The Next Right Questions on the Path to the Agentic State

Many public-sector organisations are experimenting with artificial intelligence, yet more pilots do not automatically add up to public-sector renewal. Similar work is still redesigned and procured separately across organisations, cases stall at organisational boundaries, and people are repeatedly asked for information the public sector already holds.

The next question is therefore not simply which AI tools government should adopt, but at what level public work should be understood, compared and redesigned. This working paper proposes the function as a shared level for doing so.

A function is a recurring public activity with a clear result, such as receiving an application, assessing eligibility, issuing a licence, making a payment, or providing guidance. The same function may cross organisational boundaries or recur in similar form across many public bodies.

Agentic AI changes what is possible and this builds the case for function-based thinking. AI agents can increasingly carry out bounded, multi-step work across systems and organisational boundaries. That capability creates an opportunity to redesign recurring public work as a whole rather than adding new technology to existing organisational processes.

Function-based thinking does not remove organisations, legal mandates, or accountability. It makes repeated work visible across them. It allows leaders to distinguish what can be shared from what should remain specific, identify where reusable capability could create value, and coordinate dependencies that no organisation can solve efficiently alone.

The function links strategic priorities to operational work. It sits below public purpose and portfolios and above workflows and tasks. That role makes it the level at which leaders can compare recurring work and decide where to focus attention, while the detailed human–agent boundary is defined within the workflow. Statutory responsibility remains with the responsible public body.

Function-based thinking also changes how AI opportunities should be prioritised. Technical feasibility alone is not enough. Leaders need to consider value, scale, feasibility and deliverability together: Is the function worth changing? Is the opportunity large enough? Where can AI safely contribute? And can the participating organisations actually deliver the change? Not every technically feasible opportunity is worth pursuing.

The broader opportunity goes beyond automation. Shared definitions, interfaces, safeguards and evidence can allow several public bodies to reuse parts of the same capability while retaining justified variation. Evidence from human–agent work can also travel across organisations, allowing improvements in one implementation to improve the same recurring function elsewhere. In this way, learning from one organisation can create value beyond that organisation.

Before deciding where to invest, leaders first need to understand where agentic AI may be suitable. Published with The Agentic State foundation alongside this working paper, First Moves: A Finland-Specific Assessment of Agentic AI Potential in Government maps where recurring public work may be technically and legally suitable for agentic AI. That map is a starting point, not an investment agenda: it does not show actual workflows, costs, benefits, service outcomes or organisational readiness.

This working paper asks the next right questions: How should government turn that emerging map of potential into decisions about what to redesign, test and scale?

This paper proposes that the next 18 months should be treated as a bounded learning period, not as a race to build a national platform. The objective is to learn what can be reused, governed and scaled, not to prove a predetermined solution. To do this, Finland should select one recurring function with limited legal discretion and a clear boundary for human judgement. Participating organisations should establish the baseline, redesign the workflow and data before procuring technology, and test whether a reusable capability can work across several public bodies.

Application intake, including initial processing, is a credible candidate for such a function because it recurs across the public sector, contains measurable repetition and delay, and can be separated from the final legal decision.

By 2030, Finland will need to answer harder questions about permanent stewardship, funding, standards and accountability. For now, the practical task is simpler: Find one important function that several public bodies already perform, make the shared work visible, and decide what those public bodies should redesign, learn or build together.