From Chatbots to Agent Systems
Most companies have tried chatbots by now. In 2026 the question is a different one: which processes can several cooperating agents take over, with tools, approvals and monitoring? The frameworks have matured. LangGraph has had a stable 1.0 release since late 2025, Microsoft merged AutoGen and Semantic Kernel into the Microsoft Agent Framework, OpenAI offers the Agents SDK, Google the Agent Development Kit (ADK) and Anthropic the Claude Agent SDK. With MCP and A2A there are open protocols for tools and for communication between agents.
That makes the choice of partner more important than the choice of tool. Providers fall roughly into three groups.
The Three Types of Providers
1. Research-driven AI consultancies
They bring data science depth and often their own methods for evaluation and safety. A good fit when the domain logic is complex or the industry is regulated. Examples from our directory: appliedAI Initiative (Munich) with its AI Agent Lighthouse programme, or statworx (Frankfurt).
2. Product and software builders
Software houses that turn an idea into a working product quickly, using existing frameworks. A good fit when you want to validate a use case with a prototype. Example: Netguru, which has publicly documented an internal multi-agent sales assistant.
3. Enterprise integrators
Large IT service providers that connect agents to SAP, Microsoft 365 or ServiceNow and take care of operations and change management. Example: Reply with a focus on Azure and Copilot Studio.
Which Type Fits You?
- You want to test whether a use case works: product builders with a clearly scoped prototype
- Your domain logic is complex or regulated: research-driven consultancies
- You need to roll out across the company: integrators with operations experience
How Our Directory Works
The directory is based on public information and self-reported details. For marked entries we have checked whether the company exists, where it is based, which website is official and whether it publicly offers work on AI agents. The date of the check is shown on each entry. We do not rate the quality of the work, and we do not publish guessed contact details: please contact providers via their official websites.
What DACH Customers Typically Ask For
- GDPR-compliant RAG with access rights at document level
- Hosting in the EU or on premises for sensitive data
- Human in the loop for decisions with legal or financial impact
- Documentation for the EU AI Act, depending on the risk class of the use case
Next Step
Use the checklist for your first conversations and compare providers in the directory by region and stated multi-agent experience.