This story should serve as a warning as much as an inspiration. As AI agents become capable of acting on behalf of citizens and businesses, governments are about to gain a new class of users they never planned for. The question is no longer whether agents will interact with public services, but whether the infrastructure beneath those services is ready for them. Governments are investing heavily in providing AI capabilities.
AI agents will use government services whether governments are ready or not.
The more urgent investment may be in the digital public infrastructure that determines whether third-party AI capabilities can use existing government services safely, transparently and accountably at scale.
What breaks when humans stop being the users
Today's public services were designed for human users. Humans understand implied context and take responsibility for their actions. AI agents require machine-readable ways to verify delegated authority, discover services, and understand citizen context and government processes. Without these capabilities, governments face a new category of risk: an agent could confidently use the wrong service, misunderstand a rule, act outside its mandate or make a disputed transaction impossible to attribute.
These are not problems that can be solved by building smarter agents alone. These challenges must be addressed through digital public infrastructure that underpins government itself. Without this foundation, the efficiency gains promised by AI agents risk being offset by errors, disputes, manual intervention and loss of public trust.
Trust must be built into the rails
For governments, trust in an agent-mediated world comes down to four simple questions: Who is acting? On whose behalf? Under what authority? And who is accountable if something goes wrong?
Public services will increasingly need to answer these questions before accepting actions from an agent, just as they verify the identity and permissions of human users today. The objective is not to regulate every AI system individually, but to ensure that the infrastructure itself makes actions attributable, auditable and governable by default.
Two layers of agent-compatible government
Preparing government for AI agents does not require a new AI platform. It requires upgrading the digital public infrastructure that already exists. That investment has two parts. First, government must become machine-trustworthy, with clear mechanisms for identity, delegation, consent, auditability and accountability. Second, it must become machine-readable, with services, rules and obligations described in ways agents can understand and navigate. One makes agent interactions safe and governable; the other makes government legible to agents. Both are essential. Without the first, governments cannot trust what agents do. Without the second, agents cannot reliably understand how government works. Together, they form the foundation of agent-compatible digital public infrastructure.
This matters because government efficiency is about more than processing transactions faster. The most effective digital governments combine operational efficiency with state capacity and public trust. AI agents have the potential to improve the first dramatically. Without the right infrastructure, however, they could undermine the latter two.
The next five years
Over the next five years, three shifts are likely to reshape digital government. Agent identity will become as important for software as digital identity will for citizens, allowing governments to know not only who is being represented but which agent is acting on their behalf. Service design for agents and machine-readable rules and obligations will increasingly become standard requirements in public-sector procurement, making government systems understandable not just to people but also to trusted agents. And legal frameworks will evolve incrementally, service by service, extending existing concepts of delegation and representation into the digital realm.
The most successful governments will not be those that deploy the most AI, but those that build the infrastructure that allows AI to operate safely within democratic principles. Agent-compatible digital public infrastructure is becoming a prerequisite for the three outcomes every government seeks: greater efficiency, stronger state capacity and enduring public trust.
About the author
Priit Liivak is chief government technology officer at Nortal, helping governments build secure, scalable digital public infrastructure and advising on architecture, interoperability and AI-ready public services.
Connect on LinkedIn