Google, Microsoft back 11-company protocol for AI agent discovery

Eleven tech companies released a draft standard June 17 that lets AI agents discover tools and services at runtime, bypassing the manual wiring that has kept enterprise agents from scaling.

Google, Microsoft, and nine other companies published Agentic Resource Discovery, a specification that lets AI agents find and verify tools across the web at runtime — turning today's manually wired agent ecosystem into a searchable network.

"ARD helps AI clients discover capabilities but does not replace authentication, authorization, governance, or organizational trust decisions," Microsoft said in a blog post announcing the standard.

The v0.9 draft, released June 17 under an Apache 2.0 license, relies on two components: catalogs — an ai-catalog.json file hosted on a publisher's domain listing available tools, MCP servers, and agents — and registries that crawl those catalogs and answer discovery requests in plain language. Domain ownership serves as the verification mechanism, with optional cryptographic trust metadata for production use. Contributors shipped working implementations the same day, including GitHub's agent finder for Copilot and Hugging Face's Discover Tool.

The standard threatens to reshape the enterprise AI market by positioning traditional software platforms as the gateway to AI tools, directly challenging Anthropic and OpenAI's ambition to make Claude and ChatGPT the primary enterprise AI interface. Neither company joined the initial supporter list.

The Strategic Divide

The split between the two camps reflects a fundamental strategic conflict. Anthropic and OpenAI aim to build Claude and ChatGPT into standalone enterprise entry points that workers use to access all corporate applications. Google, Microsoft, Salesforce, and ServiceNow — the companies behind ARD — want their existing platforms (Gemini, Copilot, CRM software) to serve that role instead, with ARD acting as the discovery layer that lets those platforms find and connect to any AI tool automatically.

ARD builds on concepts introduced by Anthropic's Model Context Protocol, released last year, which similarly enables AI agents to access external application data. But where MCP focuses on the connection itself — the equivalent of an API call — ARD addresses the discovery problem that precedes it: how does an agent know which tools exist, what they do, and whether they can be trusted? The protocol hands off to the tool's own connection method once a capability is selected, meaning ARD and MCP could operate as complementary layers rather than direct competitors.

Who Wins, Who Loses

The immediate beneficiaries are companies that publish callable capabilities — APIs, MCP servers, and agents that software connects to. For those publishers, ARD provides a standardized method for being found and trusted by AI agents without pre-wiring each connection. Google's Agent Registry, part of its Gemini Enterprise Agent Platform, will host and search agentic resources with enterprise governance, though native ARD support is not yet live and is planned for the coming months.

For typical content sites, the standard offers no clear action today. Google's John Mueller has argued that large language models cannot use files like llms.txt to distinguish one site from another, advising publishers to focus on current needs rather than future agent-oriented strategies.

The standard's reach depends on registries that can crawl and index catalogs at scale — an ecosystem still in its early stages. The v0.9 draft is open for changes through the project's GitHub repository, and the contributors are inviting industry feedback before a final version.

The $64 Billion Question

The enterprise AI platform market represents a battleground worth tens of billions in annual software spending. Traditional tech giants trade at established multiples — Microsoft at 35x forward earnings, Google at 22x — with AI revenue already baked into growth projections. OpenAI was valued at $300 billion in its latest funding round, and Anthropic at $61.5 billion, with both companies' valuations premised on capturing enterprise AI budgets currently flowing to incumbents.

If ARD gains adoption, it strengthens the moat of traditional platforms by making them indispensable intermediaries for AI tool discovery. If it stalls, Anthropic and OpenAI's direct-to-enterprise strategy gains credibility. The next six to 12 months, as Google's Agent Registry goes live and registries scale, will determine which vision prevails.

This article is for informational purposes only and does not constitute investment advice.