Was ist ein KI-Agent einfach erklärt?

Ein KI-Agent ist ein Softwaresystem, das ein Ziel erhält und selbstständig Arbeitsschritte ausführen kann, um dieses Ziel zu erreichen. Dazu kann es Informationen beschaffen, Werkzeuge einsetzen und andere Anwendungen ansprechen.

Was bedeutet Agentic AI?

Agentic AI bezeichnet KI-Systeme, die stärker eigenständig und zielorientiert handeln. Sie reagieren nicht ausschließlich auf einzelne Prompts, sondern können Aufgaben planen und über mehrere Schritte weiterführen.

Was ist der Unterschied zwischen einem KI-Assistenten und einem KI-Agenten?

Ein Assistent stellt Informationen oder Empfehlungen bereit und unterstützt den Menschen. Ein Agent kann zusätzlich selbst Aktionen ausführen und einen Prozess weiterführen.

Ist ChatGPT ein KI-Agent?

Ein reines Chat-System ist zunächst ein generatives KI-System beziehungsweise Assistent. Werden ihm Werkzeuge, Berechtigungen, Memory und die Möglichkeit gegeben, Aufgaben über mehrere Schritte selbstständig auszuführen, kann es Teil eines agentischen Systems werden.

Sind KI-Agenten immer autonom?

Nein. Es gibt unterschiedliche Autonomiegrade. Viele Enterprise-Anwendungen arbeiten mit Human-in-the-Loop oder Human-on-the-Loop.

Was ist ein Multi-Agenten-System?

Dabei arbeiten mehrere spezialisierte KI-Agenten zusammen. Jeder übernimmt einen bestimmten Aufgabenbereich; eine Orchestrierung koordiniert ihre Arbeit.

Können KI-Agenten andere KI-Agenten beauftragen?

Ja. In Multi-Agenten-Systemen können Agenten Aufgaben untereinander weitergeben oder spezialisierte Agenten aufrufen. Dafür entstehen standardisierte Agent-to-Agent-Protokolle.

Ersetzen KI-Agenten klassische Automatisierung?

Meist ergänzen sie diese. Regelbasierte Workflows bleiben für stabile, eindeutige Abläufe sehr effizient. Agenten bieten zusätzliche Flexibilität bei komplexen und variablen Aufgaben.

Brauchen KI-Agenten immer ein LLM?

Nein. Agenten existieren als Konzept bereits deutlich länger als Large Language Models. Moderne agentische Enterprise-Systeme nutzen allerdings häufig LLMs für Sprache, Planung und Kontextverständnis.

Welche Daten brauchen KI-Agenten?

Nur diejenigen, die für ihre Aufgabe erforderlich sind. Dazu können Unternehmensdatenbanken, Dokumente, Fachverfahren, CRM- oder ERP-Systeme und externe Datenquellen gehören.

Können KI-Agenten mit Legacy-Systemen arbeiten?

Ja, sofern sich die benötigten Funktionen und Daten über geeignete Schnittstellen erschließen lassen. Gerade die Integration historisch gewachsener Anwendungen ist ein wichtiges Einsatzfeld.

Können KI-Agenten vollständig On-Premises betrieben werden?

Ja. Abhängig von Modellen und Architektur können Agenten vollständig im eigenen Rechenzentrum betrieben werden. Andere Anwendungen nutzen Public Cloud, souveräne Cloud oder hybride Architekturen.

Wie verhindert man, dass ein Agent unerlaubte Aktionen ausführt?

Durch begrenzte Berechtigungen, kontrollierte Tools, technische Identitäten, Sandboxing, Freigaberegeln, Monitoring und Abschaltmöglichkeiten.

Wer haftet für Entscheidungen eines KI-Agenten?

Ein Agent übernimmt keine organisatorische Verantwortung. Unternehmen und Behörden müssen Zuständigkeiten und Verantwortlichkeiten für Entwicklung, Betrieb und Nutzung klar regeln. Die konkrete rechtliche Bewertung hängt vom jeweiligen Anwendungsfall ab.

Fallen KI-Agenten unter den EU AI Act?

Ein Agent kann unter den AI Act fallen, sofern er die Definition eines KI-Systems erfüllt. Entscheidend für die Anforderungen sind der konkrete Einsatz und die damit verbundene Risikoklasse. „Agentic AI“ bildet keine eigene Kategorie des AI Acts.

Wie beginnt man mit Agentic AI?

Am besten mit einem klar abgegrenzten, messbaren Anwendungsfall. Ziel, Datenzugriff, Berechtigungen, menschliche Kontrollpunkte und Erfolgskriterien sollten vor der technischen Umsetzung feststehen.

Sind Multi-Agenten-Systeme grundsätzlich leistungsfähiger?

Nein. Sie sind sinnvoll, wenn Aufgaben sinnvoll auf unterschiedliche Rollen verteilt werden können. Für kleine und klare Anwendungsfälle ist ein einzelner Agent oft wirtschaftlicher und leichter kontrollierbar.

Was ist die größte Herausforderung beim Einsatz von KI-Agenten?

In Unternehmen ist es häufig weniger das KI-Modell als die kontrollierte Integration in Daten, Prozesse und bestehende IT. Hinzu kommen Governance, Sicherheit, Verantwortlichkeiten und der richtige Autonomiegrad.

Germany Stack

The Deutschland Stack creates a common technological foundation for the digitalisation of public administration. Through uniform standards, reusable components and open interfaces, it aims to help improve the integration of digital public services, make it easier to adopt existing solutions, and strengthen the digital sovereignty of public administration.

This page explains what lies behind the Deutschland Stack, the building blocks and principles that shape it, and what its introduction means for public administration, businesses and citizens.

Adobe Stock ©

What is the Deutschland-Stack and what does it comprise?
Why is the Germany Stack needed?
Is the Deutschland-Stack mandatory?
What is the Deutschland-Stack standard?
FAQs on the Deutschland-Stack
Five principles for an effective Deutschland-Stack
Making the Deutschland-Stack operationally viable

What is the Germany Stack and what does it include?

The Deutschland-Stack is the national sovereign technology platform for the digitalisation of government and public administration. It is intended to create a common technological foundation on which the federal government, the Länder and local authorities can develop, operate and interlink digital administrative services.

However, the Deutschland-Stack should not be understood as a single piece of software or a central government platform. It comprises technical standards, architectural principles, reusable components, as well as organisational and strategic frameworks. The aim is to create a public administration IT system in which solutions work together more effectively, are reused more frequently and can be deployed more easily across different levels of government.

A key component is the so-called ‘platform core’. The basic services defined to date include, amongst others:

Identity and trust: for example, eID and, in the future, the EUDI wallet for digital identification
Data exchange: FIT-Connect for the standardised transfer of data between IT systems
Data and evidence retrieval: NOOTS for the cross-agency retrieval of existing evidence
Payment processing: the Central Basic Service for Payment Processing, including ePayBL
Mailbox and communication: ZaPuK for more standardised digital communication with the public administration

From a technical perspective, the Deutschland-Stack ranges from infrastructure, development, security and operations, through platforms and core services, to applications, access points and overarching guidelines for strategy, architecture and governance.

Why is the Germany Stack needed?

IT in public administration in Germany has developed in a federalised manner over many years. As a result, there are now numerous different systems, specialist procedures and ad hoc solutions at federal, state and local government levels. This is not inherently problematic. However, difficulties arise when similar functions are developed multiple times or when systems cannot easily communicate with one another.

The Deutschland-Stack therefore primarily addresses three structural challenges:

Fragmentation: There are often different solutions for comparable tasks. This makes reuse difficult and increases the effort required for development, operation and further development.

Lack of reuse: Similar digital functions are sometimes developed multiple times. Shared basic services and platforms are intended to ensure that existing solutions can be reused more frequently.

Lack of interoperability: Different systems and data structures lead to complex interfaces, media breaks and duplicate data entry. Shared standards and interfaces are intended to simplify data exchange.

Added to this is the aim of greater digital sovereignty. Public administration should be able to design its core digital functions in such a way that they remain controllable, secure and, as far as possible, independent of individual technology providers in the long term. The Deutschland-Stack therefore focuses, amongst other things, on open standards, open source, interchangeable components and greater consideration of European providers.

Is the Germany Stack compulsory?

The Deutschland-Stack is not universally mandatory for every public authority, every existing IT system or every business. However, binding requirements have now been adopted for certain areas.

In March 2026, the IT Planning Council established the defined technical standards as the binding basis for Deutschland-Stack solutions. For new and further developments, the federal government, the Länder and local authorities are also aiming to use the designated solutions and components.

In June 2026, the binding nature of these requirements for key basic services was further clarified: the federal states have undertaken to integrate the components for identity and trust – comprising eID and the EUDI wallet, FIT-Connect and NOOTS – and to make them available and use them nationwide. Operating and funding models are still being developed for other components, such as payment processing and the digital mailbox.

However, this does not mean that existing specialist procedures must be completely replaced in the short term. The Deutschland-Stack is being rolled out in stages. How existing solutions are adapted, integrated or, in the long term, migrated depends, amongst other things, on the specific procedure, cost-effectiveness and the technical and organisational framework conditions.

It is also important to note that the publicly available Tech Stack Roadmap currently serves primarily as a guide and a basis for further development. In itself, it does not yet constitute a legal or contractual basis.

What is the Germany Stack Standard?

There is no single ‘Deutschland-Stack Standard’. Rather, the Deutschland-Stack defines a set of common technical standards and architectural principles that solutions should be based on.

The standards cover various areas of technology. These include, for example:

Standards for APIs and interfaces such as REST and OpenAPI
Data and document formats such as JSON, XML or PDF/UA
Guidelines for cloud and managed services
Standards for IT security, including BSI guidelines
Standards for DevSecOps and software development
Standards for semantic technologies
New standards and protocols for artificial intelligence, for example for communication between AI agents

In addition, there are architectural principles such as API-first, loose coupling, reusability, zero trust, end-to-end digitalisation and the preferred use of European solutions.

The so-called ‘tech stack’ thus describes the technical layer of the Deutschland-Stack: it defines which technologies, standards and principles are relevant for interoperable, secure and future-proof public sector IT. The guidelines are continuously being developed, as technologies and requirements also change.

FAQ on the Germany Stack

The Germany Stack operates largely in the background. Citizens should therefore not interact with the Stack itself, but rather experience its effects when using digital public services.

In the long term, for example, digital identities, mailboxes, payment functions and the exchange of supporting documents can be used in a more standardised way. The ‘once-only’ principle is particularly important here: data or supporting documents already held by a public authority should not have to be resubmitted time and again, provided there is an appropriate legal basis and consent.

As a result, dealing with public authorities could in future function more as a seamless digital process: identifying oneself, submitting an application, providing the necessary supporting documents, making a payment and receiving feedback, without having to switch between different systems or analogue and digital channels.

However, these changes are taking place gradually. The Deutschland-Stack will not replace existing administrative services overnight.

For businesses, the ‘Deutschland-Stack’ has two different meanings.

As users of public administration services, businesses are set to benefit in the long term from simpler and more standardised processes. Data and supporting documents could be reused more frequently, digital identities and payment options standardised, and the exchange of data with public authorities simplified.

For IT and digital companies developing solutions for public administration, the architectural principles and standards of the Deutschland-Stack are becoming increasingly important. Issues such as open interfaces, interoperability, cloud readiness, IT security, reusability and digital sovereignty should therefore be taken into account right from the start when making architectural and product decisions.

Particularly in the context of public tenders, it is to be expected that compatibility with the Deutschland-Stack and the use of central components will feature more prominently in requirements in future. Companies should therefore continuously monitor the further development of the tech stack and the mandatory standards.

Open source is an important component of the Deutschland-Stack, but it is not the only permissible form of software.

The aim is to reduce dependence on individual manufacturers and to make solutions more transparent, interchangeable and easier to reuse. The Deutschland-Stack therefore relies on open standards and interfaces and envisages a strong open-source approach for any in-house developments required by the state.

At the same time, it should be possible to utilise existing products and market-ready solutions. Among other things, the Deutschland-Stack follows the ‘Prefer Buy over Make’ principle: where suitable solutions are available, they should not be redeveloped unnecessarily. In procurement, particular consideration should be given to European and digitally sovereign offerings.

Open-source and commercial products are therefore not mutually exclusive within the Deutschland-Stack. The key factor is that solutions must be interoperable, controllable and secure, and must be able to be operated and replaced without problematic dependencies wherever possible.

The Germany Stack is not intended to replace existing administrative IT systems all at once. Rather, it is gradually changing the way in which new digital solutions are planned and existing systems are further developed.

Rather than developing a new, stand-alone solution for every project, the first step is to assess which standards, platforms and basic services are already in place and can be reused. Existing specialist processes can generally be retained and linked more closely with one another via standardised interfaces and integration platforms.

This shifts the focus from isolated digitalisation projects towards a shared digital infrastructure. For public authorities, this means fewer bespoke solutions in the long term, greater reusability and easier data exchange. At the same time, a transformation task arises: established IT landscapes must be gradually integrated with the new standards and components.

The Deutschland-Stack is therefore less a single IT project than a long-term regulatory and development framework for the digitalisation of public administration.

The Germany Stack does not have a single launch date. It is already being set up and is due to be developed in stages until 2028.

In 2026, important foundations were laid for this. In March, the federal and state governments agreed on the initial platform core and binding standards. In June, responsibilities, funding and the mandatory use of selected basic services were further defined.

Further components, operating models, governance structures and implementation pathways are still under development. The federal government is due to present a detailed, binding milestone plan for further implementation at the 52nd meeting of the IT Planning Council in March 2027.

The Deutschland-Stack is therefore not expected to be completed by 2028 either. It is designed as a digital ecosystem that will undergo continuous development and is intended to adapt to new technologies and requirements.

Five principles for an effective Germany Stack

Standards and technical building blocks alone are not enough to create an interoperable public administration. Based on practical experience with public sector IT, Materna has identified five patterns that demonstrate how the components of the Deutschland-Stack can be used effectively.

Once Only

Collect data once wherever possible and then make legitimate use of it on multiple occasions.

Human-in-the-loop

AI supports decision-making, but responsibility remains with humans.

Integration rather than migration

Make existing systems compatible, rather than replacing them outright.

Cross-platform orchestration

Reusable platforms link processes, data and business procedures.

Scalability right from the start

Take safety, operation and future use into account right from the architectural design stage.

Making the Germany Stack operationally usable

Materna regards the Deutschland-Stack as a long-term transformation framework, the effectiveness of which is determined by its practical implementation. As a long-standing digitalisation partner to the public sector, Materna combines strategic architectural goals with experience in the development, modernisation and operation of critical administrative systems.

The focus is not on completely replacing existing solutions. Materna supports public authorities in gradually making specialist procedures interoperable, integrating core services, establishing platforms for end-to-end processes and designing secure and autonomous operating models. In this way, the Deutschland-Stack can be translated from a technological regulatory framework into concrete, scalable administrative IT.