DESIGNING THE AI WEB: ARCHITECTURAL PRINCIPLES FOR INTEROPERABLE ARTIFICIAL INTELLIGENCE SYSTEMS

Abstract

With the field of artificial intelligence undergoing constant evolution, AI systems are rapidly shifting from single models to complex networks that involve countless specialised agents. This transition is supported by several studies that have established multi-agent systems to offer advanced capabilities when compared to single LLM-powered agents. While it is aimed for these to work together by interacting with one another, the absence of a standard structure and mode of communication proves to be a challenge.This paper introduces and elaborates on the idea of an AI Web, a unified network of interoperable artificial intelligence agents capable of collaborating seamlessly across numerous diverse frameworks and applications. We draw inspiration from the internet, where machines use certain protocols for communication. Building upon previous studies, we propose a nine-layer structure embedded with Model Context Protocol (MCP) and Agent-to-Agent (A2A) protocols to allow effective communication between agents. The model was conceptually evaluated through application scenarios in enterprise and business domains.