Agentic Government
RESEARCH PROGRAMME — Governance. An emerging programme. Grounded in the present; its forward-looking companion sits in the Futures strand.
Agentic Government
This programme studies what happens to public decision-making when the state acts through autonomous AI agents — systems that do not just advise but plan, delegate, call tools, and execute across steps. It asks how a citizen can still see, trace, contest, and stop a decision when the actor on the other side is an agent rather than a caseworker, and what an accountable agentic administration would actually require.
The algorithmic state is already familiar: agencies use models to route cases, flag risk, and prioritise applications. The agentic turn is different in kind. When an agent delegates a task, passes context between steps, or hands work to another agent, the ordinary record preserves events without preserving who acted under whose authority — and “a system did it” becomes a literally accurate account of a decision no one can trace. This programme treats that traceability gap as the central governance problem of AI in public administration: not whether the state may use agents, but whether it can remain answerable for what they do.
How it connects. The programme applies the admissibility method to the agentic layer — a public agent needs identity, bounded authority, a human boundary, and reversibility before it touches a consequential decision. Its present-tense analysis pairs with the Futures strand, which projects the same question forward: see Three Paths for the Agentic State, 2026–2030 for the scenarios, and the Agent Identity signpost for how to read an agent’s traceability.
Work in this programme
- Admissibility for AI Agents: A Record-Based Test — the record-based admissibility test for autonomous agents (primary home: Admissibility & Evidence; foundational here).
- Related reading in the Futures strand: Three Paths for the Agentic State and the Agent Identity signpost.
What we are working toward. A traceability standard for public-sector agents that survives delegation; the conditions under which an agent may be admitted to a government decision chain; and the line between an agent that assists an official and one that has effectively replaced the official’s judgment.
