What changes when AI no longer just answers, but takes action?
Tino Mager: AI is the new colleague. Take logistics, for example. A driver photographs a delivery document. The AI agent identifies the details, checks the address and completeness, transfers the data to the ERP system, and initiates the next step in the process. The case no longer moves from desk to desk for manual data entry. People can focus on unclear or incorrect cases.
AI has made its way into companies: According to Eurostat, one in five EU companies with at least 10 employees used artificial intelligence in 2025. Bitkom reports current figures for Germany: 41 percent of companies with at least 20 employees now use AI, while another 48 percent are planning or discussing its use. The experimental phase is coming to an end. AI now needs to demonstrate the concrete value it delivers for processes, employees, and customers. Through the study „Der ROI-Kompass für KI im Mittelstand,“ Telekom MMS and mind digital collected data and insights from 100 early adopters on maturity, investment rationale, and measurable success.
Sven Giesselbach: The next stage of development in enterprises involves AI agents that independently handle clearly defined tasks. The study shows that 91 percent of the frontrunners surveyed use knowledge assistants, while 83 percent rely on automated document creation and text processing.
AI agents introduce a new level of responsibility. An inaccurate response from a chatbot is frustrating. With an AI agent in manufacturing, logistics, energy, or transportation, an error can halt a facility, delay a delivery, or create a safety risk. It becomes sensitive when an AI agent triggers an order without control, changes master data, or accesses sensitive information. The benchmark for AI agents must not be what is technically possible, but the tangible benefit they provide with manageable risk.
Telekom is introducing two offerings: the T-AI Agent Platform and the T-AI Agentic Hub. Why are two platforms needed?
Tino Mager: Because companies start from different points. Many midmarket customers want to solve a specific problem: too many manual emails, recurring service requests, or data that employees have to laboriously transfer between systems. They want to begin with a manageable process, see value quickly, and expand the AI agent step by step. They can start with a preconfigured agent, adapt existing workflows themselves, or work with us to develop a custom agent. With predictable package pricing.
Sven Giesselbach: In organizations where agents handle an increasing share of processes, AI agents often emerge independently across different business functions. One may be based on n8n, while others use LangGraph, CrewAI, or AutoGen. They do not "understand" one another, use different language models, and have different levels of autonomy. T-AI Agentic Hub brings together AI agents from different vendors and technical frameworks on a shared operational platform. It serves as the central control point for meeting requirements and managing costs. It is the Operating System for agents.
Tino Mager: Put simply, T-AI Agent Platform automates clearly defined business processes. T-AI Agentic Hub coordinates entire agent landscapes—from fixed workflows to agents that independently plan and distribute tasks. T-AI Platform executes processes, while T-AI Agentic Hub keeps their interactions under control. The difference, then, lies less in the technology than in the complexity, risk, and organizational starting point.
What specific problem does the T-AI-Agent Platform solve for an enterprise?
Tino Mager: In practice, we see the challenge of turning an idea into a working business process. Which task takes up time today? Which process can be automated reliably?
The T-AI Agent Platform combines n8n-based workflow automation with AI models and capabilities such as speech and IT security. It offers standard agents, an agent builder, and custom agents. For example, you can use n8n's graphical interface to adapt prebuilt workflow templates to your existing email, CRM, or ERP system. Operations, security, user and access management, lifecycle management, monitoring, billing, and support are part of the platform. Telekom MMS supports you in selecting the process and integrating it into your organization's IT landscape.
A good first AI agent has a clearly defined task. You can verify its results. Your organization can then see whether effort, processing time, or even errors decrease.
At what point is a single workflow solution no longer enough?
Sven Giesselbach: An AI agent in customer service can usually still be clearly assigned to a single process and team. It is different when AI agents work in parallel across IT operations, financial analysis, risk management, or the supply chain, accessing different data and enterprise systems.
As their scope of action grows, identities, permissions, traceability, cybersecurity, and human approvals need to be considered from the outset. Control is not something added afterward. Only then can you reliably scale their deployment.
In addition, when AI agents plan and assign tasks independently and collaborate with other agents or systems, companies need to rethink their processes: Which reusable building blocks can be used across processes? How much autonomy makes sense, and where is a fixed workflow a better fit? T-AI Agentic Hub provides an appropriate framework for both approaches. It connects agents across system boundaries and shows which model each AI agent uses, which systems it is authorized to access, and which actions it has performed. Data remains in the source systems; access takes place through controlled interfaces, activities remain traceable and testable, and can be managed through role- and permission-based concepts. Cost limits can also be defined for specific users or models.
Different models? Would that mean companies have to sign a separate contract for each AI model?
Sven Giesselbach: No. T-AI Agentic Hub uses AI Foundation Services, among other things. This is T-Systems' marketplace for AI. It gives companies access to many AI models through one provider, one technical interface, and consolidated billing – instead of separate contracts and integrations with each individual model provider. Usage is billed based on consumption.
Tino Mager: T-AI Agent Platform also accesses AI Foundation Services. The practical advantage is freedom of choice. A company can make the right choice for speed, cost, quality, or protection requirements.
Sven Giesselbach: You raise the topic of sovereignty here, Tino. This freedom of choice is an important aspect of sovereignty. Let’s remember: in June, the U.S. government temporarily required AI company Anthropic to block foreign nationals from accessing its latest AI models. A supplemental question from the ifo Institute shows that nearly nine out of ten German companies use digital products from U.S. providers. Thirty-one percent consider themselves highly dependent on them. Sovereignty means that a company understands its dependencies, has alternatives, and can switch or intervene if necessary.
Not every model needs to come from Germany. Models differ in performance, price, and level of sovereignty. You can define the infrastructure on which models run—for example, in your own IT environment, on sovereign infrastructure such as T Cloud, or with a hyperscaler. Based on the data's protection requirements, T-AI Agentic Hub determines which environment may process data in accordance with your company's policies. This lets you combine different models within an agent system based on protection requirements, cost, and performance, including your company's own models. T-AI Agent Hub can adapt to the operational and sovereignty requirements of a corporation or public authority.
Topic: Sovereignty: IT and compliance want control, while business units want fast results. How can you address this conflict?
Tino Mager: Focusing only on speed risks creating a pilot that works but cannot be scaled later. Treating every initial use case as an enterprise-wide high-risk application prevents teams from gaining practical experience.
I like to explain this using a manageable use case, ideally one with recurring patterns. You can define success criteria, interfaces, and rules for it. This allows a company to gain experience quickly without postponing the necessary ground rules and governance until later. Many AI projects stall because these prerequisites are considered too late.
Sven Giesselbach: Governance should not be seen as a brake. When implemented effectively, it creates the foundation for scaling. As long as a pilot is managed by a small team, many questions can still be resolved manually. Once dozens of AI agents are operating across different areas, that no longer works. Each AI agent then needs a unique identity, clearly defined permissions, logged actions, and boundaries it must not cross.
The goal is not to have as much control as possible at every point. The goal is the right level of control for each action.
Who is responsible when an agent makes a mistake?
Sven Giesselbach: Responsibility cannot be delegated to the AI agent. The organization deploying it must define which tasks may be automated and which controls are required.
That is why multiple layers of safeguards are needed. T-AI Agentic Hub is designed to test AI agents with people before they are approved and to incorporate security and compliance requirements. T-AI Agentic Hub classifies agent actions by risk and determines when a person must intervene. Particularly critical actions can be configured to always require human approval. Cost and usage limits can also be built in.
Tino Mager: What matters is how the specific process is designed. For example, an AI agent can prepare a decision while a person confirms it. For a low-risk routine, the AI agent can complete the process on its own. We should not ask whether AI agents should generally be allowed to work autonomously, but how autonomous they should be in each process.
What specific decisions should companies make now?
Tino Mager: Companies should start with the use case, not the technology: Where are unnecessary efforts occurring today, and can the benefits of automation be measured? There is no prescribed path. Some companies start with a single business process, while others need centralized governance from the outset because of their complexity or regulatory requirements. The two offerings are not mutually exclusive. Workflows from the Platform can be integrated into a larger Hub environment.
Sven Giesselbach: They should also clarify early on what level of control each use case requires: What decisions may the AI agent make? What data and systems may it access? What infrastructure should the agent run on? And when must a person intervene?
Tino Mager: It is not the most powerful AI that determines a company's success. What matters is whether a company can translate it into real value creation, scale it securely, and govern it responsibly.
