Vision Paper: The Agentic State

The following is a summary of the agetnic state which is published at the 2025 Tallinn Digital Summit about the Agentic State. Prominent politicians from around the globe have gathered to come up with a feasible vision that goes beyond a digital government to an Agentic Government.
The main message of the paper is the urgent need for governments to adapt agentic AI in their workflows and provide use cases and the requirements for deploying Agentic AI in a fruitful way. The main reason is the gap confronted by citizens between the internet culture (e.g., Netflix) and the bureaucratic processes enforced by existing governments. The transformation to an AI-based government should be accomplished by realizing twelve layers. The twelve layers are divided into:

• Six implementation layers that represent different sectors:
  • User Interface transform from fragmented portals into centralized, personalized, and multimodal agents.
  • Workflows transform from hand-crafted compliance-first processes to workflows that are dynamically generated by AI-agents to meet a predefined outcome (e.g., A permit agent might be tasked with approving 90% of construction applications within ten days while rigorously enforcing zoning and climate standards)
  • Policy and law making transform from being deliberative and static to adaptive and simulation-based. Current governments create laws for old concerns, since it takes years to debate and enact them. An agentic government encodes law as executable logic, which is simulated on digital twins and synthetic populations. The outcome of the law or policy is continuously monitored in terms of success metrics and compliance rates.
  • Regulatory compliance transforms from being periodic and retrospective to being embedded as agents in the two sides of the compliance process: companies and regulators. Current regulator compliance is retrospective in the sense that it happens abruptly and usually addresses rule violations years after they happen (e.g., checking tax evasion years after it happens). Agents at firms can monitor in real-time input, for example, emissions, and report to regulators in real time.
  • Crisis response from being hierarchical to being predictive and simulation-based
  • Public procurement, from being bureaucratic and hence not competitive and fostering small businesses, to being completely automated. AI agents look up suppliers in real time, compare offers, and adapt contracts dynamically to shifting needs. To achieve these use cases, the following layers are needed.
• Six enablement layers
  • Agent Governance: enabling higher accountability, safety, and redress by ethical, legal, and appeal frameworks realized by the agents. The ethical frameworks provide standards and algorithms that promote fairness and transparency. The legal framework links agentic actions to authorized human officials. The appeal framework allows citizens to have control over agents that control their lives.
  • Data and Privacy: treating data as a critical infrastructure and data analysis skill as requirements for each employee. Data should be open by default to allow innovation, but still should be protected for privacy.
  • Tech Stack: five layers of multimodal interfaces, application infrastructure through APIs, orchestration and interoperability layer, digital public infrastructure that provides verifiable digital identity, and a compute cluster.
  • Cybersecurity: Protection against malicious usage
  • Public Finance: Outcome-based pricing models and adaptive contractive
  • Leadership: the strategy needed to transform government workforces