The Agentic Resource Discovery specification aims to standardize how AI agents find and use enterprise tools, with notable absences from the supporter list
The biggest names in enterprise software just drew a line in the sand. Google, Microsoft, and Salesforce are backing a new open standard called Agentic Resource Discovery, or ARD, designed to let AI agents automatically find and connect with the tools they need across corporate software stacks.
The specification, announced on June 17-18, represents a coordinated effort to shape how workplace AI automation evolves.
What ARD actually does #
ARD uses something called ai-catalog.json manifests, which allow AI agents to query a centralized registry and discover what tools and other agents are available for a given task. Instead of an AI agent needing to be told exactly where to find every tool it might need, it can just look them up.
The system enables secure, intent-based searches. An AI agent can describe what it’s trying to accomplish, and ARD helps it locate the right resources without requiring custom-built connections for every possible combination of enterprise software.
Organizations can publish their AI capabilities under their own domains, maintaining control over what gets shared and with whom. It’s a federated approach, meaning no single company owns the registry. Each enterprise manages its own catalog while participating in a broader discovery network.
The specification operates under Apache 2.0 licensing and is hosted in open repositories, with coordination involving open working groups and associations with the Linux Foundation.
The coalition, and who’s not in it #
Beyond the three headline backers, the coalition includes Cisco, Databricks, GitHub, GoDaddy, Hugging Face, NVIDIA, ServiceNow, and Snowflake.
OpenAI and Anthropic are nowhere on the initial supporter list. OpenAI and Anthropic have largely built their enterprise strategies around chatbot-centric solutions. ARD represents a fundamentally different vision: one where AI agents operate autonomously within existing software ecosystems, discovering and using tools without a conversational interface as the bottleneck.
The design of ARD actually favors integration with established applications like Google Workspace and Microsoft’s productivity suite.
ARD is explicitly designed to complement existing protocols rather than replace them. That includes Anthropic’s Model Context Protocol, which launched in late 2024, and Google’s own Agent2Agent protocol. The specification operates as what its creators describe as a lightweight federated layer compatible with current enterprise frameworks.
Why enterprise AI needs a standard right now #
ARD addresses fragmentation by providing a standardized method for publishing and discovering AI capabilities, significantly reducing the need for custom integrations that have historically plagued enterprise software deployments.
Without a common discovery layer, every autonomous workflow requires bespoke engineering. With ARD, an AI agent handling a procurement request could dynamically locate the approval system, the budget tool, and the vendor database without anyone having to hardcode those connections.
What this means for investors #
The ARD specification doesn’t involve any crypto tokens, blockchain components, or digital assets. It’s a pure software standards play.
ARD’s Apache 2.0 licensing means any company can implement it. That lowers the barrier to adoption but also limits the ability of any single backer to extract monopoly rents from the standard itself.
If ARD becomes the default discovery mechanism for enterprise AI agents, tools that don’t publish ARD-compatible manifests could find themselves invisible to automated workflows. Google, Microsoft, and Salesforce aren’t building ARD out of altruism: a world where AI agents automatically find and use Google Workspace, Microsoft 365, and Salesforce CRM is a world where their subscription revenue keeps growing. Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our