In 2014, Estonia introduced e-Residency, giving entrepreneurs abroad a government-issued digital identity with which they could establish and run an Estonian company online. The programme connected people to Estonia’s digital business environment without requiring them to move to the country or hand control of their company to a local representative.
More than 142,000 people from around the world have since become e-residents, and e-residents have established more than 43,000 companies. In 2025, the programme generated €124.9M in direct economic impact for the state against €10M in programme-related expenditure. The additional came from the companies and business activity connected to the programme, including work for Estonian banks, accountants, lawyers, payment providers and technology companies.
The next question is whether Estonia’s digital society can support entrepreneurs whose businesses are run in part or in whole by AI agents.
The Eesti.ai Council announced in June that Estonia would explore digital identities for AI agents. The government has described the idea as a way for an agent to act for a person, company or organisation within clearly defined and controllable boundaries. The project is being coordinated by the Enterprise Estonia‘s e-Residency team together with the Eesti.ai team and public and private-sector partners.
Liina Vahtras, former managing director of Estonia’s e-Residency programme, now minister of economic affairs and communications of Estonia
Liina Vahtras, former managing director of Estonia’s e-Residency programme, says: “Just as e-Residency gave global entrepreneurs a trusted digital identity, AI agent codes can give AI agents a trusted and accountable digital identity. This positions Estonia at the forefront of the next wave of digital governance, creating new economic opportunities while reinforcing our reputation as a nation that turns bold digital ideas into practical global infrastructure.”
Since writing this article, Vahtras has been appointed as the minister of economic affairs and communications of Estonia.
The next kind of digital actor
AI agents are different from chatbots. A chatbot answers a question. An agent can carry out a task, with or without supervision, at a set level of reasoning. It may prepare a report, organise information, communicate with a service, file a declaration or prepare a payment after receiving an instruction from its user.
As agents take on more work, the question is no longer whether they can act. Across various systems, from ChatGPT to Claude AI and more, people are already operating their companies through these tools. So the challenge now is how a bank, government agency or business partner can know who the agent represents, what it is allowed to do and who is responsible for the result.
At present, an agent may operate through a person’s login or through a service-specific token. That can make the machine invisible to the receiving service. It also creates a difficult choice for the user. Giving an agent full access to a personal identity may expose far more information and authority than the agent needs.
Estonia’s proposed answer is an agent code: a separate digital identity linked to the person who authorises the agent. One person could use the code with several agents, while the service provider would be able to see that a machine was acting rather than the person directly.
The agent code would not make an AI a citizen, a company or a legal person. The person or organisation granting the authority would remain responsible. The code would instead make delegated authority visible. It would allow permissions to be limited, monitored and revoked.
From reading data to running routine administration
The first proposed use cases are deliberately practical. An agent could summarise health test results, analyse a bank statement, or prepare an invoice payment. A user could allow it to read information, prepare an action for approval and so on. A service provider could apply additional controls of its own.
Estonia’s financial sector is already testing part of this model. LHV has introduced a beta service that lets customers connect account data to AI applications such as Claude and Cursor through the Model Context Protocol. The initial access is read-only, requires customer authentication and consent, can be revoked and expires after 30 days. More companies will probably follow soon with similar releases.
A specific integration grants an agent access to a single service. An agent code could provide a common way to identify the machine and the person behind it across several services. Over time, that could allow an agent to collect information for an accountant, prepare company reports, communicate with public authorities or handle routine administration without using the entrepreneur’s full identity at every step.
This is also why the proposal differs from simply granting AI access to an existing e-ID. The service provider needs to distinguish a human from a machine. The user needs to give the machine only the authority it needs. A company owner may allow an agent to view account information but not make payments. Another agent may prepare a payment for approval. A third may make small payments within a defined limit. The nature of agentic work offers almost infinite possibilities, but requires clear rules and frameworks.
An extension of the e-Residency model
The economic case for agent codes follows the logic of e-Residency. Estonia did not create a market by selling digital identities as an end in themselves. It used identity to connect entrepreneurs worldwide to a functioning digital business environment.
AI is now lowering the cost of starting and running a company. People who previously lacked the skills, capital or team to start a business may be able to use agents to manage parts of its operations. Existing companies are also beginning to use multiple agents for reporting, contracts, customer service, finance and compliance.
Those companies will need a place where their agents can operate with clear rules. Estonia’s opportunity is to make its digital state usable by agents in a controlled way, while creating new activity for local service providers and attracting more AI-era companies to the country.
The proposed system is not a finished product. Legal, business and technical teams are still working on registration, authentication, permissions, audit trails and revocation. The first stage is focused on natural persons, including e-residents. Registration and use by legal persons, as well as the recognition of agent codes across borders, require further analysis.
A separate project led by the Information System Authority, Reason Reserve / Aruait, is examining related questions for autonomous AI agents in the public sector. It is developing an Identity 2.0 framework for machines, a trust registry for public AI assistants and interoperability standards. The projects are related but separate initiatives.
The current project timetable calls for continued legal analysis and draft legislation in September, technical development and testing in October and November, and a pilot with selected use cases and partners in December 2026.
Estonia’s experience with e-Residency offers a useful precedent, now proven as a tool used by hundreds of thousands of people around the world. Still, at the time of its launch, the country began with a practical digital service before anyone could know exactly how entrepreneurs would use it. It learned from real users, expanded the system and built an ecosystem around it.
Agent codes apply the same question to software. If an AI agent is going to act on behalf of an entrepreneur, it needs more than just computing power. It needs a recognisable relationship with the person it represents, a defined mandate and a clear line of responsibility.
Estonia has already built much of the digital infrastructure needed to make those relationships visible. The next experiment is to determine whether that infrastructure can help provide AI agents with a controlled place in the economy.
