An Agentic AI-based Radio Access Network for Efficient Video Delivery in Internet of Vehicles

Vehicular networks are increasingly relied upon to support real-time, bandwidth-intensive applications such as video streaming and immersive media services. However, the high mobility of vehicles, intermittent connectivity with roadside base stations, and fluctuating user demands introduce substantial challenges for optimizing Quality of Experience (QoE). This paper presents a novel AI-driven Radio Access Network (AI-RAN) architecture that embeds intelligent agents based on Large Language Models (LLMs) into the near-real-time RAN Intelligent Controller (near-RT RIC). These agents are designed to anticipate vehicle trajectories, predict service demands, and dynamically adjust bitrate and routing strategies in a context-aware and coordinated fashion. By leveraging semantic reasoning through prompt engineering, each agent interprets mobility patterns, network conditions, and QoE indicators to jointly optimize video delivery for mobile users. We develop a realistic co-simulation testbed integrating SUMO, ns-3, and DASH streaming modules, and benchmark our framework against traditional heuristics and deep reinforcement learning approaches. Experimental results demonstrate that the proposed system significantly enhances QoE, reduces re-buffering events and bitrate fluctuations, and enables proactive, efficient content delivery at the network edge. Our findings highlight the potential of LLM-based agents to drive edge intelligence and autonomous optimization in future AI-RAN deployments.