AI visibility expands as Google, Microsoft back agent discovery spec

AI visibility is moving one layer deeper, from pages people click to machine-readable catalogs that agents can crawl, verify, and trust. Google, Microsoft, GitHub, GoDaddy, and Hugging Face are backing the Agentic Resource Discovery project, a draft spec dated May 28, 2026, that treats agent discovery as a new surface area for search.

Google announced the Agentic Resource Discovery specification on June 17, 2026, framing it as an open standard for finding and verifying tools, skills, and agents across the web. The listed authors include Junjie Bu of Google, Scott Courtney of GoDaddy, R.V. Guha of Microsoft, and Shaun Smith of Hugging Face. The pitch is simple but consequential: agents do not search the web like humans do, so they need a way to discover capabilities at runtime instead of relying only on prewired connections.

The core of the system is publisher-hosted catalogs. Google says domain ownership becomes the cryptographic foundation for identity and trust, which means the publisher’s own domain is doing more than hosting a landing page. Registries crawl those catalogs, index them, and return matching capabilities along with metadata needed to verify the publisher before any connection is made. That is a big shift for anyone building tools, services, or APIs, because the new visibility layer depends on whether an agent can find, verify, and trust the resource without a human in the loop.

Google’s own Agent Registry shows where this is heading inside enterprise software. In Gemini Enterprise Agent Platform, Agent Registry is the central governance and inventory hub, with a unified catalog to securely store, discover, and manage MCP servers, tools, and AI agents. Google says built-in Google agents and agents deployed through the SDK can be discovered without additional configuration, while GKE-deployed agents can be registered through annotations. That turns infrastructure choices into discoverability choices.

Hugging Face has described ARD as a draft, open specification developed by contributors from Microsoft, Google, GoDaddy, Hugging Face, and others, not as a product or marketplace but as a shared standard anyone can implement. That distinction matters. If the spec catches on, the practical work will shift toward structured, machine-readable endpoints for products, services, and tools, much like schema and sitemaps once became mandatory for human search. Microsoft’s broader agent push adds weight to that reading: at Build 2025, Microsoft Discovery was presented as an extensible platform to accelerate scientific discovery with agentic AI. The message from both vendors is clear: visibility is no longer just about being found on a webpage. It is about being legible to the software that now does the searching.