LLM-Driven Agentic AI Approach to Enhanced O-RAN Resilience in Next-Generation Networks
Xingqi Wu, Yuhui Wang, Junaid Farooq, Juntao Chen · 2025
The open radio access network (O-RAN) architecture has revolutionized RAN design by enabling flexibility, interoperability, and the deployment of automated and AI-driven management solutions. However, achieving efficient resource allocation remains challenging due to the dynamic nature of network slices, each with distinct and evolving quality of service (QoS) requirements. These challenges are further amplified in real-time environments, where traditional machine learning (ML) approaches, relying on offline training, struggle to adapt effectively to changing system conditions. To address these issues, this paper introduces a large language model (LLM)-Driven Agentic AI framework for enhancing resource management and resilience in O-RAN systems. The proposed approach leverages intelligently crafted prompts to guide LLM agents in dynamically optimizing resource allocation across network slices, ensuring adaptability and improved system performance. Experimental evaluations conducted on the Open AI Cellular (OAIC) testbed demonstrate significant gains in average data rates, slice reliability, and responsiveness, showcasing the potential of LLMs to drive resilient, real-time decision-making in next-generation wireless networks.