The research programme

What this page is

A plain description of a research direction at an early stage. It says what the question is, what the work rests on, what has been done, and — at more length than is usual — what has not.

This page describes the research programme behind this site. The programme has no formal name of its own, and the term agentic state is not ours: it is the title of a vision paper by Luukas Ilves, Manuel Kilian, Simone Maria Parazzoli, Tiago C. Peixoto, Ott Velsberg and other contributors, launched at the Tallinn Digital Summit in October 2025 [1]. This programme works inside the field that paper helped name and makes no claim of priority over it.

The research question

Governments have begun to use AI systems that act on cases rather than answer questions about them. The programme's umbrella question is:

How can counter-power be built into the same architecture that makes a state capable of acting through AI agents — at the same time, rather than after the first failure?

That question is too broad to study directly. The programme breaks it into eight narrower questions, each with its own literature, method and way of being wrong. They are set out in the research agenda:

1. Does a publicly funded, structurally independent citizen-side AI agent work better than a government-run or market-run one?
2. Is the natural-language instruction that steers an AI system's discretion a distinct form of law, and does letting a person see the one that governed their case change anything?
3. Can structural-separation logic from antitrust be applied to a state's own AI functions, not only to AI markets?
4. Would adversarial pre-deployment testing of a rule against synthetic hard cases have caught known administrative disasters?
5. Do federal systems already hold data at different granularity at different levels of government while keeping rights uniform?
6. Can an "off-switch" test distinguish alliance from dependency in AI and compute partnerships?
7. Can blameless incident review be adapted into an institutional memory for policy failure that resists being gamed?
8. Do three risk mechanisms — proactive service without a real refusal channel, individual prediction, and model consolidation — interact, or merely add up?

Theoretical basis

The programme does not rest on one theory. A prior-art audit placed its claims in six established fields: digital-era governance and public administration; administrative law and algorithmic accountability; street-level bureaucracy and organisational learning; democratic theory and AI-mediated participation; state capacity, federalism and antitrust as institutional design; and the international political economy of AI and compute.

Two findings from that audit shape everything else.

Several starting ideas are already in the literature. On our audit's reading, the claim that reactive, transaction-based digital government has reached a structural limit is close to Dunleavy and Margetts on the third wave of digital-era governance [2]; organising government around life situations rather than services is close to the "no-stop shop" of Scholta and co-authors, in which services are delivered proactively around life events [3]; the asymmetry between what a state's systems can do and what a citizen's can is related to Fourcade and Gordon's account of data-driven statecraft [4] and to Sharp and co-authors' preprint on agentic inequality [5]; and AI support for democratic deliberation is the subject of Tessler and co-authors' paper in Science [6]. The programme treats these as foundations to cite, not findings to claim.

"Agentic government" is not yet a settled academic field. Our audit found recent work from several groups — an oversight-structures paper and a typology of AI systems in public administration [7][8], papers on democratic authorisation and discretion [9][10], a risk-mechanism framework [11], and the RADAR preprint on how accessible government services are to AI systems [12] — but no shared canon and little cross-citation between the public-administration, AI-safety and administrative-law strands. This is an assessment from a limited search, not a bibliometric result.

Method

Three things have been done, all desk research.

• Claim extraction. Thirty candidate claims were extracted from a book-length manuscript by the author.
• Prior-art audit. Each claim was tested against six literature corpora, about 115 papers in total, searched on 2 October 2026. Each claim received a verdict: established, emerging in parallel elsewhere, or apparently open.
• Research design. The eight strongest open questions were developed into concept cards stating the question, the closest existing work, the method, the evidence required, and what result would falsify or weaken the claim.

The methods the eight questions call for are standard: comparative institutional and doctrinal analysis; survey and vignette experiments; design science, including retrospective tests against documented cases; and comparative case study using existing dependency-mapping techniques.

What exists

Stage

Research design. Two texts exist: the research agenda and a working paper on current government cases. Neither is a study. No journal article has been drafted. No survey has been fielded, no prototype built, no dataset assembled, no ethics approval sought. Nothing has been submitted to a journal or a conference. The programme's own internal assessment is that drafting should not begin until four follow-up literature searches are complete: independent fiscal institutions; automation complacency and cognitive offloading; right-to-be-forgotten doctrine; and scholarship on social-credit systems.

The programme has no academic partner and no institutional affiliation. Its own planning notes record that the experimental questions would need a collaborator trained in experimental methods, and that two of the legal questions would benefit from a co-author with formal legal training.

Conclusions and practical implications

There are no research results to report. What the programme offers today is narrower:

• a map of which ideas about agentic government are already established and should be cited, which are being developed in parallel by others, and which appear open;
• eight questions stated precisely enough that someone else could take them up, with the result that would count against each;
• a working paper that applies the same discipline to current cases, separating announced from piloted from deployed systems, and proposes an architecture that has not been tested.

For practitioners, the usable part today is the agenda's list of falsifiers: the evidence a government should ask for before accepting any of these ideas.

Limitations and open questions

• No peer review, no external review. Nothing here has been read by an independent academic reviewer.
• The literature search was limited. It ran without access to Scopus, Web of Science or JSTOR, through open indexes, publisher pages and preprint servers. A finding that a question is "open" means not found in that search.
• The audit was AI-assisted. Bibliographic records for the works cited on this page were re-checked on 4 October 2026. On 5 October the descriptions of what those works argue were compared with the works themselves, using the full text where an open copy exists and the abstract otherwise, and several were reworded.
• The programme starts from one author's manuscript. The claims audited are that author's; the audit was designed to test them against the literature, not to defend them, but it was not independent of the author.
• The field is moving. Several cited works are preprints. Novelty verdicts made in October 2026 will date.
• Open question for the programme itself: whether eight loosely connected questions are a programme or a list. The internal answer is that they share a source and a concern but not a theory, and should be pursued as independent projects.

Sources

Bibliographic records and URLs were opened on 4 October 2026 and checked again on 5 October 2026.

1. Ilves, L., Kilian, M., Parazzoli, S. M., Peixoto, T. C., Velsberg, O., et al. (2025). The Agentic State: Rethinking Government for the Era of Agentic AI (vision paper). https://agenticstate.org/paper.html
2. Dunleavy, P., & Margetts, H. (2023). Data science, artificial intelligence and the third wave of digital era governance. Public Policy and Administration. https://doi.org/10.1177/09520767231198737
3. Scholta, H., Mertens, W., Kowalkiewicz, M., & Becker, J. (2019). From one-stop shop to no-stop shop: An e-government stage model. Government Information Quarterly, 36(1). https://doi.org/10.1016/j.giq.2018.11.010
4. Fourcade, M., & Gordon, J. (2020). Learning Like a State: Statecraft in the Digital Age. Journal of Law and Political Economy, 1(1). https://doi.org/10.5070/lp61150258
5. Sharp, M., Bilgin, O., Gabriel, I., & Hammond, L. (2025). Agentic Inequality. arXiv:2510.16853 (preprint). https://arxiv.org/abs/2510.16853
6. Tessler, M. H., Bakker, M. A., Jarrett, D., Sheahan, H., et al. (2024). AI can help humans find common ground in democratic deliberation. Science. https://doi.org/10.1126/science.adq2852
7. Schmitz, C., Rystrøm, J., & Batzner, J. (2025). Oversight Structures for Agentic AI in Public-Sector Organizations. REALM workshop, ACL Anthology. https://aclanthology.org/2025.realm-1.21/
8. Rystrøm, J., Schmitz, C., Davies, N., Hammerschmid, G., Meijer, A., & Russell, C. (2026). A Technical Typology of AI Systems in Public Administration. arXiv:2606.31755 (preprint). https://arxiv.org/abs/2606.31755
9. Salah, M., Abdelfattah, F., Al Araimi, A., & Noor, A. A. S. (2026). Before agentic AI scales in government: the democratic authorization gap. Frontiers in Political Science. https://doi.org/10.3389/fpos.2026.1956489
10. Salah, M., & Alnoor, A. M. (2026). Agentic AI, Street-Level Bureaucracy, and the Recomposition of Discretion in Public Administration. International Journal of Public Administration. https://doi.org/10.1080/01900692.2025.2605225
11. Alibašić, H. (2026). Governing Agentic AI in the Administrative State: Human Oversight, Cybersecurity Accountability, and Risk in Autonomous Digital Systems. Future Internet. https://doi.org/10.3390/fi18100528
12. Jordan, L., Peixoto, T. C., & Ramos-Maqueda, M. (2026). RADAR: Readiness for AI Discovery and Agentic Reach. arXiv:2609.28480 (preprint). https://arxiv.org/abs/2609.28480