Internet of Agents (IoAT): Sensing, Communication, Decision-Making and Execution

Advances in sensing, communication, and reasoning are driving deeper integration of AI into IoT systems. However, progress remains siloed across two communities: agent research centers on task decomposition, semantic planning, and coordination, whereas IoT research prioritizes sensing, communication, and edge computing. Bridging these perspectives and integrating semantic reasoning with system foundations remains a key challenge. Therefore, we present the Internet of Agents (IoAT), a framework that reorganizes existing AI agent research from a systems perspective, following the classical IoT paradigm of sensing, communication, decision-making, and execution. IoAT reconceptualizes system nodes as agents rather than traditional IoT endpoints. For sensing, IoAT broadens awareness to include reasoning depth, knowledge gaps, and external tool capabilities. In communication, IoAT adopts a three-layer architecture (environment, interaction, and semantic), designed to align with the five-layer Internet protocol stack. In decision-making and execution, IoAT frames the shift from rulebased control to semantic planning and from reactive execution to proactive execution. We further identify open challenges and future research directions across the four modules of IoAT.