Agentic AI for 6G Satellite Networks: Achieving Scalable and Low-Latency Resource Scheduling With the AERS Framework

Low Earth orbit (LEO) satellite networks are essential for enabling global 6G connectivity, but their resource scheduling faces fundamental challenges from limited visibility windows, constrained onboard power, and a scarcity of beams relative to diverse service demands. Existing schemes struggle to simultaneously meet real-time processing requirements with computational constraints while ensuring user quality-of-service (QoS). To address these issues, this paper introduces AERS (Agentic Energy-Aware Resource Scheduling), a decentralized framework grounded in Agentic AI principles, including autonomy, adaptability, proactivity, and coordination. In AERS, each beam operates as an independent agent that collaboratively optimizes beam hopping, user scheduling, and power allocation based on local observations. Coordination is achieved implicitly through shared rewards and environmental coupling, such as co-channel interference and power constraints, without requiring continuous interaction between agents. Case studies show that AERS achieves superior tradeoffs between delay and packet loss, while achieving a 2.98× inference speedup compared with centralized methods, making it highly suitable for satellite onboard deployment. This work establishes a foundational paradigm for employing Agentic AI in satellite resource management, offering a scalable and efficient solution for future 6G satellite networks.