Communication of Internet of Agents (IoA): A Survey on Transmission, Interaction, and Evaluation

With the deep integration of large language model (LLM)-driven agents and the Internet of Things (IoT), a new paradigm known as the Internet of Agents (IoA) has emerged. However, existing studies mainly focus on high-level capabilities such as semantic reasoning and multi-agent collaboration, while often overlooking the underlying communication infrastructure. This results in an architectural disconnection between upper-layer intelligence and lower-layer connectivity, thereby constraining the large-scale deployment of IoA. To bridge this gap, this paper reviews the state of the art in IoA communications from a communication system engineering perspective, covering three key dimensions: transmission, interaction, and evaluation. At the lower-layer transmission level, targeting the highly dynamic and bandwidth-intensive characteristics introduced by LLM-based interactions, this paper systematically surveys network adaptation and optimization mechanisms ranging from reliability assurance to scalability support. At the upper-layer interaction level, this paper extends the scope of communication from mere data exchange to semantic interoperability, provides an in-depth analysis of the standardization evolution of interaction semantics, and reveals capability-driven matching and cognitive collaboration mechanisms in dynamic environments. To validate system effectiveness, this paper further reviews multi-level evaluation frameworks spanning software simulation and physical testbeds. Finally, this paper discusses key challenges in IoA communications, including dynamic networking and semantic consistency, and outlines promising future research directions.