How an AI speakerled me to Agent Residency

In 2025, we created an AI keynote speaker called Lydia K, inspired by the Estonian poet Lydia Koidula. After she spoke at an event in Tallinn, other organizers began asking whether she could appear at theirs.

I wanted to know whether people found this useful enough to pay for. So I asked. They were willing, and in June, Äritehnoloogia reported on Lydia K’s paid appearance.

A person or company still had to handle the contract and payment. I also asked a bank about an account for an AI agent. As I understood its response, identity was central to the question.

That experience became the starting point for my work on Agent Residency. If an AI agent could do work people were willing to pay for, how could it take part in the economy on behalf of people and companies?

I began developing a framework around that question. An agent would need an identity that a service could verify, a clear connection to whoever authorized it, and limits on what it could do. The person or company behind it would remain accountable.

Estonia seemed a useful place to explore this. Its digital identity systems, X-Road and e-Residency program offered infrastructure to build on and people with practical experience of using it. I wanted to find out how much we could do with what was already there.

I discussed the idea with Tarmo Virki, who helped bring it to a wider audience through his December 3, 2025 article in Äripäev. His accompanying LinkedIn post describes our earlier conversation.

The following day, I sent a written proposal, “E-residency to AI-Residency — Pilot for Estonia,” to Liina Vahtras at e-Residency and contacts at the Ministry of Economic Affairs and Communications and Accelerate Estonia.

I proposed three to five scoped sandbox use cases. We would test real tasks, identify where existing systems and law fell short, and use those findings to recommend what needed to change. I wanted practical evidence to guide those decisions.

Company administration for e-residents was one of the use cases I presented in December. These were people who already used Estonia’s digital services and could help us understand which tasks agents might usefully handle.

On February 5, 2026, I published the Agent Residency manifesto and shared it with people involved in the discussions.

The manifesto explored how Estonia could extend its experience with digital identity to agents working for people and companies. It proposed verifiable agent identities, explicit authority and records of their actions. It described different levels of autonomy, with oversight and a way to withdraw permission. The starting point was controlled experiments with a small number of agents and organizations.

There was an economic ambition too: make Estonia a place where entrepreneurs could build useful agent services on infrastructure they could trust. The website has since been redesigned and expanded; an April 2026 archive preserves the earlier page.

On April 1, I sent an institutional briefing proposing a scoped pilot intended to operate within current law. A person would authenticate using their existing digital identity, then authorize an agent to perform a limited task.

One example was preparing an annual report. The agent could collect company data and pre-fill the report, then send it to the director for review and signature. Its authority would expire or could be withdrawn, and its actions would be recorded.

The briefing also mapped applications in finance and healthcare: agents making payments or managing company treasury within set limits, and agents accessing patient records under a doctor’s authority, with access logged. The proposed pilot would cover one sandbox workflow, with legally binding actions confirmed and signed by a person.

The documents were one part of the work. Over several months, I brought the proposal directly to government, published the manifesto, sought feedback from people in technology and research, and answered institutional questions about its application. The April briefing followed a direct request from e-Residency for more detail.

I authored Agent Residency and continued developing it through that process. My contribution included the technical framing, the proposed pilot approach and the sustained work of bringing both to government and other stakeholders. That work has a specific, documented history within Estonia’s wider AI efforts.

On June 17, the government announced plans to create digital identities for AI agents.

I'm the architect of Agent Residency, the framework that informed Estonia's national initiative to give AI agents digital identity. I brought this work to the Estonian government in December 2025.

The announcement brought wider attention and difficult questions. What would registration establish? How would a service check an agent’s authority? What could existing systems already do?

On September 29, I hosted the Agent Identity Roundtable in Tallinn to discuss practical cases and objections. Some participants questioned whether agents needed a separate identity at all. Others asked how authority could be verified across organizations and what evidence would remain afterward. The anonymized brief keeps those disagreements visible.

Through V, we also submitted seven practical comments during the agent-code bill’s consultation. The Ministry accepted one and partly accepted another.

The work on Agent Residency continues. Its open specification brings together four parts: identity establishes which agent is involved; authentication checks its credentials; authorization determines what it may do; audit preserves evidence of its actions.

I want those checks to work across services. An agent might need information from a government service, an accounting platform and a bank to complete one task. Each should be able to verify the authority relevant to its part of the work.

Existing identity and authorization systems already address parts of this. Testing should establish what they cover and where additional work is justified. Identity and permission also cannot guarantee that every decision an agent makes will be a good one. Human review and careful evaluation remain important.

The specification is published through V, our autonomous institutions lab. Our implementation work continues through Agency.AI.

V’s work on governed AI agents is included in a Horizon Europe project selected for funding. Our planned contribution is to test how agents can handle company administration for e-residents, with explicit delegated authority and records of their actions.

I live and build in Estonia, and I want more people here to have a part in developing useful AI. Through Founderly, we have run Tallinn AI Week, New Nordics AI Week, AI/ML Forum, TI-Kirjaoskus Päev and EstoniAI meetups to attract global AI companies and investment to Estonia, and to help people here build skills and find opportunities in AI.

I’m grateful to everyone who brought agent identity onto the agenda at the Prime Minister’s roundtable. E-Residency’s founding team helped me discover Estonia in the first place. Their work showed how a small country could build something useful for people around the world.

Lydia K gave me a concrete starting point. My ambition now is to make it easier for people to give agents useful work while retaining control over what they are allowed to do.

If you build agents, operate services they need to access, or work on identity and security, I invite you to read the specification and share a practical case with us. Examples of where the approach falls short are especially useful. You can reach me at ai@v.ee.