The infrastructure that makes agent-compatible public services possible

Guest blog by Priit Liivak, Chief Government Technology Officer at Nortal

Citizens are no longer the only users of public services. People already let personal assistants book their appointments, file their expenses, and chase documents on their behalf. Third-party services are springing up to do the same at scale. Non-human traffic already accounts for roughly half of internet activity, and the agent's share of it is the fastest-growing part.

That shift surfaces a problem: a private agent can act on a citizen's behalf, with a valid mandate and the right credentials, and still do the wrong thing. It may submit an application for a benefit the citizen is not eligible for, because the eligibility rules are not clearly stated, or it may submit a form with made up data, because the validation rules require a value it does not have.

What is missing is machine readable semantic layer. The existing service descriptions that explain prerequisites and dependencies have been written for humans and because of this are simplified to retain readers' attention. Agents don't have the same restrictions and can consume much larger sets of information easily. At the same time humans approach public services with a lifetime of background knowledge they never had to articulate, but agents don't have this kind of context. Experiences a human absorbs by living in society must be stated explicitly if a machine is to use the service reliably.

Why data foundations decide the outcome

In a research programme with the Open Data Institute, Nortal examined ten AI use cases across UK councils. The finding was consistent: pilots succeeded or stalled on the data environment around the model, not the model itself. Metadata coverage, agreed identifiers, versioning and automated validation separated the pilots that scaled from the ones that quietly retired. [^1] The same pattern repeats one level up, at service and data level. An agent-compatible government is the product of providing detailed metadata about services and data, and keeping this information maintained over time.

Weak or out-of-date data foundations lead an authorised agent to interpret ambiguity and take the wrong action in good faith: wrong document type, wrong eligibility category, wrong dependency missed. Wrong actions at machine speed produce invalid submissions, rework for caseworkers, degraded register quality, and higher public cost. None of that is the agent misbehaving. It is the agent doing exactly what an under-specified service told it to do.

What differentiates countries, when agents are commoditising, is the environment the agents operate in: whether the rules and services are discoverable, the boundaries explicit and the semantics owned by the public.

What semantic government means

Semantic government is the machine-readable description of public services, rules, data, and the relationships between them. If a web page explains a service to a person, the semantic description instructs a machine on how to use it. The two are not substitutes but rather complement each other.

The minimum digital government semantic layer covers three things:

Regulation — common source; State Gazette provides all the information but often lacks semantics and discoverability. For example, how to extract all regulations applying to companies operating in the shipping industry.

Services — description of the interface between the government, humans, and businesses. Semantic layer here means publishing information about prerequisites, upstream and downstream service dependencies and larger processes where individual services are playing a part.

Data — metadata about the data is not usually sensitive and can assist AI tools significantly. This includes information about the services that provide specific data and semantics within the context of individual public authority.

All of this does not have to be defined on state level. Each institution can describe its own semantic layer on those three dimensions and publish the description openly, versioned, and maintained as public infrastructure. It does not require global consensus on what a "person" or a "business" is. A tax authority's definition of "resident" need not match an election commission's. It just needs to be written down, discoverable, and linked to the neighbouring definition it connects to. Agents read those descriptions directly and move between them through those published links.

Semantics as public infrastructure

If governments do not publish service semantics, vendors will define them. A commercial or community project will apply reverse engineering for each service and become the de facto interface between agents and the state. The interface to the state then shifts from law and public registers to reverse-engineered commercial interpretation. Citizens and authorised agents transact with a re-seller's reading of the rules rather than with the rules themselves, and the government loses the ability to change a service without renegotiating with the layer that sits in front of it.

The remedy is to treat the semantic layer the way public registers are already treated: open standards, named ownership, change logs, procurement conformance. Built that way, semantics become public infrastructure rather than shadow infrastructure, and agents pull their information from the authoritative source rather than from a re-seller.

References

[^1]: Nortal and Open Data Institute, "Insights from UK councils on standards, readiness and reform to modernise public data for AI," 2025. Analysis of ten public-sector AI use cases finding that adoption outcomes were determined by metadata, data standards, ownership and pipeline automation rather than by model selection.

techUK - Transforming Public Services

public_sector_icon_badge_stroke 2pt_final.png

techUK members are transforming public services in the UK. Our community help to shape a smarter, digitally empowered public sector.

techUK drives public sector digital transformation by uniting the public sector and tech industry. Through early market engagement, efficient procurement, and innovative technology adoption, we help to modernise legacy IT, and enable efficient, secure, and personalised services.

Get involved: We run a busy calendar of activity including events, reports, and insights that demonstrate some of the most significant digital transformation opportunities for the sector. Our Transforming Public Services Hub is where you will find details of all upcoming activities. We also send a monthly public services newsletter to which you can subscribe here.