Why the Digital State Must Evolve into an Agentic State

Europe has spent years building the foundations of the digital state. We have invested in identity, registries, data exchange, cybersecurity, interoperability frameworks, digital services, and common rules for trustworthy transformation. That work matters. Without it, there is no serious public sector digitalisation. But that foundation is no longer enough.

The next phase of state capability will not be defined only by whether services are online, whether data can be exchanged securely, or whether a process has been digitised end to end. It will be defined by whether governments can act intelligently, proactively, and coherently across institutional boundaries. In other words, the digital state must now evolve into an agentic state.

I do not mean a state run by autonomous machines. I mean a state that can use AI and data in a structured, governed, and interoperable way to help institutions act with more context, better timing, stronger coordination, and greater capacity. The agentic state is not a rejection of the digital state. It is its next layer.

This matters for Europe in particular. European public administrations operate in a complex environment: legally dense, institutionally fragmented, multilingual, and accountable to high standards of rights, transparency, and democratic legitimacy. In such an environment, simply adding AI tools on top of fragmented systems will not produce transformation. It will produce islands of automation.

That is why Europe should think of agentic capability not as a procurement category or a generative AI trend, but as digital public infrastructure.

From digitising processes to orchestrating outcomes

Much of public sector digital transformation has focused on moving from paper to digital workflows. That was necessary. Then came the next phase: connecting systems, enabling once-only data use, improving service design, and making services more proactive. But the public sector challenge has changed.

Today, governments must operate in environments shaped by rising complexity, shrinking administrative capacity, growing public expectations, and accelerating technological change. Citizens and businesses no longer compare a ministry only to another ministry. They compare their experience with the best digital services they encounter anywhere. At the same time, governments are expected to remain lawful, explainable, inclusive, and secure.

This creates a new imperative. The state must not only provide digital services, it must develop the ability to coordinate decisions, actions, and information and data flows dynamically across the public sector. That is where the idea of the agentic state becomes useful.

An agentic state is one in which public institutions can rely on governed AI agents and agent-like systems to support service delivery, policy execution, administrative coordination, and decision preparation. These systems are not free-floating. They operate within legal mandates, technical guardrails, institutional responsibilities, and human oversight. Their value is not in replacing the state, but in making the state more capable.

This distinction is crucial. Too much discussion about AI in government still starts from the tool. The conversation quickly becomes about large language models, assistants, copilots, or automation products. But the real issue is not the tool. The real issue is whether public administration has the architecture, governance, and institutional design required to use such tools meaningfully. Without that, “AI adoption” remains superficial.

The agentic state is infrastructure, not a feature

One of the biggest mistakes governments can make is to treat agentic systems as isolated applications. If every authority buys its own assistant, builds its own orchestration logic, uses its own data model, and interprets governance rules in its own way, the result is predictable: fragmentation, duplicated investments, inconsistent quality, and low trust.

The agentic state therefore has to be understood as a new infrastructure layer. Just as digital identity created a trusted layer for authentication, and just as secure data exchange created a trusted layer for interoperability, agentic capability must create a trusted layer for action, coordination, and context-aware execution. That layer includes several elements.

First, it requires high-quality, interoperable data environments. Agentic systems are only as useful as the information environment in which they operate. If data remains inaccessible, inconsistent, or legally unclear, agents will not create value at scale.

Second, it requires shared orchestration and interaction models. Public services are rarely delivered by one institution alone. They involve registries, sectoral systems, decision logic, workflows, permissions, notifications, and escalation paths. Agentic systems must be able to operate across this reality rather than inside a single organisational silo.

Third, it requires governance by design. Human oversight, traceability, contestability, risk classification, logging, and accountability cannot be retrofitted at the end. In the public sector, they are part of the infrastructure itself.

Fourth, it requires institutional clarity. Agentic systems must support clearly allocated public authority. Citizens should never be left guessing who is responsible for an outcome, which rules were applied, or how a decision can be challenged.

And fifth, it requires strategic public capacity. If governments do not understand how agentic systems work, how they should be tested, where they add value, and where they should not be used, then the state will become dependent on vendors for not just technology, but judgment. That is not digital sovereignty. That is outsourced statecraft.

Estonia as a proof point, not a template to copy blindly

Estonia is often described as a digital frontrunner because of its long-standing investments in digital identity, interoperable registries, and secure data exchange. X-Road has served as the secure backbone connecting many public and private sector systems, while the once-only principle has helped reduce repetitive bureaucracy and support more seamless service delivery. Estonia has also pushed further toward proactive services, where the state can initiate service delivery based on existing data rather than waiting for citizens to navigate institutional complexity alone.

These achievements matter because they show an important truth: agentic government does not begin with generative AI. It begins with the quality of the underlying state architecture.

In Estonia, this is precisely why the discussion is moving beyond basic digitisation. The idea behind Bürokratt was never simply been to build a nicer interface. It has been to create an interoperable network of public and private sector AI solutions that can appear to the user as a coherent channel for services and information. Estonia’s Digital Agenda 2030 similarly framed a future in which major public services are linked to a national virtual assistant ecosystem rather than remaining separate entry points that citizens must understand on their own.

That shift has been important because it changes the ambition of the state. The question is no longer only: can a citizen access a service digitally? That is a fundamentally more agentic model.

At the same time, Estonia should not be misunderstood as a template to be copied mechanically. Europe is too diverse for that. Legal systems differ, administrative cultures differ, levels of digital maturity differ, institutional incentives differ. What Estonia offers is not a one-size-fits-all solution. It offers a proof point that when identity, data exchange, registries, trust architecture, and political ambition come together, the state can move from being digitally available to becoming intelligently coordinated.

Why Europe needs an agentic public infrastructure strategy

Europe now faces a strategic choice. One option is to allow agentic capabilities to emerge in a scattered way: one procurement here, one pilot there, some isolated copilots, some local automation experiments, and a growing patchwork of disconnected systems. This may generate impressive demos. It will not generate systemic public value.

The other option is to treat agentic capability as a European public sector modernisation priority and build toward it deliberately.

That does not mean centralising everything. On the contrary, Europe’s strength is federated governance. But federation only works when there is enough shared logic underneath it: common approaches to interoperability, shared governance patterns, reusable components, and mechanisms for scaling what works.

The Interoperable Europe agenda already points in this direction by stressing interoperability-by-design and cross-border public sector coordination. The same logic should now be extended into the age of agentic systems. Europe needs not only interoperable data and services, but also interoperable models for how AI-enabled public action is structured, supervised, audited, and improved. This has at least four implications.

Europe should invest more systematically in public sector orchestration capacity, not just front-end service interfaces.

It should develop shared testing and sandbox environments for agentic systems, especially in higher-impact public use cases.

It should promote reusable public components and open architectures, reducing dependency on fragmented, opaque deployments.

And it should ensure that AI governance is operational, embedded into delivery structures, procurement, and institutional routines rather than discussed only at the level of abstract principles.

In other words, Europe needs an agentic public infrastructure strategy.

Human-centric does not mean institution-light

There is often a false tension in debates about AI in government. On one side sits efficiency, automation, and speed. On the other sits rights, safety, and democratic legitimacy. That framing is too simplistic. The real public sector challenge is not choosing between innovation and control. It is designing systems where innovation is governable.

A mature agentic state must therefore remain deeply human-centric. But human-centricity should not be confused with manualism. It does not mean every process must remain slow, fragmented, or dependent on administrative navigation by the user. It means systems should be designed around human needs, institutional responsibility, and societal trust.

That includes preserving meaningful human oversight where required. It includes transparency about system roles and limitations. It includes protecting the ability to challenge outcomes. And it includes being honest about where agentic systems should not be used, or where their role should remain assistive rather than determinative. For instance in Estonia, we provide citizens an overview on how and for what reasons the public sector processes personal data through data tracker, give individuals right to decide who can process their government held personal data via consent service, we are making algoritmic transparency mandatory for the public sector.

This is particularly important in Europe. Public legitimacy does not come from technological sophistication alone. It comes from the combination of effectiveness and constitutional reliability. The agentic state will only succeed if people feel that it makes the state not only faster, but fairer; not only smarter, but more understandable; not only more proactive, but more accountable.

The coming divide: digital states and agentic states

Over the coming years, I expect a growing divide to emerge. Some governments will remain digital in the narrow sense. They will have portals, databases, workflows, APIs, and online forms. Their services will function, but their institutions will continue to operate largely in silos. AI will be added in pockets, with uneven outcomes and limited scalability.

Other governments will build the next layer. They will treat agentic capability as a core state function. They will redesign service delivery, administrative coordination, and policy implementation around trusted, interoperable, supervised AI-enabled systems. They will know where automation creates value, where human judgment must stay central, and how to combine the two. They will move from isolated digitalisation to coordinated intelligence. Those governments will not simply provide better user experience. They will possess greater state capacity.

That is why this conversation is more important than it may first appear. The agentic state is not a futuristic slogan. It is a practical governance question. How should public institutions operate in an environment where AI can increasingly help interpret context, propose actions, coordinate tasks, and support execution?

The right response is not to resist that shift, nor to rush into it carelessly. It is to shape it as public infrastructure. Europe has already shown that it can build trusted digital foundations. It has the regulatory seriousness, the public sector expertise, and the interoperability tradition needed to do this well. What is needed now is the next step in ambition.

The digital state was about making government available in the digital age. And that is why the digital state must evolve.