What It Takes to Build the Agentic State: Q&A with Luukas Ilves

"Agentic AI works by inverting the relationship between citizens and bureaucracy. Success means every citizen interacts with the government through conversation, not forms.”

Behind the world’s most advanced digital nations are the people who built them. In this Q&A, we speak with Luukas Ilves, one of the driving forces of Estonia’s e-state and now an advisor to Ukraine’s Ministry of Digital Transformation. Having helped make Estonia one of the world’s leading examples of e-governance, Ilves is now guiding Ukraine’s next leap - toward the Agentic State, where AI agents become an active part of governance itself.

Q: You’ve been doing a lot of work around the “Agentic State.” How do you personally define it, and how can it transform government or governance in practice?

AI agents combine the ‘brain’ of reasoning systems (LLMs) with the ‘hands’ of automation (APIs, RPA, digital tools), creating software that can perceive, reason, and act with minimal human supervision. Agentic systems can manage end-to-end processes, learn, self-optimize, and collaborate with humans and other agents.

The Agentic State is about using AI agents throughout public administration - and enabling individuals and businesses to interact with public services using AI agents.

In practice, this means services that flow around people's lives rather than forcing them through bureaucratic mazes. When someone says, "I need help after my house was damaged," agents orchestrate responses across insurance, housing, permits, and utilities without requiring the citizen to understand which department does what. Workflows become self-orchestrating, regulations adapt based on real-world outcomes rather than waiting years for legislative updates, and compliance shifts from periodic audits to continuous monitoring.

The transformation is from doing things right - following procedures perfectly - to doing the right things - achieving the outcomes citizens actually need. This isn't about replacing humans but freeing them from routine tasks to focus on judgment, empathy, and strategic thinking.

Q: Should we imagine the digital state as a single, intelligent system, or as a network of smaller, domain-specific agents working together?

With the current state of technology, I would say it’s going to be a network of networks, with humans setting up complex multi-agent systems. We also need to look at the bigger picture, which isn’t just the footprint of agentic systems operated by public administration, but also the agents acting on behalf of enterprises and individuals that will talk to those systems.

Q: Beyond Ukraine and Estonia, where else do you see the Agentic State taking shape? Are there any emerging or unexpected cases globally that inspire you?

All over the world. The barrier to entry for building a standalone agent for a specific function is minimal. There are cool projects in Latin America, Africa, and Southeast Asia.

The hard thing will be to scale those into production (with issues of interoperability, data quality). Here, established govtech powers have an advantage, but we will also see some surprising new entrants, both emerging markets and mature countries that have not been previous govtech leaders. In large, federal countries (like the US and Brazil), some of the most innovative uses will be at the state and local levels.

Q: You helped build Estonia’s digital foundations and are now helping Ukraine design its AI-driven future. What lessons from Estonia’s experience are most relevant, and what should be done differently now?

On a human level, building digital government in Estonia was a massive collaborative effort, without multiple owners and power centers. To build a successful digital state and society over the long run, you need to have initiative and strong delivery all over government, at the national and local levels, and a strong will for everyone to collaborate. In Estonia, this collaborative element comes from the pressure of being small. In larger systems, you have to engineer that collaboration more consciously, or everyone ends up working in their silos (as is the case in most countries).

Today, Estonia is no longer the world’s fastest-moving govtech ecosystem. We have a group of established vendors and bureaucratic planning and procurement processes. Don’t get me wrong - it’s still much more agile than most countries - but Estonia doesn’t move at the speed of startups and agile enterprise IT - or at the speed of Ukraine - anymore. We saw this during COVID, when the Estonian tech sector and volunteers built amazing tools in days, but it took months for the public sector to figure out how to properly procure these solutions.

Q: The WINWIN AI Center of Excellence has set an ambitious goal to transform public administration, defense, and industry through AI. What makes Ukraine uniquely positioned to lead this effort?

Ukraine’s superpower, developed over the last 5 years, has been its ability to turn challenges into an impetus for innovation and broad diffusion. This was true already before Russia’s large-scale invasion: Diia reached nearly 50% of the adult population in 1.5 years (an amazing adoption curve!) because it had a “killer” function - Ukraine tied its COVID pass to the app. Over the past three years, Ukraine has done the same in defense. In the coming years, we are making the bet that Ukraine can repeat this challenge-to-innovation cycle in areas of major national challenges - reconstruction, health, education, energy, mobility - and turn this into growth.

Of course, none of this comes from simply having a difficult situation. You have to combine this with top talent, visionary leadership, and strong execution.

Q: How does WINWIN’s focus on AI sovereignty (developing a national LLM and domestic AI infrastructure) connect to broader questions of national security and independence?

Sovereignty (digital or otherwise) isn’t about building everything yourself - that is a surefire route to failure - it’s about meaningful control over your tech and your long-term development. You achieve this through a combination of your own innovation and control over systems, and using the best that global tech has to offer without ending up dependent on any one provider.

In the past three years, global technology providers have been massive enablers of Ukraine’s sovereignty. Cloud, satellite communications, and cyber tools have kept Ukraine’s society running. Similarly, Ukraine is leveraging global tech (like LangChain and Gemini) to build its public sector agents.

Ukraine’s LLM and domestic AI infrastructure are complements, not replacements, to cloud services and AI models developed by major labs. They will build deep operational understanding in Ukraine, fill in gaps (like understanding the nuances of Ukrainian language), and hopefully incubate niches where innovative Ukrainian solutions conquer the world.

Q: Could Ukraine’s Agentic State model become a new exportable framework, much like Estonia’s e-governance once did?

Yes, but with a twist. Estonia’s e-government model really grew on its own for the first 5-10 years. That made it internally coherent but also made it harder to translate to other countries. The world is much more networked and connected today than in the early 2000s. So I think the model of an Agentic State will co-evolve among a group of leading countries - it won’t be uniquely Ukrainian. But the size of the opportunity (the public sector is more than 30% of world GDP) means that the group of countries that are early in this development will all have massive global opportunities.

Q: You’ve described government as a system made up of many interconnected layers - from how services are delivered to how decisions, budgets, and procurement are managed. Which parts of that system do you think are most in need of reinvention today?

Our Vision Paper on the Agentic State identifies 12 layers of how governments operate and organize themselves that need to be reinvented for the Agentic State.

First, government workflows - the invisible plumbing that determines whether a permit takes days or months. Most are digitized paper trails with bottlenecks hard-coded in. Agents can dynamically sequence tasks, route based on complexity, and continuously optimize processes.

Second, regulatory compliance and supervision. Today's episodic enforcement catches violations after harm occurs. Agentic systems enable continuous monitoring where firm-side agents generate compliance proofs while regulator agents verify in real-time, creating "minimal disclosure, maximal assurance", protecting trade secrets while ensuring oversight.

Third, the data and tech stack of the government are going to need to be reorganized to support the Agentic State. Without reliable data and technical infrastructure (e.g. APIs, model and agent registries, identity systems) that can work with agents, it will be very difficult to scale up agentic systems that really improve on current workflows.

Q: In practical terms, how could AI help fix what’s “broken” in government - not just speeding things up, but changing how public administration works at its core?

Agentic AI works by inverting the relationship between citizens and bureaucracy. Instead of forcing people to understand government structure, agents translate user intent into action.

Take procurement: today's months-long RFP cycles exclude small firms and inflate costs. Agentic procurement can, in contrast, continuously scan markets to find market offerings, negotiate at machine speed within policy constraints, and adapt contracts based on performance.

Or a complex life event like the birth of a child. When you have just gotten home from the hospital, tired and ready to start life with a new family member, the last thing you want to think about is scheduling doctors’ appointments or filling out forms for different government registries. The agentic solution just says, “Solve this for me.”

Q: You’ve highlighted Ukraine’s “build-in-the-open” approach with public betas on Diia and other platforms. Why is transparency so critical in developing AI for government?

Transparency in AI development serves three critical purposes:

• First, it builds trust - citizens see not just decisions but the reasoning behind them. When an agent denies a benefit application, it must show what rules it applied, what alternatives it considered, and how to appeal. This is more transparency than most human bureaucrats provide.
• Second, public betas create rapid feedback loops. When thousands of users test a service simultaneously, edge cases surface immediately.
• Third, openness prevents capture. When code, training data, and decision logs are public, vendors can't lock in governments, special interests can't hide influence, and civil society can verify that systems work as claimed.

Q: What have you learned about citizen engagement and iterative design from this process?

User and customer feedback are one of the most valuable signals in building any product. It’s the difference between a good idea and one that people want to use.

The Agentic State (in Ukraine and globally) is a brand-new idea - Ukraine’s first agent launched this autumn - so we are now only starting to get that feedback. I am sure many of our concepts and ideas of where AI agents add value will turn out to be wrong.

Q: If you were designing a digital state from scratch in 2025, what would be the core architecture or principles you’d start with?

We’ve already touched on them:

• Data governance that powers reliable automation. Are the data good enough that good algorithms and systems will also produce the right outcome?
• Composability that leads to a broad community of builders. Is the barrier to entry low enough that my entire ecosystem can contribute to building new solutions?
• Failure tolerance that leads to reasonable risk-taking. Are safeguards and redress mechanisms good enough that the human consequences of the system making a mistake are minimal, which allows us to make more mistakes and learn faster?

Q: What will success look like for Ukraine in 5-10 years, and what would make it truly agentic rather than just digital?

Success means every Ukrainian interacts with the government through conversation, not forms. You tell Diia what you need - start a business, get medical care, report infrastructure damage - and services assemble themselves. And the government anticipates needs before citizens ask: Your business gets regulatory guidance before compliance problems arise. Benefits arrive when circumstances change, without applications. Infrastructure repairs are scheduled before failures occur.

Behind this simplicity, agents continuously optimize. Reconstruction funds route automatically to verified needs. Defense systems coordinate autonomously while preserving democratic oversight. Regulations adapt to rebuild the economy without enabling corruption.

This contributes to building the society and economy Ukrainians deserve - free, open, prosperous, with institutions that enable, protect, and are accountable to citizens.

The question isn't whether it's possible, but whether we move fast enough to make Ukraine the global model rather than a follower. Success means other countries come to study the "Ukrainian Model" the way they study Estonia today.

About Luukas

Luukas Ilves is a technology and policy leader shaping the next generation of digital governance. As Advisor to the Deputy Prime Minister and Minister of Digital Transformation of Ukraine, he works on advancing AI-driven public services, national digital infrastructure, and the development of Ukraine’s Agentic State. Formerly Undersecretary for Digital Transformation and Government CIO of Estonia, he helped drive the country’s digital transformation and strengthen its leadership in cybersecurity, connectivity, and e-governance.

He is also a Non-Resident Fellow at the International Centre for Defence and Security and a mentor at Creative Destruction Lab Estonia. His work bridges technology, policy, and governance, exploring how AI can renew democratic institutions and state capacity for the digital age.