The distinction
Most governments use AI. Almost none is AI-native.
If you could remove the model and the service would still work as designed, the service is AI-enabled. If removing the model would make the service impossible, the service is AI-native.
Most governments now use AI. There are assistants on portals, transcription in meetings, summarisation in the back office, translation at the counter. This is AI in government. It leaves the service itself unchanged: the citizen still works out which service they need, still gathers the documents, still applies, still waits.
Most governments use AI. Almost none is AI-native. | Today | AI-native |
|---|
Who starts | The citizen applies | The state offers and the citizen confirms |
|---|
How they ask | Find the service, then the form | Say what they need, in their own words |
|---|
Who does the legwork | The citizen collects and submits | The state does what it already can |
|---|
What the rule is | Guidance a person interprets | Code that executes, tests and explains |
|---|
What the person does | Assesses from scratch | Confirms or corrects — and is accountable |
|---|
Who can act | The citizen, through a browser | The citizen, or an agent they authorised |
|---|
By that test, very little in production anywhere is yet AI-native. That is not a criticism. It is the size of the opportunity.
What it is not
Not a chatbot on the portal. Roughly two in five European government portals already offer automated live support. That is the previous generation, at scale.
Not a count of deployed use cases. Inventories are growing quickly; most entries are pilots, internal productivity tools, or assistive features on unchanged services.
Not a finished state anywhere yet. The gap between the narrative and the volume is universal. Closing it, service by service, is the programme.