Google, Microsoft lead open specification for AI agent discovery

Image credit: Search Engine Journal

Eleven major technology companies, including Google and Microsoft, released a draft open specification Monday to standardize how artificial intelligence agents discover and verify tools and capabilities across the internet.

The Agentic Resource Discovery (ARD) specification aims to enable AI agents to autonomously find and utilize necessary resources at runtime, addressing a key challenge in developing more interoperable and capable AI systems.

Published on June 17, ARD was developed collaboratively by Google, Microsoft, GitHub, Hugging Face, Cisco, Databricks, GoDaddy, NVIDIA, Salesforce, ServiceNow, and Snowflake.

The specification outlines the use of ‘catalogs’—ai-catalog.json files hosted on publishers’ domains—and ‘registries’—crawlers that index these catalogs—for resource discovery.

Verification relies on domain ownership and trust metadata, according to the published draft.

Several contributing organizations have already implemented solutions based on the ARD specification. GitHub introduced an agent finder for its Copilot service, while Hugging Face launched its Discover Tool on the same day.

Google announced plans to integrate native ARD support into its Gemini Enterprise Agent Platform’s Agent Registry in the coming months. This integration will allow organizations to connect their internal AI registries to the broader ARD network.

The ARD specification is currently a v0.9 draft and is licensed under Apache 2.0. It builds upon the existing AI Catalog data model developed under the Linux Foundation.

John Mueller, a prominent figure in web standards, has been involved in discussions surrounding the specification, which is designed to foster a more coordinated environment for AI agent development.

The companies involved stated that standardizing resource discovery is important for the future scalability and utility of AI agents, moving beyond systems that require pre-wired connections to tools and skills.

Source: Search Engine Journal