Is your company also one of those who didn't even think of using artificial intelligence just a few years ago? And today, almost 60% of companies in the DACH region are experimenting with AI agents. 10 to 30% are already using AI productively. 1
In order to have a positive effect — e.g. on productivity, costs, efficiency, goals or market position — AI agents must work in a targeted manner.
Because unlike a “normal” IT implementation, AI agents are often the drivers themselves. Some are gradually introducing themselves “from below” — the notorious “Shadow AI” — and transform the business without a strategy. Although other stand-alone solutions carry out tasks properly, they use different guidelines or do not comprehensibly support the strategy of the company or organization.
Or in other words: Some tools or agents initiated by some employees and pursuing some micro-goals are already active in your company without IT or management knowing anything about them.
It is important to “capture” these phenomena and make them usable — as they often also include value-driving ideas — or to replace them with a better, strategic solution.
For example, through a multi-agent system that works reliably, in compliance with data protection regulations and effectively. And this is how agents do it:
If we understand AI agents as “digital employees,” agentic systems are digital teams or departments with several such employees who ideally correspond with each other. And voice models, AI chatbots, pattern and image recognition are the digital assistants. Depending on their skills and setup, the AI agents receive special tasks and competencies to carry out work autonomously.
No, because using current agent technologies, we connect the AI models with secure data via various interfaces and abstract complexity in advance — for example through tool use (standardized access to functions and systems) or via the Model Context Protocol (MCP). MCP is an interoperability standard that can be used to connect systems from different providers and provide structured contexts for large language models (LLM) — such as Figma files, code repositories, SharePoint documents, or Notion tasks. This enables secure integration across system boundaries without the need to implement each connection individually.
At the same time, we check authorizations: The AI agent waits for approval, so to speak, before accessing sensitive (e.g. budget-relevant or public) resources.
On this basis, internal AI agents can go much further than an external chatbot. Instead of just answering individual questions or helping with a task, they plan actions themselves within the given framework. They implement these using tool use and connected systems without the need for constant human control.
Then it is not “spooky” at all, but harmless in practical use. Because AI agents work comprehensibly:
But we have also discovered that less creativity and more reliability are often more appropriate for companies than autonomy. By very clearly defining and testing the individual steps that the agent should complete, the result is all the more effective and precise.
This also makes sense to set up the workflow of the AI agent (s) exactly as it suits the corporate strategy — and not (only) for individual employees.
The Agentic Loops — the loops of reflection and iteration, also known as “Re-Act” (Reasoning and Acting), are particularly exciting. This is where agents differ from simple chatbots, but also from conventional software — and they make the technical solution very robust and efficient.
After each step, the AI agents check whether the result is good and whether the approach achieved the goal in the best possible way. And if not, what mistakes were made. They learn from mistakes and adapt the procedure until they are able to perform their task optimally. This is done without programmers having to intervene and readjust.
As a result, AI agents that are specifically deployed in the right places in the company or organization are suitable for increasing efficiency: for recurring, self-optimizing task solving.
This also helps people: AI agents are an enormous relief for human teams. They experience less stress, have to perform less time-consuming manual work and use their labor for more valuable, exciting tasks — resulting in higher quality and satisfaction.
AI agents can be carefully integrated into almost any everyday work routine. General quality assurance is achieved, for example, through robust guidelines, individual source control and evaluation according to defined criteria.
The human-in-the-loop approach helps to minimize errors and risks — for example due to undetected, changed framework conditions or hallucinations in AI.
People monitor critical steps: In our example above, specialists could be involved to ensure quality assurance before sending the results to external stakeholders such as auditors or supervisory bodies. Or the AI agent highlights uncertain information and sends it to a person as an intermediate step for control.
A “human-on-the-loop” concept is also conceivable if people can intervene at any time in particularly sensitive tasks and stop the process.
Ideally, the task is coordinated with the usual processes in the company or in the organization. In other words, put it in such a way that, just as in “real life,” the AI directly requests the involvement of defined teams on sensitive topics, e.g. for finance or law.
For recurring requirements and questions from our customers, we have developed several ready-made, ready-to-use and specialized AI agents.
These vertically integrated agent systems specialize in a specific industry or clearly defined business processes and are specifically equipped with only the relevant data, tools and rules.
Our most popular AI agents handle sales and market research workflows from start to finish.
Curious about how the individual agents work? We're linking to a few articles that you can dive deeper into.
How Europe can become competitive again with AI - and why now is the time not to lose touch. Read the white paper now!
There is no question that AI is exciting. But the path to successful, meaningful applications is often unclear, uncertain and fraught with challenges. In this article, I want to share with you how we turned initial doubts into real AI success - and what you can take away for your business.
AI is transforming data management: from weather forecasts to medicine. Learn how she analyses huge amounts of data - and how you can benefit from it!
