AI Agents Take Control of Mobile Networks

As 5G networks become increasingly complex with dense deployments and diverse service requirements, traditional manual management approaches are proving inadequate. The transition to 6G will amplify these challenges, creating an urgent need for intelligent, adaptive systems that can handle network operations autonomously. This research demonstrates how agentic AI—systems that use large AI models as central controllers—can manage mobile networks with human-level cognitive abilities including planning, memory, and continuous learning.

Researchers have developed a framework where AI agents can interpret network operator goals, monitor performance in real-time, and autonomously execute optimized policies. These systems use large language models (LLMs) enhanced with capabilities for reflection, tool use, and multi-agent collaboration to handle the dynamic nature of modern telecommunications networks. The approach represents a significant evolution from classical rule-based systems to autonomous agents that can decompose complex goals and adapt to changing conditions.

The methodology integrates several design patterns that define how intelligent systems execute tasks. Reflection enables agents to evaluate and revise their outputs through self-assessment and iterative refinement. Tool use allows AI models to interface with external systems beyond text generation, retrieving real-time information and interacting with databases. Planning involves breaking down tasks into manageable steps, while multi-agent collaboration coordinates specialized agents performing different subtasks. These patterns work together in a cohesive architecture that maintains awareness of network state and adapts to new information.

In a practical case study focusing on 5G Radio Access Network (RAN) optimization, the system demonstrated its ability to monitor Key Performance Indicators (KPIs) across multiple hierarchy levels—from individual cells to regional clusters. The AI agents process performance measurement data streams collected at 15-minute intervals, identifying patterns, anomalies, and performance degradations. Through reflection loops, the system continually compares predictions against new data, refining its understanding of network behavior and relationships between different hierarchical layers. The implementation shows how agents can hypothesize about performance issues, investigate potential causes, and propose corrective actions when permitted to execute them.

The real-world significance lies in addressing the growing complexity of mobile networks that conventional approaches can no longer handle efficiently. Industry applications already show promise: Deutsche Telekom and Google Cloud developed a system using Gemini 2.0 to detect anomalies and initiate corrective actions, while Telenor and Ericsson created an agentic AI system that adjusts configurations to balance throughput and energy consumption. These developments indicate that autonomous network management could improve performance, resilience, and reliability in next-generation telecommunications.

However, significant limitations remain. The computational and energy costs of running large AI models present practical challenges, particularly for distributed edge and cloud nodes. Security and privacy risks are amplified when sensitive user data and network control signals are involved, requiring safeguards against adversarial behaviors and information leakage. Current frameworks also struggle with efficient communication protocols between agents, and maintaining consistent awareness across dynamic network conditions requires sophisticated infrastructure for data retrieval and planning. Addressing these challenges will require advances in system architecture, communication protocols, and governance frameworks before widespread deployment becomes feasible.