Automatic transcription, there could be errors.
Alessandro Oppo (00:00)
Welcome to another episode of the Democracy Innovator podcast. Today’s guest is Simone Maria Parazzoli. Thank you for your time.
I know you have been researching Civic Tech, GovTech, and governance solutions for the digital era we are living in. I also know you did research with the OECD. Would you like to tell us what you learned through your research?
Simone Maria Parazzoli (00:48)
Yeah, sure. A little bit of background about me: over the last couple of years I was indeed at the OECD, where my work was on three fronts.
The first was innovation in public services. The second was innovative participation—how emerging technologies can improve civic participation. And the third was AI in the public sector.
After those years at the OECD, I’m now part of a new initiative, partly funded by the World Bank, called the Agentic State. The goal of the initiative is to envision—and then lay the foundations for—governments to adopt agentic AI solutions.
There are many things I learned and that caught my attention over the last years, but the red thread throughout was always this tension between technology—specifically AI—and government. That includes the governance of these technologies when deployed in complex systems that are partly social, partly political, and partly technical, which is exactly what happens in government use cases.
Alessandro Oppo (02:33)
How did you start? AI is now everywhere, and many people are thinking about how it can influence social and political life. But how was it for you—when did you get the idea that AI was going to change governance?
Simone Maria Parazzoli (02:59)
My curiosity started when I was a student at Sciences Po, and I was lucky to have as one of my teachers Swazik Peniko, an independent researcher and journalist working on algorithms in the public sector.
She ran a course about algorithms in the public sector—how deployments could be done better—and we went through some of the most famous scandals around the adoption of algorithms in government.
I fell in love with the topic because it raised many sets of questions with no simple answer. We had technical questions to untangle—like what it means for an algorithm to be discriminating.
Take the famous COMPAS example in the judicial system in the US: there was a ProPublica article arguing the algorithm was discriminatory, and then an endless discussion followed on the technical details—how you measure discrimination, and to what extent an algorithm can satisfy different definitions of fairness.
In practice, that is a normative question, but also a very tangible one—because it affects real outcomes, like who goes to jail and under which conditions.
Fundamentally, what attracted me was that in this field there are many questions I don’t have answers to—and I find that very interesting. So I kept working on the topic.
I did my master’s thesis on AI in the public sector, trying to understand how algorithmic accountability can be ensured through policy. The idea is that governments have a palette of options: they can pick different policies, and each policy choice will shape the sociotechnical system—the complex system of technology, institutions, and humans acting together when an algorithm is deployed.
Among all those possible choices, the question is: what should governments do?
Like most research, I wasn’t coming in with a single answer. I was trying to offer a framework to think through the matter.
Then I stuck to the topic: I did data science research, NLP for governments, then OECD, and now the World Bank.
What is still extremely interesting to me is that there is really no clear answer to most of these questions. Even the basic ones. Even when we talk about deploying very simple algorithms in the public sector, we immediately face issues of governance, data management and sharing, privacy, and cultural tensions—like deterministic versus probabilistic approaches in public administration. These are questions that don’t have settled answers.
Alessandro Oppo (07:29)
I also have a lot of questions without easy answers.
You mentioned policy. In the Agentic State paper there was something about policy making—how AI agents can help draft policies and even update them almost in real time, based on data. That’s very interesting.
Simone Maria Parazzoli (08:06)
Yes. The Agentic State report is a vision paper. It’s a serious “science fiction” attempt to describe what governments could look like when AI agents are widespread.
To describe that future somewhat rigorously, we developed a framework with 12 layers, and one of those layers is policy making and rule making.
For that layer, we claim that agentic policy making and rule making will show four characteristics.
The first is dynamic policy simulation: AI agents stress-test policies in digital twins of society. Through simulation, they can explore how different outcomes could be achieved through different combinations of laws.
The second feature is machine-readable law: regulation is not ambiguous text, but precise and executable code. That enables more consistent application of the law and automated compliance.
The third feature is adaptive rule refinement: rules are not frozen in time. They are continuously monitored for impact and, through evidence-based analysis, adjusted to optimize outcomes identified by policymakers.
The fourth feature—probably the most interesting for the civic tech community—is participatory intelligence: using agents to collect more and higher-quality feedback from citizens, businesses, and frontline systems.
Those continuous data points can be leveraged to develop better policies and regulations. In near real time, you can learn from the effects of regulation and improve it—while trying to preserve democratic accountability in the best possible way.
Alessandro Oppo (11:10)
The idea of testing policies makes me think of what big tech companies do. When they change software, they don’t roll it out to everyone—only a small percentage. They see if it works, if there are bugs, and then they release it broadly.
And machine-readable law also makes me think about “code is law.”
Simone Maria Parazzoli (11:48)
On simulation: yes, that process is not far from A/B testing, which is common in private companies.
And at the same time, private companies continuously update terms and services. That’s a form of dynamic change of rules, driven by changes in regulation where the product is deployed, but also by changes in the product itself.
On machine-readable law: we have the common idea of “rules as code.” That isn’t new—it has long been considered a way to reduce ambiguity in regulation and make laws immediately operational.
But the truth is that now we might be able to accept some degree of ambiguity thanks to generative AI, because systems like ChatGPT do not require perfectly logical sentences. Within the ambiguity of natural language, they can still work well enough.
So we might see a new, updated version of “rules as code.”
As for “code is law,” it depends on what we mean—what code, and what law. In practice, in government, when we say “code is law,” what are we referring to?
Alessandro Oppo (14:47)
I was thinking that law can be interpreted, while code is executed. So if we have laws that can’t be interpreted but are just executed—removing ambiguity—then…
Simone Maria Parazzoli (15:08)
But I wonder—this is again an interesting question mark—because to what extent can an agentic solution be considered deterministic? It depends on the complexity of the solution.
In that sense, a “code is law” approach does not necessarily imply there is no ambiguity at all, or that the code is 100% deterministic.
Which raises the key question: to what extent can governments accept non-deterministic solutions? That’s probably the fundamental issue when we talk about AI in the public sector.
Alessandro Oppo (16:14)
Good question. I have to think about it.
Simone Maria Parazzoli (16:18)
It’s a big one, and we didn’t nail it.
There are regulatory mandates that administrative practice should be explainable and transparent, which means that as a citizen you have the right to understand why your case was treated one way or another.
But in the everyday practice of public administration, are human agents operating with zero ambiguity and fully deterministic logic? That’s a good question.
And especially when it comes to testing: when we are able to test algorithmic and agentic solutions, and evaluate their accuracy and recall, we’ll be able to assess how they perform against a human benchmark—which is not 100% accuracy.
Alessandro Oppo (17:52)
I think we are bringing back very old philosophical questions—like whether human beings are deterministic, whether we can choose what to do or not—and old problems without solutions.
Thinking about the agentic state—many AI agents, lots of data, many connections between entities—is this the beginning of civic tech, or the end?
Because in the paper there were examples where a citizen could say, “I want to open a company” or “I want to do something,” and the AI agents would guide them to the right institution.
So basically every question from a citizen could be satisfied by AI agents.
Simone Maria Parazzoli (19:32)
I think we need to draw a line between GovTech and Civic Tech.
GovTech generally refers to the application of technology in government to solve mainly efficiency problems inside government—procurement processes, internal workflows, database management, and so on.
Civic Tech, on the other side, is about the democratic practice of governance: the idea that technology can improve dialogue between citizens—understood as political actors—and policymakers, who are their representatives.
In Civic Tech, we look at people not only as service users, but as political actors in the “theater of democracy.”
The Agentic State paper takes more of a GovTech lens. But there are ideas here and there about civic tech.
I worked in civic tech for a while, and Tiago Peixoto, one of the co-authors, is one of the brightest thinkers in the field—so there is civic tech thinking underlying the vision paper.
But this distinction is important. And what could be interesting to discuss—and I’m curious to hear your take—is to what extent we should make deliberate efforts to blur the line between Civic Tech and GovTech.
Alessandro Oppo (22:07)
Often it seems to me that the distinction is already blurred. Everything related to states and government can be called GovTech, and software for citizens’ assemblies can be Civic Tech.
At the same time, if we reach a state that is so interconnected, can the outcomes of a citizens’ assembly become inputs for the GovTech part? Then where does GovTech end and Civic Tech begin?
Of course, this opens many questions.
And when we talk about this possible future, I don’t even know if we are talking about five years, ten years, twenty—or if we will ever see it in our lifetime.
Simone Maria Parazzoli (23:25)
The timeline is a big question mark for the Agentic State initiative. We don’t really know when the agentic state will be “live,” or to what extent it might be implemented in three years or five years.
There are a few governments moving fast toward something like this—Ukraine is the leader and one of the only countries starting to say out loud, “we want to build the agentic state.”
Ukraine is the most central voice, but Italy has taken first steps as well, and Singapore and Estonia too.
Still, the timeline is uncertain. We’re not fully sure that the “AI 2027” vision will happen as described. 2027 is very close, and we’re not seeing deployments that precisely match that timeline.
Alessandro Oppo (25:05)
I also wonder: nation-states are tied to territory. In the future, if we see agentic states, and everything is interconnected, states might become very interconnected too.
There’s also the theory of decentralized states. And we are Italians, but we use Google—so in some way…
If we think about the state as the entity that has the monopoly of violence, then the state is not so tied to territory. Where is the state?
Will we see, maybe in 100 years, a global state—a global agentic state?
Simone Maria Parazzoli (26:24)
I have no clue.
But the state has been dealing with this question for at least 30 years, since the internet: the physicality of territory, space, and how territorial aspects of the internet matter for sovereignty and the monopoly of violence.
For the agentic state, so far we are sticking to nation-states as they are. We are not imagining a global entity or the dissolution of nation-states. There is no strong reason to imagine that right now.
Yet there are more international collaborative efforts, especially around cloud and, most importantly, digital public infrastructure—an international movement helping governments build solutions through common “bricks.”
That is relevant for the agentic state too: what tools and resources could be developed internationally to make it easier for national governments to build their solutions?
It’s an open question. If you have suggestions, I’m happy to take them—we have ideas, but we’re not sure what will be needed.
Alessandro Oppo (28:55)
Something that struck me in the paper was this contradiction: government procurement has impeccable intentions—fair competition, preventing corruption, maximizing value for taxpayers—yet it is channeled into bureaucratic processes that undermine those very values.
This also reminds me of a conversation with Tiago: participatory budgeting can improve political life, citizens are happier, and politicians who implement it have better chances of being re-elected. There are many advantages.
But often we don’t see these methodologies adopted—because of power dynamics, or other political reasons.
Simone Maria Parazzoli (30:20)
There is no purely technical answer here.
Civic tech solutions are increasingly present, but still not widely diffused. And when they are adopted, they are often implemented in ways that do not necessarily empower people.
A key question is whether the output of an assembly should be mandatory for a municipal council. There are good reasons to say yes, and good reasons to say no—because assemblies are not elected. So their output might be taken as an informed recommendation.
More broadly, the low diffusion of civic tech is fundamentally a political problem: there is limited political interest in scaling it. That’s why movements are needed—deliberate efforts to make governments accountable and ensure they hear people’s voices.
When these efforts are effective, they are usually bottom-up.
I’m thinking about Taiwan, where relatively quickly the “top” heard the push and embedded these practices into institutions.
And also SEDEM: there was support from the top, but also strong bottom-up effort to ensure the solution was active and deployed successfully across many use cases.
Alessandro Oppo (32:40)
I have questions—and I also know they might not have answers.
I’m thinking about democracy as we know it now and how it might look in the future.
Sometimes we use the same name for different political systems. We call what we have now “democracy”—representative democracy—but in the future it could be something else, like “agentic democracy.”
Simone Maria Parazzoli (33:34)
How?
Alessandro Oppo (33:36)
For example: how can we make sure people’s voices are actually heard?
I think it’s important that citizens understand they have weight, because sometimes politics feels very far away—especially in Italy.
People have problems in their towns, but politics feels distant. How can politics be more accessible? How can people feel they have power?
Simone Maria Parazzoli (34:24)
Probably the underlying question is whether current representative democracy is working well or not. Big question mark.
But how can we do it better? I think technology can help a lot—on both the GovTech and Civic Tech sides.
On the GovTech side, it means listening more to users and making sure services address real needs—starting from the problem and building solutions that evolve as needs evolve.
On the Civic Tech side, it means using technologies to bring together different voices and opinions. Pol.is is the easiest example of clustering opinions, and it’s widely known.
With more advanced technologies, we can do more: define boundaries, identify groups, and facilitate dialogue between different sets of participants—different groups of citizens.
So I’m optimistic that technology can empower people. But both GovTech and Civic Tech improvements require buy-in and political support.
No civic tech solution becomes central in the life of a polity without political support. That support should not be expected as a gift—it can be fought for through political movements and political effort, including within representative democracy.
There needs to be agreement and power behind the adoption of technology if it is to truly empower people.
Alessandro Oppo (37:41)
In the paper there are several paragraphs about the cost of inaction—what happens if states do not implement technological solutions?
It seems they have to, otherwise the cost of inaction is very high.
Simone Maria Parazzoli (38:13)
We think so.
We believe governments will be better off if they start working on agentic solutions relatively soon, because we expect the world to look quite different in the coming months and years.
There’s a big opportunity: build better services, be more proactive, better address people’s needs, and become more efficient, more effective, more transparent—regaining trust.
So the opportunity is big. The cost of inaction is definitely high, and it will increase over time.
A key question is how unevenly that cost will be distributed across countries—advanced countries versus middle- and low-income countries.
Can agentic AI be an opportunity for leapfrogging? Possibly, but in different ways.
Still: we believe the cost of inaction is high, and governments should experiment now—pilot now—and collaborate to develop the tools, resources, frameworks, and standards needed to build the “railways” toward the agentic state.
Alessandro Oppo (40:21)
This relates to which resources states have.
It also connects to the question “where is the state,” because if AI is hosted by big tech, then…
In the paper you also raised questions like: should governments use open source models, host inside the territory?
And there were questions like: should government provide universal baseline agents to ensure equitable access to agentic services? Or can market-based approaches deliver adequate public benefits?
Simone Maria Parazzoli (41:16)
Yes—these are open questions. We don’t have a definitive answer. We have ideas, but we’re not sure.
The key concept is that we might be seeing new sources of inequality. If agents become the interface between citizens and governments, differences in agentic capabilities could translate into unequal access to public services—and unequal ability to benefit from them.
That is unacceptable, and we should fight it. The question is how.
Will governments be able to provide a trustworthy, satisfactory solution for everyone, while allowing some people to use private solutions? That’s one possibility.
Or maybe governments should lay the foundations—the “railways”—and private actors build on top. Governments develop digital public infrastructure: identity, payments, data interoperability and management. Then private actors build services, with incentives to ensure equitable access.
We don’t know yet.
What is certain is that many private vendors are approaching governments saying, “we have a solution for this and that,” which is exciting.
But today, governments often don’t know precisely what they need in terms of agentic solutions—and they don’t have the means to understand how solutions could be deployed in practice.
That’s why we see our initiative as important: to close this gap and empower governments to take bold decisions toward the agentic state.
Alessandro Oppo (44:23)
I was curious about the feedback you received. From the names, there were many institutions and people working on digital innovation inside states. What was the feedback?
Simone Maria Parazzoli (44:53)
The contributors to the paper participated in their personal capacity, so we are not claiming official endorsement from their agencies.
But what’s exciting is that these people are leading thinkers and doers in government technology across the globe. The fact that they engaged is already an important signal that the vision resonates.
The feedback has been exciting. We received some good criticism too—useful points.
But the main request from governments is: how do we move from vision to implementation?
For us that means: what deliverables, tools, and artifacts can the Agentic State Initiative produce to help governments move from vision to implementation?
And if you have questions or ideas, we’re happy to take them.
Alessandro Oppo (46:52)
I have thoughts more than ideas.
I’m thinking that once there are many agents interacting with each other, there could also be agents that try to simplify the system.
Laws can overlap or contradict each other. Agents might identify inconsistencies and simplify.
If bureaucracy becomes lighter, everything could change.
And then there’s the “when” question—five years, ten years. But those are my thoughts.
Simone Maria Parazzoli (49:02)
Yeah, that’s useful. Very useful.
Alessandro Oppo (49:11)
So many questions, thoughts, ideas, but not many answers.
It’s interesting to think that this could lead to a different society—a different world.
It’s an extremely important topic, but there are no clear answers.
Simone Maria Parazzoli (50:00)
There are no definitive answers, but there are attempts—and that’s what matters to drive action.
We’ve been provocative with our proposal of the agentic state, and now the world feels full of question marks.
The exercise for the next months is to go one by one through each question mark and look closely: how can we actually build this vision?
We need to find balances between the tensions in government adoption of AI and technology.
It’s an exciting attempt. Our humble effort is to get closer to the issues, unveil some of the truth, surface pieces of answers—and hopefully that will help governments move forward.
Alessandro Oppo (51:34)
Now I’m thinking again about A/B testing: if there is a new policy, maybe it can be implemented on a smaller part of the population first. Maybe this approach can also be used for agentic states.
Simone Maria Parazzoli (51:56)
Definitely.
Iteration, testing, experimentation—that will be the bread and butter of the agentic state.
Culturally, that will be one of the biggest shifts. So far, innovation often happens only in a few limited teams—innovation units, digital government units.
The agentic state we envision requires experimentation across whole government activities. That’s a big cultural change, but it should lead to better outcomes.
And when we say better outcomes, in the end it’s improvements for citizens: efficiency, effectiveness, trust—ultimately benefits for people.
Alessandro Oppo (53:23)
From the citizens’ point of view: the paper seems for people working in the field or highly passionate about the topic.
So I wonder: for a normal citizen, what could be their role?
If the agentic state is implemented—and if the cost of inaction is so high that this might be the path—how do we explain it to citizens?
What can citizens do, actively? Right now they seem passive. They could become active if there are citizens’ assemblies and their output becomes an input for the agentic state.
Simone Maria Parazzoli (55:30)
What can citizens do?
First: citizens’ experience in the agentic state will be very different—and much better.
For example, doing taxes would not be a constant back-and-forth between an accountant and a bizarre, contradictory portal. It would be quicker, more efficient, more personalized, without technical problems—tailored to each person’s situation.
What can citizens do to push this agenda forward? Honestly, not much in the direct sense: this is state transformation, and most citizens don’t care about changing public services. They can provide feedback.
But for those who are more curious and active, an important thing is to help build the pieces of the agentic state.
There’s a lot to do: build the vision, refine the nuances.
If you have technical skills—or imagination—prototype what a public service would look like in the agentic state. Explore how procurement could work. If you’re into governance, help define governance frameworks. If you’re into civic tech, help define how participation might change.
This imaginative effort is accessible to everyone. You don’t need to be in government to do it.
I talk with people in government, but I’m not in government. You can imagine alternatives and contribute—and that itself can drive change.
Alessandro Oppo (59:14)
This is something special about the technological space: it allows everyone to participate.
It also feels similar to open source products. The agentic state should be open and transparent, so if there’s a problem, citizens can participate and propose solutions.
Simone Maria Parazzoli (1:00:05)
Yes. In general, the agentic state will be open, transparent, and built on open source solutions as much as possible. That’s the way to go.
The approach is that of open source communities: the code is there for everyone to improve.
Alessandro Oppo (1:00:24)
I’m sure a lot of people will like it.
I also haven’t asked you much about you personally. You mentioned your academic background and work, but would you like to share something more personal? I know you play the guitar.
Simone Maria Parazzoli (1:00:51)
Yeah, I do.
I published some songs with a friend—my music colleague. And I’m working on a new album. Let’s see if it comes out soon.
That’s my effort beyond helping governments build the Agentic State.
Alessandro Oppo (1:01:22)
Do you have a message for people working in similar fields?
Simone Maria Parazzoli (1:01:35)
My main suggestion is: have a look at the vision paper and see if it resonates.
There are so many open question marks. Any input is more than welcome.
Any layer of our framework comes with many problems. We can start with the first one—public services.
For instance: how should the interface of new agentic public services look, as a digital interface? I have no clue. Will it be a chat, like the classic ChatGPT experience? Maybe. It could be part of it.
Then we go to workflows, to regulation.
For example: how will rules as code change under generative AI and new NLP capabilities?
That’s a question we didn’t have the time and space to explore, but it’s important.
Everyone can contribute to this movement toward the agentic state. It’s not really a “movement” in the political sense—it’s governments trying to move toward a vision—but any input is welcome, and we’re excited to build it.
It’s a big effort, but we’re trying to get there.
Alessandro Oppo (1:03:34)
Thank you, Simone. It was very interesting. Do you want to add anything else?
Simone Maria Parazzoli (1:03:43)
I’m just a fan of the podcast, so I’ll enjoy the future episodes. Ciao ciao!
