Integrating Causal AI, Agentic AI, and LLMs for Intelligent Telecom Network Management

Abstract

The rapid evolution of telecommunication networks toward 5G, beyond 5G (B5G), and emerging 6G infrastructures has introduced unprecedented complexity in network management due to massive device connectivity, dynamic traffic patterns, virtualization, and distributed cloud-native architectures. Traditional network management approaches rely heavily on statistical learning and rule-based automation, which often fail to distinguish correlation from causation, resulting in suboptimal decision-making during unexpected network events. Recent advances in Artificial Intelligence have demonstrated that combining Causal AI, Large Language Models (LLMs), and Agentic AI can significantly improve autonomous network operations by enabling reasoning, adaptive planning, root-cause identification, and intelligent decision support. This paper proposes an integrated telecom management framework that combines causal inference for understanding network behavior, LLMs for knowledge-driven analysis and operator interaction, and Agentic AI for autonomous planning and execution of network optimization tasks. The proposed framework is evaluated using simulated telecom network environments containing traffic congestion, base station failures, routing anomalies, and fluctuating user demands. Experimental findings demonstrate improvements in fault localization accuracy, network throughput, latency reduction, resource utilization, and decision-making efficiency compared to conventional machine learning-based approaches.