The web is being rebuilt for agents: Google and 11 others publish ARD, an "index for AI agents"

If search was an index for people, ARD is an index for AI agents. The Agentic Resource Discovery (ARD) specification, published by Google on June 17, 2026, is an open standard that lets agents find and verify the capabilities they need across the web. Microsoft, GitHub, Nvidia, Hugging Face, and 11 companies in total built it, and released it under the Apache 2.0 license. It signals that the one doing the discovering is shifting from people to agents. ASAP summarizes how this standard changes content discovery, from the primary announcement.

The two doors ai-catalog.json actually opens

ARD works through two primitives. A site hosts an ai-catalog.json manifest at a well-known path on its domain, and a registry API crawls and indexes these catalogs and returns ranked matches to natural-language queries. If robots.txt and sitemap were for search engines, ai-catalog.json is a sitemap for agents. Google designed it so an agent can ask at runtime where a capability lives, which one to use, and whether it is safe to connect to.

Why a "registry," not just a sitemap

The point where this design breaks from existing web discovery is the second primitive, the registry API. Hosting a manifest looks a lot like a sitemap, but a search engine's sitemap ends with a crawler reading pages and handing a human a list of links. ARD's registry instead takes a natural-language query and returns ranked capabilities. That means what you pick is a candidate to invoke, not a link to click. As the distance between discovery and execution shrinks, the axis of competition moves from which site gets surfaced to which capability gets selected.

What practitioners should read into it

For anyone running content or services, the announcement poses a clear question. Content has been optimized for people to click, but now it must be optimized for agents to query and select. Being discovered and cited by AI is becoming the new gateway to traffic. Cisco, Databricks, GitHub, GoDaddy, Google, Hugging Face, Microsoft, Nvidia, Salesforce, ServiceNow, and Snowflake all signed on as launch contributors, and GitHub shipped a reference implementation the same day. It is rare for a group this size to push one standard at once, which is reason to move up manifest experiments rather than wait and watch.

How to read "open standard"

ARD is an open standard published under the Apache 2.0 license, built on the data model from the Linux Foundation's AI Catalog Working Group. Because it is a shared spec that no single company owns, it does not lock anyone into one closed ecosystem, which lowers the barrier to adoption. But being an open standard does not by itself guarantee real usage. What is confirmed so far is the spec, the contributor list, and a same-day reference implementation; whether agents actually treat this registry as their query path cannot be judged from this announcement alone. The direction of the signal is clear, but its magnitude is still open.

Source: ASAP summary of Google's Agentic Resource Discovery (ARD) specification release (June 17, 2026; Apache 2.0 license, ai-catalog.json manifest plus a registry API, 11 contributors including Cisco, GitHub, Microsoft, Nvidia, and Hugging Face, with a GitHub reference implementation the same day).