Policy, legislation, guidance, implementation arrangements and the available evidence change continuously. The system should allow authorised users to establish what government considers authoritative now, what was considered authoritative when an earlier decision was made and what has subsequently changed or been superseded.
The underlying official records must remain the source of truth. AI-generated summaries and interpretations should operate as a derived layer and should never silently overwrite or replace the authoritative record. Every material synthesis should clearly identify the evidence on which it is based, ideally the relevant passages or data rather than merely a document as a whole. Changes, corrections and overrides should themselves be auditable so that government can reconstruct how its knowledge evolved.
This knowledge must be subject to appropriate requirements for classification, privacy, retention, access and records management. Different users should see information according to their role and authority. Together, these protections would create decision lineage: the ability to reconstruct a decision from the public purpose and authority behind it, through the evidence, advice and assumptions that informed it, to its implementation and eventual effects.
A governed institutional memory would allow authorised ministers and officials to interrogate the accumulated experience of government: what has been tried, what happened, why earlier decisions were made and which lessons may be relevant to a current problem. It would reduce dependence on individual memory while making it easier to transfer learning across departments and successive governments. It would also allow decision-makers to distinguish between what government knows now and what it knew – or believed – when an earlier choice was made. This is a valuable function for democratic scrutiny.
As government becomes increasingly dependent on shared digital foundations, it will also create new forms of technological dependence and vulnerability. Systems that connect information or perform actions across institutional boundaries could be compromised or manipulated, potentially exposing sensitive information or carrying out unauthorised actions. The greater the authority and access granted to a system, the more consequential these risks become.
Recommendation: Embed security and operational resilience into the state’s digital foundations.
The government should establish common controls covering instruction injection, data leakage, compromised external content, unauthorised access and manipulation of connected systems. These should be supported by continuous security testing and operational monitoring, with responsibilities for identifying incidents, escalating risks and restoring services assigned in advance. The level of protection should increase with the sensitivity of the information a system can access and the consequences of the actions it is authorised to perform. In the age of AI, cyber-security is going to be increasingly important as the offensive cyber-capabilities of frontier models and agents grow.[_]
When building the state’s new AI and digital foundations, it is essential for governments to plan for failure and build resilience into the system from the outset. Models may become unavailable, suppliers may experience outages or new security vulnerabilities may be uncovered, forcing rapid changes to a digital system. If AI is embedded in essential government functions, the state must be able to continue delivering even in the face of disruption. In other words, systems must be portable and adaptable when individual components fail or need replacement.
To do this, the government should avoid architectures in which the application, model and supplier are so tightly integrated that changing one component requires rebuilding the whole service. Instead, governments must develop digital architectures that rely on modular systems and are flexible and adaptive by default. Common interfaces and contractual exit provisions should allow models or providers to be substituted as performance, security, risk or data-classification requirements evolve. Public bodies should maintain fallback arrangements and the ability to disable, roll back or replace AI capabilities without losing access to the underlying data, workflow or public service.
Deliver the AI-Enabled State
With more adaptive institutions and the right digital foundations in place, governments can begin using AI to transform how the state operates. This transformation will take two forms. First, AI can augment the capabilities of individuals and institutions, helping them perform existing functions more effectively. Second, it can also enable new ways of organising and coordinating government, allowing information, expertise and eventually action to move more easily across institutional boundaries. Governments should therefore experiment with these new capabilities while creating mechanisms to scale what works across the state. As these systems become increasingly agentic and autonomous, governments will also need clear rules that enable adoption while safeguarding important principles such as transparency and accountability.
Governments are not short of technology pilots. Departments regularly test new tools, often with promising initial results. However, experimentation rarely produces transformation across the state. Successful applications often remain isolated and evidence is rarely shared across institutions, leading to many institutions purchasing or piloting the same technology at the same time. Without pre-emptive action and guidance, AI is likely to reproduce this pattern, with some institutions moving faster than others and differing technical solutions being built across government.
Recommendation: Create a common route from experimentation to system-wide adoption.
To address this, the centre should maintain visibility of significant AI projects across government and establish a common pathway for significant applications to move from experimentation to adoption. Applications should progress through use-case definition, prototyping or controlled testing, evaluation, assessment of production readiness, deployment and continuous monitoring, followed by an explicit decision to scale, revise or retire them. This creates a common learning cycle in which implementation generates evidence to inform adoption elsewhere, rather than each institution having to discover what works independently. The evaluation of these systems should focus on the outcomes they deliver, rather than on their technical novelty or impressive capabilities demonstrated only in controlled conditions.
Procurement is an important tool in supporting this. If departments purchase technology under incompatible standards and contractual terms, the state will continue to buy fragmentation. Contracts should meet common requirements for interoperability, data and model portability, security, evaluation, auditability and exit. They should provide audit rights; require appropriate logging, observability and incident management; protect government intellectual-property and data-use rights; and allow models or providers to be substituted without rebuilding the application layer. This does not mean prescribing a single model or supplier, as AI will continue to develop rapidly and different systems will be appropriate for different tasks and data classifications; today’s preferred model may quickly be superseded.
Together, these mechanisms would allow AI capabilities to spread across government without requiring either technology or experimentation to be centralised. But some of the most important applications will sit at the centre itself, where AI can expand the information and analysis available to political leaders as they make consequential decisions. The ambition should be to build a new centre of government capable of harnessing AI to improve its strategic intelligence.
Recommendation: Build an AI-enabled strategic intelligence and simulation capability at the centre.
Today, the centre often makes decisions based on data assembled for a specific meeting or decision point. This information often arrives at the centre through departmental processes, with different institutions providing different data. This often means that the data are not linked, and decision-makers may be missing important connections between data assets. AI can help create a more persistent strategic intelligence capability. It can retrieve and synthesise relevant evidence, map dependencies between priorities, compare departmental positions, and identify where several objectives depend on the same infrastructure, resources or institutional decisions. It can also connect internal delivery information with relevant economic, scientific, technological, security and societal developments.
Importantly, it can do this continuously and in real time. The assumptions underlying a strategy may change or a new development may alter the feasibility of a plan. The centre should be able to identify these signals across dispersed sources and direct expert attention to developments that may require political reconsideration.
This capability can also help leaders explore the potential consequences of the choices they face. Governments routinely make major decisions without systematically examining how their effects may interact across the state. The centre should therefore combine AI with economic modelling, forecasting, simulation, operational data and structured expert judgement to explore potential interactions. For example, decisions about taxation, infrastructure, regulation, public services or technology policy may have consequences in several sectors or depend on assumptions that are rarely tested together. Better strategic simulation capabilities would allow leaders to explore these interactions more quickly and at lower cost. This would enable leaders to expand the range of possibilities they can consider and the assumptions that underpin them.
As the capabilities of AI continue to advance, the technology will move from a passive role in supporting strategic capabilities such as simulation or planning to a more active role. At present, most AI use cases are focused on specific tasks, such as drafting material or summarising documents, but as agentic capabilities become more widespread, AI will also play a role in supporting government work by pursuing bounded objectives over time, interacting with several systems and taking permitted actions when specified conditions are encountered.
Recommendation: Introduce bounded AI agents to coordinate routine government activity.
Agents should be introduced only where the underlying infrastructure, such as data and identity management, is in place and sufficiently mature to allow their actions to be audited and monitored. What agents can do must be clearly defined, and each agent will need a named human owner. Agents may exercise delegated operational authority, but they cannot acquire political mandates: consequential decisions about priorities, trade-offs or the exercise of state power must remain with accountable humans.
As more agents enter the broader government ecosystem, institutions will be required to adapt by implementing new rules and policies that take into account emerging agentic capabilities. Over time, this will lead to the creation of a new form of governance, in which agents are part of the state’s core infrastructure. While this will increase effectiveness and efficiency, it will also bring with it risks that demand important safeguards related to transparency and accountability.
Recommendation: Ensure AI-enabled government remains aligned and democratically accountable.
Just as institutions must remain organised around the outcomes political leaders have prioritised, AI systems operating within them must remain anchored to a legitimate public purpose. In practice, this means ensuring that the tasks they perform, the data they use and the measures by which they are assessed all serve that purpose. Because institutions adapt to the targets against which they are judged, AI may allow government to optimise for one value at the cost of another.[_] As systems become more capable and autonomous, this risk becomes more consequential.[_] To help address this risk, systems should be monitored and tested as part of a broader AI-assurance ecosystem to ensure they are safe, reliable and working as intended.
Finally, as the capability of the AI-enabled state grows, it is essential that its activities also remain transparent and accountable.[_] Personal and sensitive information should be governed by data minimisation (collecting only the data that is necessary), the once-only principle (avoiding asking people to provide the same information repeatedly) and appropriate access controls. People must be able to understand why a consequential decision was made about them and have appropriate opportunities to challenge or appeal it. This will require systems to generate clear logs detailing the purpose and authority under which the system operated; the inputs and authoritative sources used; the relevant system, model and policy versions; the recommendations, classifications or actions produced; the points at which human judgement was exercised; and any subsequent overrides, appeals or corrections.
Chapter 5
In the age of AI, the quality of a country’s institutions and its leaders’ ability to govern effectively through them are central to its strength and resilience. Those who can build new institutions that adapt to and harness new technologies will be best positioned to translate political choices into action. Those who cannot will fall behind, and when they fail to deliver, they will have to explain why to their citizens.
Leaders will have to make choices about how their countries respond to this period of geopolitical and technological disruption. However, most governments are still not organised for this new world, and many institutions remain too slow and too inflexible. As the pace and complexity of change increase, these limitations will become harder to sustain.
In the age of AI, effective leadership means building a state that can adapt as circumstances change and use new technologies to become more capable. AI can help government understand the world more clearly and respond more effectively. But greater technological capability does not replace political leadership; AI cannot determine what a country should seek to achieve or make the difficult choices that governing requires. Those remain the responsibility of political leaders.
The Reimagined State is a model for bringing these elements together. It combines stable institutions that preserve expertise with more flexible structures that can organise around changing priorities. It is built on shared digital foundations that allow information and knowledge to move across institutional boundaries. It uses AI to make individuals and institutions more capable, while enabling new ways for government to learn, coordinate and act. The objective is not to replace existing institutions or transfer political authority to machines, but to build a state that is better able to translate political direction into results.
Much of this transformation must begin at the centre of government. This is where political authority connects with the distributed capabilities of the state, and where leaders can reshape how those capabilities are organised around their priorities. The centre must set direction and create the common foundations that allow institutions to work together. But the objective is not greater centralisation. Expertise, experimentation and delivery should remain distributed throughout government, while the centre creates the conditions that allow these capabilities to combine and adapt as priorities change.
The countries that succeed in the age of AI will therefore not simply be those that develop or acquire the most powerful technologies, but those that can translate technologies into greater national capability. That requires political leaders with the judgement to set a direction, but also institutions capable of making it a reality. The capacity of the state to adapt and deliver will increasingly shape a country’s prosperity, security and power. Leaders who recognise this will have the opportunity to transform government for the emerging world. Those who do not will find that the institutions they inherited are no longer capable of delivering the future their citizens expect.