China Mobile With the widespread adoption of AI Agents, traditional network architectures can no longer meet the demand for efficient collaboration between agents and networks. This document proposes a new paradigm of "AI Agent Policy-Aware Network", enabling three key transformations: from Flow-aware to Agent-aware, from QoS-based to Policy-intent-based, and from Network-driven to Agent-network collaborative. By defining core components such as the Agent Policy- aware Controller and Agent Policy-Aware Device, this paradigm establishes a dynamic mapping mechanism between Agent intents and network policies, supporting key scenarios including autonomous performance measurement, path optimization, SLA assurance, and secure transmission. This document outlines the background, scenarios, use cases and requirements of Agent Policy-aware Network.