Test the Grand Continent. The ideas and analyses you won’t find anywhere else. But you’ll see them everywhere soon. In your inbox starting at eight euros per month
On September 29 last year, Donald Trump inaugurated America.gov, a portal allowing you to pose a question in plain language (how to renew a passport, how to enroll in Medicare) and to obtain an answer drawn from 29,000 federal sites. The site was designed under the direction of Joe Gebbia, cofounder of Airbnb and administrator of Tesla. His first mission within the Department of Government Effectiveness (DOGE) was to digitize the retirements of federal employees, at a moment when the administration was pushing them toward exit. At the launch of America.gov, Donald Trump also thanked Edward Coristine, 20 years old, known at DOGE by the nickname “Big Balls.” The site runs on Gemini, Google’s model, and on Grok, Elon Musk’s. A presidential decree compels federal agencies to link to it as quickly as possible. The digital doorway of the world’s leading power thus bears the signature of an administration that has turned the chainsaw into a government program.
While in France and Europe the democratic state is being challenged for its slowness and its heaviness, artificial intelligence opens possibilities that digital technology alone could not: breaking down administrative silos, moving from a gatekeeping posture to a proactive logic, and organizing permanent feedback loops with users and citizens.
The same tool can also expose personal data, facilitate surveillance, and create dependencies on private actors whose power now rivals that of states.
But the status quo offers no guarantees of safety or quality: it prolongs the complexity of administrative journeys and the existing technological dependencies.
France could set a vast ambition: to become, with its European partners, the world reference for a democratic government in the age of agentic AI. To demonstrate, through the quality of service delivered and the effective protection of rights, that Europe can develop and master cutting-edge technologies and bring them to its citizens, while staying faithful to its democratic principles.
“Hello, State” is no longer science fiction
Recent technological advances allow us to envisage the creation of a complete and ambitious digital public assistant — let us call it “Hello, State.”
This tool would be capable of answering real-life questions (“Hello, State, I’ve just lost my job, what are my rights?”, “My mother is becoming dependent, where do I start?”, “I have a small business project, what steps should I take?”), pre-fill the relevant forms, guide to the right human interlocutor when the situation requires, and report in real time the recurring misunderstandings and friction points, so that they trigger a revision of the rules or the interfaces involved.
That ambition may seem distant. Yet the first bricks of such an agentic government already exist and are being deployed rapidly.
The first level is the informational assistant, already offered by public institutions and private companies. GOV.UK Chat in the United Kingdom and Gov.pt in Portugal allow querying, in writing or spoken form, vast sets of administrative information in a single interface and to answer citizens’ questions from official data. Estonia is also rolling out Bürokratt, a network of assistants designed to facilitate access to the various administrations. On the private side, Polimorphic and CivicPlus market among American local authorities assistants capable of answering residents’ questions.
The digital doorway of the world’s leading power thus bears the signature of the administration that made the chainsaw a government program.
Alexandre Pointier
The second level is the transactional desk, which can process and complete certain procedures. Diia.AI in Ukraine and TAMM in Abu Dhabi already allow certain administrative operations to be performed through conversation: obtaining a certificate of income or residence, renewing a license, or settling certain bills.
The third and final level would be the companion assistant, which follows the citizen through life events (coming of age, career orientation changes, birth, etc.), mobilizes the relevant services, preserves the context of exchanges, and can take initiatives under user control.
Imagine a citizen who loses their job. Their assistant could identify rights to unemployment benefits, prepare the steps with France Travail, check eligibility for other benefits, propose suitable training, and ensure follow-up of the whole process, without the person having to repeat their situation to every administration.
Silos also separate the territorial levels. The school cafeteria belongs to the commune, the middle-school cafeteria to the department, the high school cafeteria to the region, and each has its own tariffs and rules. A family with children in school, middle school, and high school thus juggles three invoices and often three portals. Even without touching the competencies of the communities, “Hello, State” could connect to the different portals, perform the procedures, and present the three invoices in a homogeneous way.
Several actors are already working on the necessary capabilities: TAMM organizes its services around life events and automates some recurring tasks; ServiceNow develops orchestration of procedures across administrations; the United Kingdom is preparing with Anthropic a personal assistant intended, in a first phase, to accompany job seekers. But their integration into a universal service still requires significant work on data, information systems, processes, and responsibilities.
The cost of inaction
The benefits of a “Hello, State” are potentially substantial: a massive reduction in bureaucratic complexity and thus in the denial of rights, increased user satisfaction, and continuous service improvement thanks to the constant incorporation of citizen feedback. AI could also liberate public servants from a portion of repetitive administrative tasks and give them back time to perform their core functions.
The risks are evident: loss of autonomy and de-skilling of public agents, intrusion into privacy, an authoritarian drift facilitated by the technology. Automation could also reproduce, or even amplify, existing errors and discrimination, while making it harder to identify accountability. Automated controls are thus regularly criticized. In the Netherlands, the use of algorithmic risk profiles contributed to a discriminatory system of controls on child care allowances, unjustly affecting tens of thousands of families. The third Rutte government resigned in January 2021 partly due to this scandal. In Australia, the automated recovery of overpayments, called Robodebt, was deemed illegal as early as 2019, and later the subject of a damning royal commission report in 2023.
Against this backdrop, the temptation is strong to slow down, to advocate a moratorium on the use of AI within administrations. Several voices have already risen in this sense. These concerns are legitimate, and the misadventures of online dematerialization invite caution. In 2017, the online switch of vehicle registrations left users with months without a title and allowed private sites to charge fees for a service that was free on the official site.
But by fearing Orwell too much, one risks resigning to Kafka. We must weigh the consequences of non-choice.
The democratic European state is weakened in its model. It has become unable to organize, at scale and systematically, high-quality relationships with its users, whether individuals or businesses, to provide high-quality public services across its territory, and to meet the challenges of climate and technological transitions. 61% of French people report sometimes or often encountering difficulties in their administrative procedures. 43% of French people approve the idea that there should be “less democracy and more efficiency.”
Public opinion of the state is formed first by daily interactions with the administrations.
Alexandre Pointier
The sense of inefficiency is amplified by comparison with the private sector. Users now expect more personalized, more flexible, more easily accessible services, like those offered by private platforms in finance, online shopping, travel planning, etc. These comparisons are partly unfair, because the state does not choose its users, does not select its missions, cannot refuse a complex file. But they cannot be dismissed, as public opinion of the state is formed first by daily interactions with the administrations.
The status quo also carries its own technological risks. The aging and fragmentation of certain IT systems complicate their maintenance and security. Agents already use AI tools, sometimes without an appropriate framework. A poorly designed agentic transformation could indeed multiply sensitive access points and vulnerabilities; but doing nothing does not better protect administrations against cyberattacks or service outages. A bold program would thus be an opportunity to modernize systems, control access, and rebuild public technological competencies.
Inaction would only accelerate distrust of public power, and encourage those who wish to dismantle it or reduce it to its security apparatus. It would also let private companies develop their own administrative assistants, capable tomorrow of informing citizens, knowing their personal situation, and acting on their behalf. The movement is underway. In the United States, HelloGov, which recently merged with visa specialist iVisa, is preparing passports and visas for private individuals, for a fee. The company has also launched GovSchema, an open standard designed to make administrative procedures directly readable by AI agents. Even when open, this standard was born from the initiative of a commercial actor, for lack of a state to take it up. It is this actor who is beginning to write the grammar of future interactions between machines and the state. The gray cards have shown this at a small scale: where the state fails, private intermediaries take over, without the same guarantees of transparency and privacy. The gateway to public services could thus gradually escape from the administrations themselves.
Yet this intermediation is only the first degree of dispossession. The second, more serious, occurs when tech companies install themselves at the heart of administrations, take control of data and steer decisions. In the United States, ICE, the federal immigration service, relies on Palantir to run its operations. In April 2025, the agency ordered from Palantir, for 30 million dollars, a real-time migrant-tracking platform. The company’s software also equips the Pentagon and several federal health agencies.
The bet of a return of democratic power goes far beyond AI in public services. But transforming the relationship between citizens and administrations could become one of its pillars: one must take the initiative, rather than simply endure the technological and political choices of others.
Proving a democratic agentic government
Far from the Chinese model, where the mobilization of AI by the state can be conflated with the expansion of a surveillance apparatus, and far from the American model, where private companies become essential infrastructures for public data analysis, a French-style “Hello, State” could become the European spearhead for a bold investment in a public technological good.
The mastery of this common good must remain public: data governance, permissions granted to AI agents, and control over decisions. Its architecture must be open and interoperable, to allow the use of several technology providers without creating irreversible dependence. This implies, among other things, that the administration keeps within its own ranks teams capable of auditing its tools and, if necessary, organizing a switch of provider.
Public mastery must not lead to concentrating all citizens’ data in a single file, which would increase risk. The SAFARI project, which in 1974 aimed to interconnect administrative files around the social security number, became the subject of a scandal that led to the French Data Protection Act and the creation of the CNIL. Registers kept by responsible administrations and linked by authorized and traceable exchanges allow cross-cutting services without relinquishing information control.
Two guarantees must be at the heart of “Hello, State.” The first is a firewall: information entrusted by a citizen to obtain a service cannot be used to target controls. This eliminates a major source of risk. Fraud controls retain their own means, under judicial oversight. The second is transparency: each data query leaves a trace, which the citizen can access. To strengthen user trust, Estonia has thus developed a data tracker, which shows who has accessed a person’s information in public registers, when and why. It currently covers only a portion of public databases, but the government plans to generalize the tool. These guarantees must be enshrined in law, but also in the system’s technical architecture, to withstand political changes.
As with any digital transformation, this project will be all the more useful and effective if accompanied by a transformation of the internal functioning of administrations. The difficulties encountered by “Hello, State” in completing procedures will reveal organizational inconsistencies. In order not to simply add an interface to existing complexity, there must at minimum be a program to simplify procedures attached to the project — and not deregulation.
By freeing human time, “Hello, State” should enable a massive redeployment of agents to handle complex situations and, above all, to support people whom automated systems risk excluding. This must be an explicit objective of the transformation. The agents who will supervise the tool should receive solid training in the critical use of AI, to understand its recommendations and know its biases.
But fear of Orwell too much leads one to resign to Kafka.
Alexandre Pointier
A project like “Hello, State” could serve as a demonstrator and incubator for a broader industrial ambition: to make France one of the countries capable of developing frontier AI models, rivaling those of the American and Chinese giants. The most advanced projects so far rely on existing frontier models: the United Kingdom works with Anthropic, America.gov with Google and xAI. Sovereignty will of course not be determined solely by AI models. It will also depend on our mastery of the infrastructures that run them — data centers, cloud services —, the software that allows AI to communicate with administrations and act on behalf of citizens, and the standards that set the rules. Europe must develop these capacities while preserving the option to use different models. By making the state a major user and a demanding client of European AI, this project would create a structuring demand for our tech companies and an exceptional testing ground for agentic systems. The complexity of administrative procedures and the demands of reliability and security would pose as many technological challenges whose solutions could then be deployed in other sectors and exported to other states. France could thus combine its ambition for technological sovereignty with a project to transform the public power, whose benefits would be immediately perceptible to citizens.
There is still work to turn this into a political project.
For twenty years, Europe has sought to influence global digital transformation through its normative power, from the General Data Protection Regulation to the AI Regulation. The latter already regards systems that allocate public benefits as “high risk” and will require human oversight and an assessment of their impact on fundamental rights starting December 2027. With “Hello, State,” Europe could become a reference by example: demonstrating that an agentic government can make access to rights effective without sacrificing civil liberties.
France could take the initiative and lead its European partners in this transformation. For the first time in a long while, Europe could assert itself through its own technological achievements, and offer other democracies a model of agentic government to draw inspiration from.