Toward the Internet of Agentic AI: Protocols, Architecture, and Challenges

With the rapid development of artificial intelligence (AI), the Internet of things (IoT), and 6G, autonomous agents are increasingly deployed in a wide range of applications. However, existing network architectures, centered on human control and fixed protocols, lack the flexibility to support dynamic, intent-driven collaboration among heterogeneous agents. To address this, this article introduces the Internet of agentic AI (IoA), a new paradigm that positions agents as primary entities for communication and coordination. We present the architecture, core protocols, and interaction models that enable IoA functionality. Furthermore, we propose an intelligence-based reinforcement learning (IRL) framework that replaces scalar rewards with multi-dimensional random variables, supporting adaptive, multi-objective agent coordination and enabling more expressive policy learning. Simulation results validate the effectiveness of the proposed approach. The article also presents representative use cases and outlines future research directions.