Mr. Corici, “Agentic AI” is considered the next step in the development of AI systems. What does this concept mean specifically for communication networks, and what is meant by an “Agentic Core”?

The idea of using agents in communication networks is not new. Research into distributed, intelligent agents began more than two decades ago. At the time, however, practical deployment largely failed due to limited computing resources, insufficient data availability, and a lack of suitable integration mechanisms.

Today, the technological landscape has changed fundamentally. Modern cloud architectures, powerful real-time data processing, and advanced AI methods enable software agents to operate reliably, perform extensive optimizations within the core network, and participate in decision-making at both the control and data plane. An Agentic Core can therefore be understood as a core network in which selected operational and optimization decisions are handled by specialized software agents. These agents operate within clearly defined rules, particularly with regard to safety and guaranteed response times.

There is no single, universal agent. Depending on the requirements, very fast and simple agents are used to react under strict delay and reliability requirements, while more complex agents carry out long-term optimizations based on extensive analysis and learning models. This combination enables both immediate reaction and strategic adaptation, while determinism, verifiability, and operational trust remain fundamental to the behaviour of the overall system. 

Open6GCore is considered a key reference implementation for future 6G core networks. What role does Open6GCore play for the Agentic Core?

Open6GCore is designed as a highly flexible implementation of core network functions that can be adapted to a wide range of service requirements, deployment scenarios, and operating conditions. This adaptability is a fundamental prerequisite for integrating software agents that continuously observe, decide, and act in changing environments.

The Agentic Core therefore builds on Open6GCore as its execution and experimentation environment. While Open6GCore provides structural flexibility and compliance with standards, the Agentic Core components extend the system with regulated, policy-based autonomy. This makes it possible to partially automate decisions at both the control and data plane. Together, the two approaches mark the transition from a configurable, 5G-derived core network to a continuously adaptable, 6G-native system.

From a scientific perspective, what are the main advantages and challenges of using AI in communication networks?

From a scientific standpoint, the use of AI in communication networks represents a fundamental paradigm shift. Instead of strictly predefined software behavior, adaptive and data-driven operating models are coming to the fore.

This shift offers clear advantages. AI can analyze complex network states, allocate resources more efficiently, and respond to dynamic changes more quickly than traditional rule-based mechanisms. It also enables a higher degree of automation in planning, provisioning, and operation – an essential capability given the expected scalability and heterogeneity of future 6G environments.

At the same time, significant challenges remain. Communication networks require predictable, interoperable, and verifiable behavior, whereas many AI methods are probabilistic and difficult to explain formally. As a result, current research increasingly focuses on regulated AI. Rather than unrestricted autonomy, agentic systems are designed to operate within clearly defined and measurable limits, remain observable, and be validated before deployment. Finding the right balance between adaptability and determinism is one of the central scientific challenges on the path toward 6G.

Communication networks are part of critical infrastructure. How can the reliability of an agentic AI-based network be ensured?

As critical infrastructure, communication networks must meet particularly high reliability requirements. Agent-based decision-making increases system complexity, especially when large AI models are involved. While such models can outperform traditional mechanisms in many situations, they may also produce suboptimal or, in rare cases, harmful decisions.

To mitigate these risks, the Agentic Core relies on multi-layered safety mechanisms. Fast and simple agents operate within very tightly constrained action spaces. More complex optimization decisions are validated in advance through extensive simulations using the OpenLANES digital twin. This emulated environment makes it possible to realistically reproduce large-scale network scenarios and test them repeatedly before new functions are deployed in live systems.

At the same time, large models can enhance reliability in rare and unforeseen failure scenarios. When predefined rules are insufficient and recovery mechanisms stall, such models can often generate responses that are not necessarily optimal, but good enough to restore partial functionality. This introduces an additional layer of resilience, embedded within the overall reliability framework of the network – a decisive factor for the use of agentic AI in critical communications infrastructures.

Meet Dr. Marius Corici and more 5G/6G experts of Fraunhofer FOKUS at Mobile World Congress in Barcelona: Hall 7, Booth 7C61 (Berlin Partner) or Hall 6, Booth 6F10–6F15 (6G Platform Germany).