# Key Lessons for Public Sector Leaders > Single front door, seamless citizen journeys, governed AI agents: the 2026 playbook for deploying agentic AI in public service without ripping out legacy systems. - Date : 2025-09-27 - Lecture : 9 min - Catégorie : strategie-ia - Tags : AI, Agentic AI, Public Sector, Governance - URL : https://www.adservio.fr/en/insights/articles/lecons-cles-pour-les-leaders-du-service-public ## TL;DR - Citizens now expect the same seamless experience from public services that they get from the private sector: this shift in expectations, more than the technology itself, is what's driving agentic AI adoption. - A single front door, orchestrated by interoperable AI agents, can guide citizens in natural language without them having to navigate multiple agency portals. - A decisive three-part test helps select high-impact use cases: technical fit, strategic fit, and mission alignment. - The recommended approach remains the 'thin slice': prove value on a narrow scope before expanding the vision, with ROI metrics defined from the pilot stage. - In 2026, agent governance (traceability, open standards such as MCP, human oversight) has become as decisive as technical performance for scaling a deployment. ## A New Standard for Public Service, Powered by Agentic AI Unified access, connected citizen journeys, empowered teams, and continuous improvement, all without ripping out legacy systems. A new standard for public service is no longer on the horizon. With agentic AI, it's already here. Yet many public sector leaders are only just beginning to explore its potential, and questions remain: Where do you start? What does it take to succeed, and how do you avoid the false starts that marked the first wave of AI projects? To explore what agentic AI can achieve and how leaders can seize the opportunity, we hosted a candid conversation with two leading voices in the field: Richie Etwaru, CEO of Mobeus, and Deon James, Head of Partner Engineering, GenAI Cohort at Google. A year on, this conversation remains a particularly relevant lens: the administrations that have progressed the most since then are precisely those that followed the principles discussed here rather than multiplying isolated proofs of concept. Here are the key takeaways, enriched with the practices observed in the field since. ## The Problem: A Citizen Journey, Not a Bureaucratic Maze Imagine trying to start a small business. You need a license, a tax ID, an environmental permit, and a dozen other approvals. It isn't a single task; it's a frustrating scavenger hunt across a dozen different agencies, each with its own website, forms, and rules. This is the reality of government for millions of citizens and businesses every day. The single biggest force driving the need for AI in government isn't a new technology or a policy, it's a permanent, irreversible shift in public expectations. ### Expectations Shaped by the Private Sector Citizens, having experienced the seamless convenience of banking apps, e-commerce platforms, or mobility services, now expect the same from their government. As Richie Etwaru, CEO of Mobeus, observed: > Governments are starting to see themselves more as an economic balance sheet. If taxes are your revenue and citizens are your customers, you're going to start thinking a bit more about the customer experience. This reframing turns the citizen experience from a 'nice to have' into a core mission priority. It's not just about service delivery, it's about public value and trust in the institution. ### The Hidden Cost of Administrative Fragmentation This fragmentation carries a measurable cost: cumulative wait times, drop-off rates on procedures, citizens re-entering the same information repeatedly, and public agents' workload absorbed by low-value, first-line tasks. Administrations that have mapped these friction points often find that a handful of journeys, starting a business, changing an address, applying for social benefits, account for most of the dissatisfaction. These are precisely the high-volume, high-friction journeys that make the best entry point for agentic AI. ## The Opportunity: A Single Intelligent Front Door The biggest potential of agentic AI in government is the power to eliminate this friction by creating a single, intelligent 'front door' for citizens. Instead of navigating a maze of agency portals, a user can simply state their need in natural language: 'I want to open a coffee shop.' ### From Conversational Agent to Cross-Agency Orchestrator The AI agent then becomes an intelligent guide, a system that not only understands the request but also knows the required steps, agencies, and documents. It shields the user from the fragmented complexity behind the scenes, orchestrating a seamless journey. In 2026, this orchestration increasingly relies on open interoperability standards between agents and information systems, letting a citizen-facing agent talk to the business systems of several agencies without costly point-to-point integration, while keeping a verifiable trace of every action taken on the citizen's behalf. ### Why Ambiguity Is Agentic AI's Playing Field This is possible because agentic AI runs on a fundamental shift in logic: it's built for fluidity, not rigidity, as Richie Etwaru put it. > Where does AI really help? It's when things aren't clear-cut and when they're going to change a lot. That's the non-negotiable principle for deployment: go where the ambiguity is. To succeed, leaders must choose use cases that thrive on flexibility, guiding a citizen through a regulatory maze, rephrasing a complex entitlement in plain language, rather than deterministic calculations that traditional systems already handle very well. > Related read: [How AI Makes Government Services Accessible to Citizens](https://www.adservio.fr/en/insights/articles/comment-l-ia-rend-accessibles-les-services-gouvernementaux): The "no wrong door" approach and the SGX platform unify public-service access with AI: automated eligibility, FedRAMP security, WCAG accessibility. ## The Deployment Blueprint: A Decisive Three-Part Test With a world of opportunity ahead of them, how do leaders avoid the 'AI POC trap', where promising projects never scale? Our conversation surfaced a three-part test for selecting initiatives built for real impact. ### Technical Fit The problem must align with AI's strengths, focusing on cases where flexibility is needed rather than rigid certainty. Think an integrated business portal capable of rephrasing a complex regulation, not a statutory tax calculation that a rules engine already handles reliably and auditably. ### Strategic Fit Avoid recycling old, unfunded projects in the hope that AI will be a magic wand. Instead, identify new problems that couldn't be solved before but now fit AI solutions like a glove, for example, automatically matching a citizen with every benefit they're entitled to, an exercise that was previously impossible to automate given the diversity of eligibility rules. ### Mission Alignment Even well-suited use cases need a strong mission and mandate. As Deon James, Head of Partner Engineering, GenAI Cohort at Google, pointed out, successful deployments are 'mission-driven', tied to clear human or economic outcomes rather than technological novelty alone. This clear sense of purpose, backed by executive sponsorship, provides the political will and cross-agency collaboration needed to turn promising ideas into meaningful impact. Without that mandate, even the best technical agent stays confined to a pilot that never scales. ## Start Small, Think Big: The Thin-Slice Method Embarking on a major transformation can feel overwhelming. The temptation to tackle everything at once (or 'boil the ocean') is strong, but it's a sure path to costly delays and frustration, especially in a constrained public budget environment. ### Choosing the Right Pilot Scope The key is to adopt a 'thin slice' approach: by focusing on a single, high-impact user journey, you can deliver value quickly, prove the concept, demonstrate real ROI, and build momentum for future phases. Deon James captured this approach well: > Prove a solution for a small subset of users, where the most manual tasks happen and where a system can deliver the most value. From there, iterate your development to scale toward the bigger vision. ### Measuring and Demonstrating ROI from the Pilot Stage This isn't about thinking small forever. It's about being strategic: using targeted, successful MVPs to learn, adapt, and earn the right to take the next, more ambitious step. The most advanced administrations instrument their pilots from day one, self-service resolution rate, average handling time, citizen satisfaction, volume of requests deflected from human agents, turning an intuition into a costed investment case for the broader goal of a transparent, citizen-centered government. > Related read: [Creating a Thriving Business Climate with Agentic AI in Government Services](https://www.adservio.fr/en/insights/articles/creer-un-climat-economique-favorable-avec-l-ia-agentive): How Adservio Seamless Government Experience uses agentic AI to simplify business paperwork, environmental permitting, and public-sector compliance. ## Governance, Security, and Interoperability: The Foundation of Trust An agent that acts on a citizen's behalf, accesses sensitive personal data, and interacts with several agency systems cannot be deployed as a simple experimental feature. This dimension, largely absent from the first pilots of 2023-2024, now determines the ability to scale in 2026. ### Agent Governance and Traceability Every action an agent takes must remain traceable, explainable, and reversible: what data was accessed, what decision was made, on what regulatory basis, and with what level of human validation. The most mature administrations define explicit autonomy tiers, an agent can inform and guide fully autonomously, but any action affecting an entitlement or a sum of money remains subject to human validation before execution. ### Interoperability and Open Standards Deploying specialized agents across agencies only makes sense if they can exchange context in a standardized way rather than through fragile proprietary integrations. Adopting open protocols for exchanging context between agents and tools, paired with interoperable digital identity registries, has become as important a technical prerequisite as the choice of underlying language model. It's also what allows an administration to switch AI providers without having to rebuild its entire architecture. ## The Final Shift: This Is the Worst the Technology Will Ever Be Ultimately, succeeding with agentic AI isn't just a technical challenge; it's a profound cultural and organizational one. It demands a new kind of leadership, capable of embracing a dizzying pace of innovation and planning for a future that's arriving faster than ever. This requires a complete mindset shift, which was one of the most compelling parts of our conversation. Richie offered a provocative piece of advice that captures this new reality perfectly: > Remember, this is the worst the technology will ever be. That one idea changes everything about how you plan, invest, and lead. It also means training public agents not to fear agentic AI, but to supervise, correct, and advance the agents they work alongside every day, a new role that deserves as much investment in training as the technical deployment itself. > Related read: [Transforming public services with AI](https://www.adservio.fr/en/insights/articles/transformer-les-services-publics-avec-l-ia): Agentic AI in public services: unified portal, automated eligibility screening, cross-agency orchestration, and a phased approach to transform the citizen experience. ## What Public Sector Leaders Should Take Away From the forces driving change to the new blueprint for purpose-driven AI deployment, it's clear we're at a pivotal moment for government innovation. The administrations that succeed won't necessarily be the ones that deployed the most agents, but the ones that combined a high-impact use case, a clear mandate, solid governance, and a team prepared to iterate continuously. The insights shared here are just the beginning, offering a glimpse into the future of public service. In the full webinar, we unpack the critical cultural shifts needed to thrive, making it a must-watch for any leader guiding their organization into the future. Disclaimer: The statements and opinions expressed in this article are those of the author(s) and do not necessarily reflect the positions of Adservio. ## FAQ ### Why do citizens now expect more from their government? Because they have experienced the seamless convenience of private-sector digital services and now expect the same from their public institutions. This shift in expectations is irreversible and turns the citizen experience from a simple nicety into a mission priority, tied to public value and trust in government. ### How can agentic AI simplify administrative processes? By creating a single, intelligent front door: the citizen states their need in natural language (for example, 'I want to open a coffee shop') and an AI agent orchestrates the required steps, agencies, and documents behind the scenes, relying on open interoperability standards and full traceability of its actions. ### How do you pick the right agentic AI use cases in the public sector to avoid the trap of a POC that never scales? By applying a three-part test: technical fit (favor ambiguous situations where AI's flexibility adds value), strategic fit (target new problems rather than recycling old, unfunded projects), and mission alignment (secure a strong mandate and executive sponsorship). It's also recommended to start with a 'thin slice' instrumented from the pilot stage, then build agent governance before any scale-up.