AI and the future of government

In government, AI is growing from writing aid into analyst, one that compares a case file with earlier decisions and regulations. The writing aid is already routine: when a national government had more than twenty thousand civil servants work with an AI assistant for three months, they reported saving twenty-six minutes a day on average, mostly on writing tasks. Analysing case files and finding the relevant rules and case law already happens. In a scenario for 2040, the analyst goes further: AI places a new case beside earlier decisions and shows where similar cases ended differently, with the sources attached. The civil servant is then handed a judgement call rather than a text: two earlier decisions point the other way, so which line does this case follow? The decision stays with the civil servant.

An agent is AI that takes the steps needed to complete a task by itself. A chatbot explains how to apply for a parking permit; an agent retrieves the details government already holds, fills in the form, spots what is missing and tells the citizen where things stand. Governments are now testing such agents in controlled settings. They are working out which steps an agent can safely take over, and whether citizens then fill in less and wait less. The step after that is bigger: agents that together handle a whole life event, such as moving house or the birth of a child, across every service involved. That is the beginning of government as a single front door.

Much of the work on a permit application is checking: are all the documents there, does the plan fit the rules, are the dimensions and drawings right? AI can do that first check. It sets the application beside the rules and the local zoning plan, and marks what is missing or out of line, with a reference to the relevant article. Automatic checks on single requirements already exist. The next step is AI that assesses many more rules together while the design is still being drawn, along with the effects on the surroundings, such as shade, noise and water. Applications then arrive almost complete and compliant, and the case officer spends their time weighing the interests of neighbours, nature and heritage.

Proactive services reverse the direction: the citizen no longer has to work out what they are entitled to, because government offers it. The starting point is a life event, such as a birth, a move, a death or reaching pension age. In some places, parents already receive child benefit after a birth without applying; they only need to confirm. AI can widen this. In data government already holds, it recognises who is likely to qualify for an allowance or service but is not yet using it. A scheme then reaches the people it was designed for who could not find their way to it. The law sets out which data government may combine for this.

A large public consultation brings in tens of thousands of responses from citizens, businesses and interest groups. Reading and sorting them all took policymakers weeks, and a strong argument was easily buried under repetition. Language models now do that sorting. One government had a language model sort more than fifty thousand consultation responses into themes. It took two hours, and its grouping matched the experts' more often than two groups of experts matched each other. The policymaker can then find the arguments under each theme, including those of the one respondent who sees what others miss. The next step is a conversation rather than a pile of responses: AI that asks follow-up questions of very many residents at once.

In a crisis, every hour counts. The further ahead you can see, the better you can prepare: where will the water rise, which way will the fire move, which district will lose power? AI models combine weather data, satellite images, sensor readings and information on roads and vulnerable buildings, and predict where and when things will go wrong. Flood forecasts arrive days earlier as a result, even in areas with few measuring stations. A crisis team can then evacuate in time or position emergency services differently. Digital twins for disaster exercises are already being developed. The next step is AI that keeps inventing new disaster scenarios and works out the effects for fire services, healthcare, grid operators and local authorities at once. The plan has been tested before the water arrives.

In public procurement, AI brings bids, requirements and contract data together, so differences between tenders show up sooner. It can read through piles of documents, set out per requirement what each bidder promises and flag where a bid deviates or something is missing. It also helps after the award, by tracking whether the terms of the contract are met. One government even gave an AI assistant the title of minister, with procurement as its portfolio and less room for corruption as the aim. AI can already search for patterns across many tenders. In a scenario for 2040, it continuously links tenders, payments and delivered work into one overview of what a government buys, where unusual patterns stand out early.

A digital identity wallet is an app holding your official details, such as your identity, driving licence and diplomas. You show only what is needed, for instance that you are over eighteen, without giving away your date of birth. A European regulation requires member states to offer their residents such a wallet. AI plays a part in identification at a distance: it compares a selfie with the passport photo and picks up signs of forged documents or recordings. Cryptography lets a service check at once whether a credential is genuine. In a scenario for 2040, an agent authorised by the citizen presents exactly the credentials an application asks for, on their behalf. The wallet then becomes the base for a single front door to all of government.

Sovereign AI is AI over which a country or government keeps control: of the data the model was trained on, the servers it runs on and the rules it follows. For government that matters, because it handles sensitive data and does not want to depend on a single supplier outside its own legal system. Several governments are therefore having open language models built and trained on their own language, legislation and public texts. A group of local authorities is testing such a model in their digital assistant. A model of this kind can be narrower than the large commercial ones. What government gains is the ability to set its own requirements for where the training data comes from, how the model is managed and how its answers are checked.

An AI-first government lets AI and new technology do the repetitive and predictive work that civil servants need not do themselves. They focus on residents and on decisions that call for human judgement. AI can sort applications, summarise incoming reports and answer questions about progress. Staff assess exceptions, weigh interests and talk decisions through with the people they affect. That can leave more room for personal contact. It does not happen by itself: it depends on how government divides the work and the time that is freed up. Civil servants also keep watch over what AI does.

AI takes over civil servants' tasks and creates new work at the same time. The algorithm manager keeps an overview of the systems involved in decisions, checks after every update whether the outcomes are still correct and fair, and steps in when needed. The data steward looks after the quality of registers and agrees who may use which data, and for what purpose. The explainer tells a citizen in plain language why their application was turned down and what they can do next. A fourth role is on its way: the agent supervisor, who directs a team of AI agents the way a team leader directs people today. They divide the work, check the results and step in where an agent gets stuck.

A team of civil servants and AI shares the preparation, the assessment and the contact with citizens. Take a department that handles welfare applications or grants. The agent collects data from the various registers, checks whether the file is complete, lists the rules that apply and drafts a decision. The civil servant weighs what no register holds: the applicant's circumstances, or the business owner who falls just outside the letter of the scheme but well within its intent. They call the applicant when something does not add up and explain why a decision turns out differently than expected. The team agrees which draft decisions one civil servant checks and which are reviewed by two people.

Leadership in the AI era does not need a new generation of government leaders. The change starts with the leaders already in post. They set the example by using AI and new technology themselves and sharing their experience with civil servants. They also offer a picture of tomorrow's government that civil servants want to help build. Leaders make room for ideas from civil servants and show how their contributions shape the government of tomorrow. Together, leaders and civil servants build a government that works better for citizens.

A government that wants to work with AI needs technology, money and rules. Yet it often gets stuck on something else: habit. It comes down to the willingness of civil servants and leaders to try a new way of working and then stick with it: letting go of a familiar procedure, leaving a check to a system, reporting when that system gets something wrong. A culture of care protects citizens from arbitrary treatment, and that has real value, because a government mistake can affect someone's income or home. It does make learning harder, because departing from the usual feels like a risk. Some agencies therefore work with sandboxes: defined settings where a team experiments within the law on real cases, with leaders taking part themselves.

In the future academy, staff learn to read AI advice critically and to explain what AI has prepared. Organisations set up such an academy themselves, because AI and new technology change the work faster than training programmes can adjust their curricula. New colleagues practise there with the systems the organisation is bringing in. The material comes from the organisation's own work and is updated as tasks change.

Who does a government build its future with? Start-ups working out new ideas, scientists measuring what works, partners bringing knowledge it does not have in-house, and other governments at home and abroad facing the same questions. Climate, an ageing population, housing and security are challenges no government solves alone. Together they develop solutions and share what works: a model that learns to sort documents in one place can be reused in another, and an approach from one country can help another move forward. That way, an innovation reaches citizens sooner.