A new AI agent standard called Agentic Resource Discovery, or ARD, has picked up backing from Google, Microsoft and Salesforce, in a move read squarely as a counter to Anthropic and OpenAI.
ARD is an open specification that lets AI agents automatically find and connect to the tools they need across a company’s sprawl of software, as reported by CIO Dive.
Instead of a developer wiring every connection by hand, an agent can discover what is available and plug itself in.
What the ARD AI agent standard actually does
Think of ARD as a directory service for agents. When an AI agent needs to book travel, pull a report or update a record, it first has to know which tools exist and how to talk to them.
ARD standardises that discovery step so agents can locate resources across a corporate stack without bespoke integration work for each one.
The coalition around it is heavy. Beyond the three headline names, ARD has drawn support from Cisco, Databricks, GitHub, GoDaddy, Hugging Face, NVIDIA, ServiceNow and Snowflake.
That is a lot of infrastructure firms lining up behind one spec, which is usually how a standard goes from proposal to default.
Crucially, ARD is designed to sit alongside existing protocols rather than replace them. It is meant to complement Anthropic’s Model Context Protocol, which launched in late 2024, and Google’s own Agent2Agent protocol, layering discovery on top of the plumbing those standards already provide.
Why the AI agent standard fight matters
Here is the nerdy subtext. Anthropic’s Model Context Protocol landed so well in the agent layer that it effectively set the terms, and that success clearly rattled the incumbents.
When Google and Microsoft, two companies that rarely agree on anything, publicly align on a rival directory standard, it signals how much MCP reshaped the board.
Notably, neither Anthropic nor OpenAI appears on ARD’s initial supporter list. That absence is the whole story. The agent layer is turning into a standards war, and control of how agents discover and reach tools is control of the ecosystem that gets built on top.
For ordinary users, agentic software is the layer that quietly books, fetches and files things on your behalf. The standard sitting underneath it decides how smoothly an agent can reach the apps you already use, which is why a directory spec most people will never see still matters to how these tools feel day to day.
For everyone downstream, the risk is fragmentation. Competing standards can mean duplicated work and tools that only play nicely inside one vendor’s garden.
The next thing to watch is whether Anthropic and OpenAI ignore ARD, adopt it, or push a counter of their own, because that choice will shape which agents can talk to what for years to come.
