SANet: A Semantic-Aware Agentic AI Networking Framework for Cross-Layer Optimization in 6G

Agentic AI networking (AgentNet) is a novel AI-native networking paradigm in which a large number of specialized AI agents collaborate to perform autonomous decisions, dynamic environmental adaptation, and complex missions. AgentNet has the potential to facilitate real-time network management and optimization functions, including self-configuration, self-optimization, and self-adaptation across diverse and complex environments, laying the foundation for fully autonomous networking systems. Despite its promise, AgentNet is still in the early stages of development and still lacks an effective networking framework to support automatic goal discovery, multi-agent self-orchestration, and task assignment. This paper proposes SANet, a novel semantic-aware AgentNet architecture for wireless networks. SANet can infer the semantic goal of the user and automatically assign agents associated with different layers of the network stack to fulfill the inferred goal. Motivated by the fact that AgentNet is a decentralized framework in which collaborating agents may generally have different and even conflicting objectives, we formulate the decentralized optimization of SANet as a multi-agent multi-objective problem, and focus on finding the Pareto-optimal solution for agents with distinct and potentially conflicting objectives. We propose three novel metrics for evaluating SANet: (the agents' objective) optimization error, (dynamic environment) generalization error, and (multi-objective) conflicting error. Furthermore, we develop a model partition and sharing (MoPS) framework in which large models, e.g., deep learning models, of different agents can be partitioned into shared and agent-specific parts that are jointly constructed and deployed according to agents' local computational resources. Two decentralized optimization algorithms, static-weighting and dynamic-weighting algorithms, are introduced to optimize the above three metrics. A bandwidth-adaptive compression framework is also proposed to enable different agents to perform in situ compression of their intermediate embeddings, dynamically adjusting to localized resource constraints and task requirements. We derive theoretical bounds for all these performance metrics and prove that there exists a three-way tradeoff among optimization, generalization, and conflicting errors. Finally, to validate our theoretical results, we develop an open-source Radio Access Network (RAN) and core network-based hardware prototype that implements three Transformer-based time-series prediction agents to interact with three different layers of the network. Experimental results show that the proposed MoPS framework achieves performance gains of up to$14.61\%$while requiring only$44.37\%$of the Floating-Point Operations (FLOPs) for inference at each agent compared to state-of-the-art algorithms. Also, compared to the static-weighting algorithm, the dynamic-weighting algorithm achieves up to$83.81\%$reduction in training errors caused by conflicting objectives.