Agentic AI Hub: What Companies Can Learn From Government
18 Pilot Projects, 400 Startups, One Insight: The Federal Agentic AI Hub shows how agentic AI works in practice and why the bottleneck is not the technology but the data
Since 9 March 2026, an experiment has been running in Germany that has implications far beyond the public sector. The Federal Ministry for Digital and State Modernisation has launched its Agentic AI Hub – 18 pilot projects in which autonomous AI agents process real applications in real government agencies. What at first glance looks like administrative modernisation is in reality a crash course for every company that takes agentic automation seriously.
| 18 | 400 | 90% | 20,000 |
| PILOT PROJECTS in 19 municipalities | STARTUP APPLICATIONS 10 selected | PUBLIC BODIES PLAN AGENTIC AI Capgemini, May 2025 | SPECIALIST PROCEDURES automatable per BMDS |
What the Agentic AI Hub Is – and Why It Concerns Everyone
The Agentic AI Hub is a virtual, cross-location “engine room” in which the Federal Ministry for Digital Affairs and the Federal Digital Service bring together AI startups and municipal administrations. For three months, from March to the end of May 2026, ten selected startups are testing their solutions in 19 municipalities of varying sizes, from Berlin-Steglitz-Zehlendorf to Flensburg to the Neckar-Odenwald district.
The response to the project call at the end of January 2026 was overwhelming: around 400 startups and nearly 200 municipalities applied. An advisory board comprising representatives from the German Association of Cities, UnternehmerTUM and Merantix selected 18 projects from these, evaluated according to strategic fit, representativeness and scalability.
| “The overwhelming response to our call shows: Germany does not just want to discuss AI, it wants to act. As a country we have the know-how, the ideas and the courage to use AI consistently for our benefit.” — Dr. Karsten Wildberger, Federal Minister for Digital Affairs and State Modernisation, BMDS press release, 9 March 2026 |
The goal is as ambitious as it is concrete: autonomous agents are to check applications for completeness, request missing documents, analyse submissions and produce draft decisions for case workers. The pilot phase is being evaluated by the BMDS and the Digital Service for effectiveness and scalability; a second cohort has already been announced.
Five Pilot Projects That Are Relevant Beyond the Public Sector
Not all 18 projects are equally instructive. Some solve highly specific administrative problems; others show patterns that translate directly to corporate processes. The following five deserve particular attention.
| Municipality/Municipalities | Use Case | Strategic Significance |
| Nuremberg, Munich | Process mining for receivables management and naturalisation procedures | Agentic access to process data, not just documents |
| Berlin, Cologne, Heinsberg | AI-supported application guide for long-term care support | End-to-end citizen interaction with automatic document preparation |
| Borken District | Sovereign AI orchestration layer for public authorities | Infrastructure play: platform-agnostic, scalable, GDPR-compliant |
| 7 municipalities | Process modelling via voice recording in BPMN 2.0 | Automated process mapping as a foundation for agentic systems |
| Neckar-Odenwald | Automated mail processing including GDPR deletion deadlines | Practical immediate benefit with a clear compliance profile |
The Real Bottleneck: Data, Not AI
Perhaps the most important finding from the Agentic AI Hub can be summarised in a single sentence: the bottleneck in agentic automation is not the AI – it is the data. Not one of the 18 pilot projects starts with model selection. All of them start with the question of whether the existing data is even in a form that an agent can process.
A study by the Capgemini Research Institute confirms this pattern well beyond Germany. It found that 90 per cent of public sector organisations surveyed worldwide plan to evaluate, pilot or scale agentic AI in the next two to three years. Yet only 21 per cent of these organisations possess the data foundations that would actually be necessary for training or fine-tuning.
| “The ability to deploy Gen AI and agentic AI depends on having rock-solid data foundations.” — Marc Reinhardt, Public Sector Global Industry Leader, Capgemini (study “Data foundations for government”, May 2025) |
The implication for companies is immediate: anyone wishing to introduce agentic automation should not begin with a proof of concept for the latest language model, but with an honest audit of their process data. Are workflows documented? Is the relevant data available in structured form? Are there clear responsibilities for data quality? Federal Digital Minister Wildberger put the potential at a workshop in January 2026 at 20,000 specialist procedures that could be automated in the federal administration. The figure illustrates both the scale and the prerequisite: each of these procedures must first be understood before it can be automated.
What Companies Can Learn From the Agentic AI Hub Right Now
The federal pilot projects are not an academic experiment. They follow a logic that every company can apply to its own automation journey. Five insights stand out.
| FOUR LESSONS FROM THE AGENTIC AI HUB 1. Data before model. Not a single pilot project starts with model selection. All start with data preparation. Those who do not know their process data cannot deploy an agent meaningfully. 2. Start small, stay concrete. The pilot projects last three months and each solve one concrete problem – not a strategy workshop. Incoming mail. Care application. Naturalisation file. Not ‘digitalising the administration’. 3. Sovereignty is not a buzzword. In Borken District, a sovereign AI orchestration layer is being built. In regulated sectors – whether public administration, law firm or financial services – data sovereignty is a hard requirement, not a nice-to-have. 4. The human stays in the loop. No pilot makes decisions without human approval. The agents make proposals – the case workers decide. That is exactly how it should work in companies too. |
From Pilot to Production – and Why Companies Can Move Faster
The public sector has a structural disadvantage compared to the private sector: procurement law, works councils, committee approvals. Each of the 18 pilot projects must navigate procurement regulations, data protection impact assessments and security clearances. Companies do not have this overhead – or at least not to a comparable extent.
This means: what the public sector pilots in three months, a well-prepared company can operationalise in three months. The key lies in preparation. The same sequence that the Agentic AI Hub prescribes – map processes, prepare data, build the agent, keep the human in the loop – applies to every sector.
Lunatec guides companies along exactly this path: from process discovery through platform selection to the ongoing operation of agentic systems. Experience from dozens of automation projects in regulated industries shows that the companies that progress fastest are those that do not start with the technology but with their data and processes. The Federal Agentic AI Hub now confirms this insight at national level.
Those who do not wait for the public sector to present its results, but begin their own process mapping today, gain a considerable advantage – over the competition and over the regulator, who with the EU AI Act from August 2026 onwards will increasingly ask for documentation and compliance.
ABOUT LUNATEC
Lunatec, headquartered in Frankfurt am Main, guides companies, law firms and the public sector in the introduction of agentic automation. As a UiPath Diamond Partner and Microsoft Partner, we combine platform expertise with regulatory understanding – from process discovery and architecture through to ongoing operations.
lunatec.de · Frankfurt am Main · Shape the Automated World