A Blueprint for Transformation

This whitepaper breaks down the broad notion of an Agentic State into ten distinct functional layers of government, ordered across three dimensions:

For each layer, this whitepaper follows a common structure, moving from present realities to future possibilities, with key questions for reflection. Each section is structured around three components:

How it (doesn’t) work today: An honest account of how the government and public administration currently operate, and where existing processes, technologies, or institutional norms fall short.

A vision for agentic government: A reimagined approach that leverages the potential of agentic AI to deliver public value in new ways, often by rethinking long-held assumptions.

Key questions: Critical reflections and practical considerations designed to support further debate and help substantiate the path toward realising the vision.

This format is designed to help practitioners and policymakers navigate the tension between today's world and the changes needed to make good use of agentic AI. The list of ten functional layers is neither exhaustive nor definitive. Rethinking how we model and think about public functions is itself a key step in transitioning to an Agentic State.

What is at Stake and What is Driving This Whitepaper

Collectively, the world’s governments are falling short at delivering public goods. In high-income countries, overall public trust in government and quality of public services are in decline. Emerging economies face a different challenge: the pressure to do more with less and to leapfrog to close the gap with wealthier states. 

The human and economic cost of poor governance is staggering: trillions lost in misallocated resources, unmet needs, and institutional failure. Worse still, most governments are structurally brittle and poorly equipped to manage disruption. Their organisations and processes were designed for stability, not for a world of cascading geopolitical, technological, environmental, and social shocks.

Technology alone is no panacea. However, when paired with deep institutional reform, it becomes a powerful force multiplier. What creates public impact is not adopting tools for their own sake but linking technology adoption and transformation to public missions. The difference between governments that have mastered the use of technology and those that have not is not measured in single digit percentages, but in multiples of cost savings, operational efficiency and service outcomes. And that is before factoring in the potential of agentic AI. 

Government digital transformation is increasingly becoming  a top priority in many countries. As governments undertake modernisation efforts, they must avoid the trap of committing significant resources solely to catch up with the present at a time when agentic AI is triggering a once-in-a-generation paradigm shift in technology, organisation and design. This whitepaper serves as a practical guide to grounding reforms and technology modernisation in the opportunities and challenges of the agentic age.

Factors for Successful Execution

There is no playbook or catalogue of best practices for adopting agentic AI in government. The technology is simply too new and developing too rapidly. Even the most digitally advanced public administrations are only beginning to experiment. Many AI transformation efforts, as done today, are bound to fail.

Public sector leaders can get a head start by looking to the private sector in four ways:

• Learn from private sector AI adoption and use cases: Enterprises are pouring billions into AI adoption and development. Their collective experience in coming years will yield a rich evidence base of best practices, useful lessons and empirical evidence on what works, and more importantly, what doesn’t (yet) work. Governments should observe closely and copy what works.
   

• Build with the private sector: The past three years have seen a wave of new products to support enterprise AI adoption. Nearly every step of the AI development and adoption lifecycle has multiple providers offering models, tools, data, much of it open source. Instead of reinventing the wheel, public administrations should seek to build on what is already there and invite enterprises to collaborate with them.
  
  

• Adopt new economic models: Governments should increasingly treat agentic deployments not just as technical builds, but as performance-based services, paid for per transaction, per outcome delivered, or per marginal cost of serving hard-to-reach groups. This shift in pricing models can help align incentives, control expenditure at scale, and ensure that AI systems deliver public value, not just operational novelty.
  
  

• Operate as a platform for private sector and individual invention: Agents reduce the technical and skills barriers to accessing and building on APIs, datasets, and digital infrastructure like eID, allowing users to interact with government systems and data using natural language instead of technical interfaces or developer tools. Governments should extend and democratise their existing role as technical and data platforms, making these resources compatible with and useful to third-party AI agents operated by the private sector, civil society and even individuals. 

The gap between rapid private sector AI development and slow government adoption also presents an opportunity. In economic sectors where public funding, infrastructure, data or standards are critical (health, education, energy, science, infrastructure, agriculture, construction, defence and public safety to name just a few), those governments that move faster will also create significant opportunities for private innovation, growth and export.

Assumptions, Caveats, and Limitations

Assumptions and predictions about the capabilities of agentic AI are grounded in tools that are either already in use or in public development by leading AI companies and research organisations, as well as the initial findings from their adoption in the private sector. While it is relatively optimistic about the reliability and performance of evolving technologies, it does not speculate about breakthroughs beyond current capabilities.

The whitepaper also assumes a certain amount of structural continuity. Rather than imagining a radically transformed state or society, it builds on the existing functions of public institutions and the lived experience of today's citizens. It assumes that human biology, psychology and social structures will remain largely unchanged. It is intended as a guide for today's policymakers and practitioners, not as a vision of a transhumanist future.

Version 1.0 of this whitepaper does not cover several adjacent but critical topics in depth:

• A comprehensive societal vision for the agentic age, including the broader economic and social disruptions AI may bring.

• The impact of generative and agentic AI on politics, media, and the public sphere.

• Governance or regulation of AI, including AI safety and ethics.

• International and global collaborative approaches (e.g. standards, guidelines, digital public infrastructure, or interoperability) to pursuing the visions outlined here.

• Forecasts or timelines for technological developments or adoption.

• Environmental, supply chain, sovereignty implications of wide-spread AI use.

This whitepaper takes a horizontal view across core government functions. It is agnostic to specific policy domains (e.g. healthcare, social services, education, or justice), which are natural candidates for follow-on work. And it presents a snapshot in time, a reflection of present technological developments.