The AI agent ecosystem is fragmenting, and a group of researchers wants to stop it before the damage becomes permanent.
A position paper published on arXiv outlines what the authors call "Web of Agents" — a minimal architectural framework for making autonomous AI agents interoperable across different platforms and vendors. The proposal centers on four building blocks: agent-to-agent messaging, interaction interoperability, state management, and agent discovery. Crucially, the researchers argue for reusing existing standards and infrastructure rather than inventing new ones, which is a rare act of restraint in a field that loves to rebuild from scratch.
The stakes are real. Agentic AI — systems that can perceive their environment, make plans, and take actions autonomously — is already being deployed across enterprise software, customer service, and developer tooling. If every major platform builds its own closed runtime, the result looks less like the open web and more like the early days of messaging apps: powerful inside their walls, useless outside them. A fragmented agent landscape would hand structural advantages to whoever locks up the most enterprise workflows first.
The Web of Agents proposal is a position paper, not a shipping standard — so the gap between argument and adoption remains wide. The history of interoperability efforts in tech is mostly a graveyard of sensible ideas that arrived too late or too early. Whether this one lands before the silos harden is the actual question worth watching.