AI and the future of government: From chatbots to the agentic state

Artificial intelligence is already changing how governments think about public services, internal operations and interactions with citizens. But adding AI to government does not automatically make government smarter.

This is the topic explored in this episode of Code the State by Yolanda Martínez, Practice Manager for Digital Development for Latin America and the Caribbean at the World Bank, and Oleksandr Iefremov, CEO of the Ukrainian GovTech company Kitsoft, which built the technical foundation of the Diia platform and the world’s first national AI assistant. 

Their perspectives come from different sides of public-sector transformation. Yolanda works with governments across Latin America and the Caribbean on digital development, while Oleksandr brings Ukraine's experience of building AI on top of an established digital public infrastructure.

Together, they explore what AI can actually change for citizens, why agentic government requires much more than an AI model, where governments should draw the line between proactive services and automated decision-making, and why the future of government may ultimately be less about AI itself than about respecting citizens' time.

https://youtu.be/Y888qOfFnKA

Why AI in government needs digital public infrastructure first

Both speakers are optimistic about the potential of AI in government, but neither sees AI as a shortcut around the fundamentals of digital transformation.

For Yolanda, the current attention around agentic AI creates an opportunity. Governments interested in AI are being pushed to confront the foundations they need first: digital public infrastructure, reliable data, interoperable registries and systems that can work together. In that sense, the hype around agentic AI can help motivate digital teams to get those basics right.

Oleksandr makes a similar argument from a technical perspective. Connecting an AI chatbot to broken processes or fragmented infrastructure does not repair what sits underneath. It can simply make the underlying problems harder to see.

Ukraine, he argues, had an important advantage when it began adding AI to public services: much of the underlying infrastructure was already there. He points to digital identity, the Trembita interoperability layer and Liquio, where public services can be represented through structured digital workflows.

This made it possible to connect AI to real registries, APIs and legal procedures rather than limiting it to answering questions. Oleksandr specifically describes connecting AI to services through the MCP protocol and says this architecture allowed the transactional layer to be deployed in weeks rather than requiring new integrations with separate government systems for every service.

His broader argument is that other governments cannot simply skip the infrastructure stage. AI pilots may still produce value, but without the underlying foundation, moving toward an agentic state becomes difficult to sustain.

As he puts it, without that foundation, agentic AI risks becoming little more than a marketing buzzword.

How governments are already experimenting with AI

Yolanda sees growing interest in moving AI from experimentation toward practical government use cases.

She describes an AI accelerator for Latin America and the Caribbean that received 59 proposals focused on three broad areas: improving the design of citizen services, increasing productivity by automating and simplifying internal government processes, and using AI to manage data for better policy decisions.

From those proposals, 18 use cases were selected. They range from healthcare and education to transactional chatbots and tools for assessing AI readiness.

One example comes from Brazil, where a team is using AI to make an AI readiness assessment framework more accessible across different levels of government. This matters particularly in a federal country with thousands of municipalities and significant differences in institutional capacity.

Another initiative brings together Costa Rica and the Dominican Republic around reusable agentic AI building blocks. The goal, as Yolanda describes it, is to create a common technology stack that can make it easier for digital authorities to extend AI capabilities across agencies and levels of government.

But technology is only one part of the accelerator's design. Yolanda repeatedly returns to governance.

Government is typically divided across ministries and agencies, while a citizen journey may depend on several institutions at once. For AI to scale, she argues, governments need a digital authority with enough convening power to bring those institutions together, manage a common technology stack and establish shared approaches to security, safeguards and implementation.

Open-source technology and increasingly accessible AI tools may reduce technical barriers, but they do not remove the need for ownership and coordination.

What AI can change for citizens

For Yolanda, one of AI's most important possibilities is making government services easier to access for people who struggle with traditional digital interfaces.

She points to older citizens as a straightforward example. Opening an app, navigating menus and completing a form may be difficult, while voice interaction could make the same service substantially easier to use.

Language presents another opportunity. Yolanda recalls that when Mexico made printable birth certificates available in the Maya language in 2018, the possibilities were still relatively limited. Today, she argues, AI creates the potential to provide services across many more languages. She also points to work in Kenya involving chatbots that use Swahili to help people access government services.

But AI does not remove the need for user-centered design. Yolanda stresses the importance of focus groups, user-needs assessments, testing interfaces and continuous iteration.

Her recommendation to government digital leaders is practical: create teams that can experiment, build prototypes and learn. Governments do not need to develop every capability alone. They can work with international organizations, academia and local startup ecosystems to build expertise and test new approaches.

AI does not make government services disappear

If citizens can eventually tell an AI what they need and receive the outcome without navigating forms, does the government service itself disappear?Oleksandr argues that it does not.

From the citizen's perspective, AI becomes another interface. A person can speak naturally, type an unstructured request or explain an intention, and AI can help translate that into structured information.

Behind that interface, however, government still needs a highly structured system: registries, standardized procedures, orchestration mechanisms and clear legal foundations.

Someone still needs to define the legal procedure. Someone needs to maintain the data model. Governments still need processes for exceptions that cannot be handled automatically.

Oleksandr expects automation to expand gradually, beginning with simpler procedures and moving toward more complicated ones. AI can help convert voice or unstructured input into the structured data required by government systems, assisting both citizens and public servants.

He also points to a more cautious approach already being used in European governments: applying AI first inside government. For example, AI can help recognize information from paper documents and put it into structured systems while the citizen-facing process remains largely traditional.

Ukraine and other countries, by contrast, are also experimenting with citizen-facing AI chatbots. Oleksandr acknowledges that this approach carries more risk, but argues that digitally confident citizens increasingly expect conversational interfaces because they already use them elsewhere.

Where should government draw the line with agentic AI?

The shift from responsive digital services to proactive government raises a more difficult question: when does anticipating a citizen's needs become making decisions on their behalf?

Oleksandr's answer centers on two principles: consent and reversibility.

A government system might recognize that someone is eligible for a benefit and ask whether they want to claim it. It could pre-fill information, perform checks, pre-approve routine steps or potentially auto-renew certain services.

But, in his view, it should not make irreversible decisions without keeping the citizen in the loop.

The citizen must be able to say no, and the system must be capable of rolling an action back.

Different governments may draw that boundary differently. In democratic systems, Oleksandr argues, the ultimate rules should be established through political and legal processes, including parliament, and then translated into technology.

That means defining what is logged, what can be reversed, what can happen automatically, what requires explicit confirmation and which actions require review by a public servant. It also means ensuring that human review is meaningful rather than simply asking an official to approve whatever an AI system recommends.

How AI changes the work of public servants

Yolanda sees AI not only as a citizen-service technology but also as a way to rethink how government itself works.

Public institutions handle enormous volumes of documentation, analysis and repetitive administrative work. Automating parts of that workload can give public servants more capacity to focus on work that requires judgment and expertise.

She describes the World Bank's own use of customized AI agents to streamline internal procedures. Preparing a lending operation previously took around 18 months, she says, while the organization is moving toward approximately 12 months.

Her point is not that AI is replacing staff. Instead, she describes it as increasing the organization's capacity to respond to demand, reuse previous knowledge and make existing processes more efficient.

But implementation requires more than access to tools. Yolanda emphasizes the need for clear organizational ownership, dedicated teams, opportunities for staff to test new tools and mechanisms for providing feedback to the people building them.

Skills are part of the same transformation. She describes a World Bank partnership with Coursera that provided 5,000 scholarships and was used differently across countries: for students in vulnerable communities, public servants and professionals who needed new AI-related skills. Across more than 12 participating countries, she says AI and cybersecurity were the most in-demand areas.

Her broader point is that AI adoption requires governments to rethink not just technology, but the way public-sector work is organized.

Why interoperability still matters in an AI-driven state

Some of the biggest barriers to seamless digital government have little to do with AI.

Yolanda illustrates this with a personal example. Mexico City offers a digital driver's license as a verifiable credential. Yet when she moved first to Geneva and later to Washington, D.C., she found that authorities would not accept it and still required a physical document.

For her, this captures a wider interoperability problem. Citizens increasingly move between countries, but digital credentials do not automatically move with them.

The same issue extends beyond driver's licenses to education, professional qualifications and other official credentials. International standards may exist, but governments still need to recognize and implement them in practice.

Oleksandr offers a contrasting experience. During the COVID-19 pandemic, he used a Ukrainian digital COVID certificate at a conference in Estonia and was surprised that it worked seamlessly because Ukraine and Estonia had implemented compatible standards.

For him, the experience demonstrates why cross-border interoperability and verifiable credentials matter.

Yolanda also points to work underway in Latin America and the Caribbean to enable mutual recognition of national digital identities through an open-source digital identity broker. Her broader argument is that regional digital integration becomes possible when government teams agree on standards, safeguards and common approaches while respecting national sovereignty.

What the Women in GovTech Challenge shows about building capacity

Technology alone will not create the next generation of digital government. Governments also need people who understand how to design, implement and improve these systems.

Yolanda reflects on this through her experience as part of the assessment team for the Women in GovTech Challenge.

The program combines practical learning about digital identity, interoperability and user-centered service design with technical tools and mentoring. In its third cohort, more than 1,000 people applied for 288 places. Yolanda says the quality of the prototypes improved significantly compared with previous cohorts as participants gained access to a more robust toolkit.

She specifically highlights Liquio as a tool that made a significant difference in the third cohort, enabling participants to build prototypes that could be ready for production if the necessary conditions were in place. She also emphasizes the importance of communities of practitioners who understand modularity, interoperability, digital safeguards and user-centered design. 

The agentic state is ultimately about citizens' time

For Yolanda, the current moment is an opportunity for digital leaders to take ownership of transformation rather than wait for AI to reshape government on its own.

“Today is a huge opportunity that we have to completely transform how government interacts with citizens and build trust between citizens and institutions.” — Yolanda Martínez.

Her closing message is to build teams, work with practitioners across government, academia, startups and international organizations, and use this moment to rethink how institutions interact with citizens and build trust.

Oleksandr concludes the conversation with his perspective on the agentic state: 

“Agentic state is not about AI. It's about respect for the citizens' time.”

Every form a citizen no longer needs to fill out, he argues, gives part of their time back. At the scale of millions of people, that matters more than any single technology.

The goal, in his framing, is not simply a smarter government. It is a government that helps people and delivers more value to citizens.

Listen to the full Code the State episode

This article is based on AI and the Future of Government, an episode of Code the State hosted by Helen Uvarenko, with guests Yolanda Martínez and Oleksandr Iefremov. Listen to the full conversation for their complete discussion on AI, digital public infrastructure and the agentic state.