An article by Karl-Heinz Land - author, investor, founder & CEO neuland.ai AG

Our cities and municipalities are broke. As early as 2024, municipal budgets faced a record deficit of EUR 24.8 billion, and the situation is getting worse. Our public administrations have reached the end of the road, financially and organisationally.

Germany does not have a knowledge problem. Germany has an enforcement problem. We know that our cities groan under rising budget deficits. Maintenance of schools, bridges, roads and parks is inadequate everywhere. Public swimming pools, museums, theatres, opera houses and cultural venues are being closed or cut. The result is frustration and anger among citizens.

We know that citizens wait months for official notices. We know that businesses despair over permits, interfaces, forms and jurisdictions. We also know that the state does not have "too much" administration, but too often the wrong kind: too slow, too fragmented, too paper-bound, too disconnected from its own systems, too weak.

And yet we keep having the wrong debate.

Reform does not mean dismantling

Cutting red tape in Germany is still treated as though public administration simply needs to be made smaller. Fewer posts, fewer rules, less state. It sounds like relief, but more often it is just the political shorthand for being overwhelmed. The reality is harsher: across Germany, roughly 500,000 to 600,000 public servants are missing. On top of that, around 1.39 million employees will retire by 2035. Anyone who, under these conditions, understands cutting red tape as merely striking out administration is confusing reform with capitulation.

The real task is different: we do not need less administration. We need more adequate administration. Administration that delivers the work it promises. That serves citizens and businesses rather than disciplining them through procedures. That becomes faster, more precise, more transparent and more resilient. In short: whoever wants to become a "permit champion" must first become a "process champion".

AI as collaborative intelligence

This is precisely where the democratic explosive force of AI lies.

Not AI as a toy. Not AI as a chat window for bored knowledge workers. Not AI as yet another "tool rollout" that stands around the organisation like an unwrapped piece of gym equipment in January. But AI as collaborative intelligence: a controllable layer across systems, knowledge and processes. A technical infrastructure that relieves people, takes over routines, understands files, respects permissions, orchestrates workflows - and leaves responsibility where it belongs: with human beings.

This is the decisive point. AI must not act as a surrogate government within public administration. It must not simulate democratic responsibility. It must support, verify, prepare, compare, sort, consolidate, explain, accelerate. AI must become "collaborative intelligence"! An AI that takes over routine work and frees up time for what matters. Decision and sign-off remain with people. That is where the constructive power lies: AI does not take responsibility away from employees - it takes away the paralysing part of the work that so often makes responsibility impossible today.

The daily reality of public administration

What does everyday life look like in many administrations? Not politics at its best. Not service to the community. But support work, interface work, searching, copying, checking, requesting documents, transferring, reconciling. Procedures move from system to system, from department to department, from inbox to inbox. Data sits in SAP, in legacy systems, in specialist applications, in files, in documents, in e-mails, in forms. The citizen believes they have filed an application. The administration sees: a case that first needs to be broken down into its component parts.

The consequence is not just inefficiency. The consequence is politically dangerous!

When procedures take months or years, even though they could be completed in days or hours, what emerges is more than frustration. What emerges is the feeling: this system can no longer cope. When businesses wait for permits, when citizens are sent from one office to the next, when public servants experience that the mountain does not shrink despite a heavy workload, then trust in self-efficacy disappears. And when people lose the feeling that their actions make a difference, they resign themselves to their fate - and they vote for the far right.

That sounds harsh. But it is the heart of the matter. Democracy does not live on Sunday speeches, parliaments and constitutional articles alone. Democracy lives on the everyday experience that the state works. That an application is processed. That a permit arrives. That a decision is comprehensible. That a public service does not vanish into thin air. That work has an effect. An administration that paralyses ultimately paralyses democratic trust as well.

That is why AI is more than a productivity tool for cutting red tape. Used properly, it creates hope for a better, less paralysing administration.

The architectural shift: from chatbot to operating system

The neuland.ai HUB describes precisely the architectural shift that is needed for this. The central insight is: it is not enough to hand individual AI tools to employees. Individual productivity gains are real, but the measurable effect at department and organisation level does not materialise if processes remain unchanged, compliance platforms are bypassed and integration into knowledge and systems is missing. That is the German digitalisation trap: we often digitalise the surface, but not the workflow.

The HUB therefore does not start with the chatbot, but with the operating system for AI in complex organisations. It is not a single application but an enterprise AI management and orchestration platform. That sounds technical, but it is politically highly relevant. Because an administration does not need AI gimmicks - it needs a controllable infrastructure: with an SDK, API access and a UI library, so that in-house applications, connectors and agents can be developed and operated. Without new dependencies. Without starting from scratch every time. Without the next vendor building the next silo.

Why European organisations need something different

This is particularly important for German and European organisations. The market is dominated by American providers. Many are strong on SaaS connectors and cloud-native environments. But German administrations and businesses do not live solely in polished cloud demos. They live in SAP, in legacy systems, in mature role-and-permissions models, in regulatory requirements, in data protection, traceability, sovereignty and the reality of public budgets. An AI that does not connect there remains folklore.

The neuland.ai HUB addresses precisely this gap: full system integration while preserving existing permissions. Connectors for standard systems, special support for SAP, and an SDK for proprietary and legacy systems. The existing role-and-permissions model is carried over. This eliminates the absurd alternative of many AI projects: either the AI is allowed to see everything - or nothing. In public administration, both are useless. AI must be allowed to see exactly what is permissible in the given procedure, the given role model, the given context.

Technological sovereignty as a prerequisite

This is also why technological sovereignty is not an ideological quirk, but a prerequisite for trust. The HUB is built on open-source technologies, in nearly all cases under an MIT licence. It deliberately forgoes proprietary hyperscaler services from AWS, Azure or Google as a necessary operating foundation. This is not a nice option but an architectural decision: full technological sovereignty. In addition: full functionality on-premise, in a sovereign cloud and in a conventional cloud - without compromise. For administrations that must reconcile sensitive data, regulatory requirements and political responsibility, this is not a detail. It is the entry ticket.

Making knowledge usable

But sovereignty alone does not make for a fast administration. What matters is the ability to make knowledge genuinely usable. The HUB relies on small language models, a knowledge graph, ontology and agentic RAG. In plain terms: the AI is not meant simply to search documents like a better full-text search. It is meant to understand connections, combine sources, validate intermediate results and work on the basis of an organisation's own data. Terabytes of data can be connected. Small language models are trained on the organisation's own data; processing happens locally, more cost-efficiently and more securely than with large cloud LLMs. The knowledge graph turns scattered data into a semantically structured memory. Agentic RAG actively plans which information is needed, combines sources intelligently and checks intermediate steps.

That is precisely the difference between digitalisation as a filing graveyard and AI as a process engine.

From application to decision: agentic workflows

When an application comes in, not everything needs to be searched, checked, forwarded and requested manually any longer. Agentic workflows can cover entire work steps: from input validation through processing to handover. The role of employees shifts consistently towards decision and sign-off. This is where collaborative intelligence becomes practical. AI handles routine, identifies gaps, compares data, prepares official notices, flags contradictions, documents steps and hands over cleanly. The human decides. But the decision no longer comes after weeks of searching - it comes on the basis of prepared, traceable, verified cases.

Months become days. Days become hours. Citizens and businesses receive faster and better service. Public servants finally get room to breathe again. And the political rallying cry of "cutting red tape" gains an operational core.

Everything prepared, the stamp is waiting: deciding instead of searching. Image: AI generated with neuland.ai HUB

Guardrails and governance

Particularly important here is the combination of generative capabilities and deterministic process steps. Public administration cannot be automated on gut feeling. It needs rules, traceability, clear competences, auditability. The intelligent orchestrator of the HUB knows what it is allowed to do in a given context and what it is not. Guardrails are embedded in the system, not bolted on afterwards as cosmetic configuration. That is the difference between a tool you operate and an agent you can trust in a controlled manner.

On top of this comes a central governance layer: compliance, security, budget control and audit logs are configured once and apply automatically to all applications on the platform. When requirements change - for instance through GDPR interpretations, NIS2, DORA or the EU AI Act - the adjustment does not need to be reinvented for every individual case. Standards such as ISO 27001 are supported, as are sector-specific requirements and monitoring and analysis functions for the EU AI Act. This is precisely what an administration needs that wants to become not just faster, but faster in a legally secure way.

The fiscal dimension

And yes: this also has a fiscal dimension that should not be talked down. For a city like Cologne, the annual savings potential is between EUR 200 million and EUR 300 million. Across Germany, it amounts to roughly EUR 500 billion to EUR 600 billion by 2035. These are not peanuts to be squandered in Sunday speeches. This is the difference between permanent municipal paralysis and renewed capacity to act. Anyone who talks about broken budgets and treats AI merely as a topic for the future has not understood the present.

It is not just about less effort in procedures. It is also about better control. Duplicate payments and fraud can be significantly reduced when data no longer gathers dust in siloed systems, but is intelligently reconciled, checked and traced while preserving existing permissions. An administration that cannot consolidate its own information pays not only with money. It pays with trust. AI can help end this blindness - not through magical omniscience, but through structured, verifiable, rule-bound process intelligence.

AI as a crane, not a wrecking ball

Perhaps the greatest mistake now would be to sell AI as a savings programme directed against employees. That would be morally wrong and strategically foolish. Given the shortage of 500,000 to 600,000 public servants and the additional retirement of around 1.39 million employees by 2035, the question is not whether AI replaces people. The question is whether the state can fulfil its tasks at all if it fails to relieve people through AI. AI is not the wrecking ball for public administration. It is the crane with which administration can be rebuilt.

That is why the debate must grow up. The opponent is not AI. The opponent is the sluggish digitalisation that has spent years pretending a PDF is already progress and an online form is already transformation. The opponent is the idea that complex organisations can be modernised with isolated tools. The opponent is the fear of genuinely redesigning processes. Because that is precisely where many AI initiatives fail: they optimise the individual but do not change the organisation. They create small islands of efficiency but no new overall capacity.

The right ambition

The neuland.ai HUB, by contrast, articulates the right ambition: to move generative AI from individual efficiency gains into genuine, controllable process automation - under the conditions that apply in Germany and Europe. That means SAP and legacy integration. Permissions models. On-premise capability. Sovereign cloud. Open source. Governance. Audit logs. Budget control. Agentic workflows. A marketplace. Coupling to existing systems. A multi-model strategy - frontier models, open source and bring-your-own. Protection against vendor lock-in. In short: not as a shiny tool, but as public infrastructure for a new culture of administration.

Of course AI does not solve every political problem. It replaces neither the setting of priorities, nor democratic debate, nor budgetary decisions. It does not automatically turn bad rules into good ones. But it can make visible where procedures get stuck. It can take over routines that wear people down. It can speed up processing without abolishing oversight. It can make administrative knowledge accessible without ignoring permissions. It can show citizens and businesses: the state is not just there - it works.

Democracy erodes in everyday life

And that is precisely where its democratic opportunity lies.

Because democracy rarely fails through a single great event. It erodes in everyday life. In waiting. In not being responsible. In "please resubmit". In "your case is being processed". In the business that gives up because a permit takes too long. In the public servant who knows what would be right but suffocates in the process. In the citizen who eventually comes to believe that none of it matters.

If AI leads the way out of this paralysis, it does not save "democracy" in the abstract. It saves democratic experience. It gives people back the feeling that applications have an effect, that work has an effect, that politics can have an effect. It transforms administration from a labyrinth into a procedure, from an obstacle into infrastructure, from a symbol of standstill into a promise of agency.

That is why the provocative thesis is right: AI can save our democracy - if we stop misunderstanding it as artificial intelligence and instead organise it as "collaborative intelligence". Not as a replacement for people, but as relief. Not as a black box, but as a controlled process layer. Not as a US cloud fantasy, but as sovereign, European, integrable infrastructure. Not as a tool, but as an operating system for a state that wants to function again.

And the quotation attributed to Friedrich Merz - "In zehn Jahren wird Deutschland ein besseres Land sein - auch dank KI!" [In ten years, Germany will be a better country - thanks in part to AI!] - is therefore either a bold punchline or a political mandate. As a slogan it would be cheap. As a reform promise it is radical. Because in ten years Germany will not be a better country because of AI. It will only be a better country if we use AI to finally build the processes a capable state requires. Then the sentence might prove true. But the "thank you" is not owed to the machine. It is owed to a society that has stopped paralysing its own administration.