New paper on The Agentic State
Very happy to share “The Agentic State: How Agentic AI Will Revamp 10 Functional Layers of Government and Public Administration”. I had the honor of contributing to this whitepaper, which doesn’t ask how to add AI to government: instead, it asks what government would look like if it were designed around intelligent systems from day one, with democratic accountability at its core.
In my experience, many digital reforms haven’t stumbled because of poor technology, but because they’ve had to operate within outdated institutional models. Thus we should be asking what a different operating logic for government might look like.
That’s exactly why we wrote this paper: not to ignore those gaps, but to provoke a conversation about what comes after. What if, instead of tweaking around the edges, we asked what government could become – once the foundations are finally ready for a leap?
Some of the ideas we explore include:
-
Second-generation Digital Public Infrastructure (DPI): The paper glimpses the next layer of DPIs: programmable payments, agent orchestration, and model context protocol. An infrastructure for agents, not just humans.
-
A dual identity system with agentic power-of-attorney: One for people and organizations (via verifiable credentials), one for software agents (via cryptographic IDs), with delegation tokens defining who can act for whom, on what, and for how long.
-
Outcome-driven internal workflows: Don’t digitize steps. Define the outcome (e.g. “approve 90% of permits in 10 days”), and let agents orchestrate tasks dynamically – escalating edge cases to human review.
-
New economic models for AI deployment: Treat agentic systems as services. Pay per outcome, per transaction, or per marginal cost of reaching underserved groups – aligning incentives around public value.
-
Professional gatekeeping: As AI enters regulated domains like education, medicine, and justice, governments will face resistance from professional bodies guarding status, roles, and privileges. Should new models of negotiation and credentialing be developed to safely integrate agents into certain domains?
-
Leadership that governs by design: As agentic systems gain autonomy, leaders must learn where to act, and where to step aside – investing, divesting, regulating, or deregulating with strategic intent and legitimacy.
Version 1.0 is just the start. It is a blueprint to provoke, not prescribe. Comments and reflections are more than welcome: they are needed. I am especially interested in hearing from people who have tried implementing any of these ideas in real environments. What worked, what failed, and what are we missing?
Finally, huge thanks to lead author Luukas Ilves, and to fellow contributors Manuel Kilian and Ott Velsberg, for the opportunity to join them in this effort, and to the Global Government Technology Centre for making this work possible.
Link to the paper in the comments below 👇