The citizen should not be the state's integration layer

Every time you carry a document from one government office to another, you are doing the state's integration work for it. The tax authority already has your address. The civil registry already has your birth certificate. The police already know whether you are free to leave the country. The reason you still collect those facts yourself, print them, and hand them to a fourth office is not that the state lacks the information. It is that the state has decided you are the cheapest way to move it between its own departments.

This arrangement has lasted because it was invisible to the people who could change it. No ministry sees the whole journey. The citizen sees all of it and has no standing to fix any of it. We think artificial intelligence changes that calculation for the first time, not because it makes the forms faster, but because it makes it affordable for the state to carry its own information. Earlier this month we took that argument to the Government Tomorrow Forum in Berlin, where we were asked what it would take for dealing with government to be as simple as saying "I lost my passport" or "my mother needs help". This is our answer.

The chatbot is the manure cart

When the first motor vehicles arrived in Manhattan, the city found an immediate use for them: hauling horse manure out to New Jersey. The new machine was pressed into service for the old problem, and it took years before anyone asked what a city without horses might look like. Most of what is sold as AI government today is the manure cart. A chatbot on a ministry website that tells you which office to contact, which form to download, and which other office to visit first is a faster way of making you do the integration yourself. It automates the relationship instead of changing it. Digitising a broken process gives you a faster broken process.

We hold a harder view than most of our industry on this. A government that buys a conversational front end and leaves the back office untouched has bought nothing. The front end is the easy part; it can be an app, a voice line, or the state's existing super app. The work is in the back office, where the rules of each service have to be modelled, where consent has to be enforced, and where a dozen registries that were never designed to talk to each other have to be made to. That is where the integration the citizen has been doing by hand actually moves, and it is the only place a vendor should be judged.

Invisible has to mean transparent

Here is what the alternative feels like. A resident whose driving licence expires in September asks, in her own language, what a renewal takes. She is told the five things required, and the first is already ticked because her licence is on file. With her consent, the proof of address is fetched from the registry, a photo slot is booked for the next morning, the fee is prepared and shown to her before it is taken, and she approves the whole thing inside one conversation. Then something happens that no form could do. Her residence permit requires a minimum number of days in the country, the border records show that the trip she is about to book would leave her seven days short, and she is told before she asks. A seventy-four-year-old with a flip phone tells the same service, in Arabic, that he has moved in with his daughter, gives his consent once in plain words, and every office that holds his address updates from that single change.

The phrase people reach for when they see this is the invisible state: a government you do not have to understand in order to be served by it, the way you do not need a diagram of your bank to receive your salary. We believe in that goal, and we also think the phrase is dangerous if it is taken literally. A state that works in the background is one step from a state that works in the dark, and people can tell the difference. Across the OECD, only around four in ten people trust their national government, while nearly seven in ten are satisfied with the administrative services they actually use. People trust the counter more than the politics, because the counter is predictable and they can see what it is doing. Switzerland approved a state electronic identity this year by the narrowest of margins, which means half the country declined the foundation that any service like this would rest on.

So the invisible state cannot ask to be trusted. It has to be seen to be trustworthy, every time. Consent is given once, in the citizen's own words, and can be revoked. Records are read at the source and never copied out. Every access is logged where the citizen can look at it. And the citizen, not the state, decides how much of the work to hand over: visibility for the caseworker first, pre-filled applications next, action on the citizen's behalf only once that trust has been earned. The invisible state is not something a government deploys. It is something it is granted, by one citizen at a time, and it can be taken back.

Reveal before you execute

The people who have actually built agentic public services say the same thing, and we agree with them: you cannot jump from paper to agents. You simplify the process in the real world, you pass the regulation that reflects it, and only then do you build. Ukraine did it with a driving licence in 2019 and now has more than twenty-four million people on a state app that is becoming agentic one service at a time. The sequence is correct. Where we part ways with the cautious reading of it is on what to do while the process is still broken, because the honest answer is that it is always still broken.

Administrations have spent thirty years writing reports about simplification without simplifying much, and the reason is not a lack of will. It is that nobody could see the whole process in one place. The ministry that owns the passport does not see the police check, the police do not see the registry, and the one person who sees every step is the citizen, who cannot change any of them. The first useful thing an agent does for a government is not to execute. It is to reveal. A system that has to model every step between "I lost my passport" and a new passport in the post makes every redundant step visible to the people who can remove it, and hands the administration its own map. Laid over a dysfunctional process without that intent, the same technology will entrench the dysfunction. The difference is not in the model; it is in whether revealing was the goal.

Two practical consequences follow. The data has to stay where it is, in the mainframe or the registry that owns it, read in place and never copied out, or no agency will ever agree to connect, and no citizen should want them to. And the first service has to be small, local, and impossible to go back from. A county council owns most of a resident's journey and its own budget, and it can connect two registries without waiting for a national framework. Pilots fail when nobody has defined what good looks like at the end and worked backwards from it. They succeed when the first service is modest, visible, and better than the counter it replaces.

The demand already exists

The usual conclusion to a conversation like this is that the shift to an agentic state has to become a political project, that it has to be made possible for the administration, and that it has to be made desirable for citizens. The first two are true. The third is already done. Citizens have been doing the state's integration work for free for decades, and they have never once been asked whether they wanted the job. The first government that stops asking them to will not have to explain why. It will only have to show that the service is honest about what it knows, asks before it acts, and can be switched off.

The technology to do that exists today. We know because we brought it into a room full of people who have run ministries and built state apps, and nobody argued the point. What they argued about was everything around it, which is exactly where the argument belongs. We will keep contributing to that work. The part that is left is harder than the technology, and it is the part worth doing.