AGENTIC-AI xAPPS: O-RAN MULTI-AGENT CONTROL FOR ENTERPRISE POLICY, SCHEDULING & SPECTRUM

The following work presents the Agentic-AI xApps in O-RAN as a multi-agent coordination framework, wherein the different agents take up the task of managing the respective portions of policy enforcement, scheduling optimization, and spectrum allocation in enterprise networks, thus doing it in an autonomous manner.In the Near-RT RIC, the agents are using reinforcement learning and LLMs as their tools for concurrent decision-making, which means they are able to adapt to changes in traffic, interference, and intents in real-time.The convergence behavior is analyzed in simulated enterprise scenarios, and very rapid stabilization is shown via shared state communication and predictive forecasting. Lyapunov-based bounds and drift detection together with tension control during high-load surges are what make stability possible. Conflict resolution involves a mediator agent that uses game-theoretic arbitration and veto mechanisms, thus blocking unsafe actions that might lead to neighbor destabilization. This is Proven in traffic steering use cases where KPIs such as throughput and SINR are applied.The results show that there is a 25% efficiency improvement over centralized RIC and at the same time stable global performance. This mechanism is a step forward in the direction of developing a trustworthy, scalable O-RAN autonomy for 5G/6G enterprises.