The Agentic Resource Discovery (ARD) Specification: What It Is, Who’s Backing It, and Why Snowflake’s Support Matters

When an AI agent needs to find a tool, an API, or another agent, how does it know where to look? Until now, developers had to wire these connections manually — a brittle, unscalable approach. The Agentic Resource Discovery (ARD) specification, announced on June 17, 2026, is designed to solve that problem by providing an open, standardized way for AI agents to dynamically discover capabilities across the web .

What Is the ARD Specification?

ARD is an open protocol for publishing, discovering, and verifying AI agent capabilities . It defines a machine-readable file — /.well-known/ai-catalog.json — that organizations host on their own domain. This file lists available agents, MCP servers, skills, and APIs, along with metadata about what each resource does and how to invoke it .

Crucially, ARD sits entirely before invocation. It helps a client find the right resource; once found, the resource is invoked through its own native protocol, such as MCP or function calling . This means ARD is not an execution runtime, a product, or a marketplace — it is a shared protocol layer being developed under a Linux Foundation working group .

The specification is currently in v0.9 draft and is Apache-2.0 licensed . An early census of 39 major sites, including all eleven named working-group members, found that none served a discoverable ai-catalog.json as of June 18, 2026, indicating adoption is still in its earliest stages .

The official specification site and a formal protocol-oriented specification are available at agenticresourcediscovery.org and agentdiscovery.io/docs .

Which Companies Are Backing ARD?

Eleven companies are listed as collaborators on the announcement :

• Microsoft (co-developer)
• Google (co-developer)
• GoDaddy (co-developer)
• Cisco
• Databricks
• GitHub
• Hugging Face
• NVIDIA
• Salesforce
• ServiceNow
• Snowflake

RV Guha, a key figure in the project, noted that Satya Nadella made the decision for Microsoft to participate, Alan Eustace for Google, and that Yahoo also joined as a co-developer. He described the effort as "fierce competitors, coming together to build plumbing the Web badly needed" .

What Each Key Backer Brings

• GoDaddy emphasized the identity-and-discovery combination in ARD, arguing that trusted discovery requires verified publisher identity .
• Google published the initial announcement and is co-developing the spec .
• Microsoft sees ARD as part of a secure common layer for publishing and indexing AI capabilities across its ecosystem .
• Hugging Face is positioning ARD as a foundational layer for an open agent ecosystem with broad community participation .
• GitHub launched Copilot Agent Finder alongside ARD, implementing the spec to let Copilot discover and invoke external agents and tools dynamically .

Snowflake's Support and the Agentic Enterprise Strategy

Snowflake announced its support for ARD on June 17, 2026, describing it as "an open protocol that standardizes how AI agents and tools are cataloged, searched and discovered across an enterprise" . This builds on Snowflake's broader strategic pivot to become the control plane for the agentic enterprise — a vision articulated throughout 2026 .

Key Elements of Snowflake's Agentic Strategy

• Snowflake Intelligence (CoWork, now GA) — a system for AI-powered data discovery, governance, and decision-making. Over 15,000 agents are already deployed in production .
• Cortex Agents (now GA) — enables developers to build data agents that can plan tasks, use tools, and generate responses from structured and unstructured data .
• Managed MCP Server (now GA) — Snowflake's hosted Model Context Protocol server for connecting agents to tools, serving as a secure bridge to external agents like Anthropic, Cursor, and Salesforce's Agentforce .
• Snowflake CoCo — a cross-platform coding agent announced at Summit 2026 .
• Snowflake Datastream — a native streaming service bridging real-time data delivery and autonomous code generation .
• Horizon Catalog — integrated with Apache Polaris for bidirectional Iceberg interoperability, extending governance and discovery .

ARD support fits specifically as the discovery layer for this agent ecosystem. It lets Snowflake's agents find third-party tools and lets third-party agents find Snowflake-hosted resources, moving Snowflake from being merely a data warehouse toward an open, interoperable agent platform . As Snowflake's own blog states, the company's support for ARD "builds on our broader commitment to open standards, from MCP to Apache Iceberg and Open Semantic Interchange" .

Related Vendor Developments

• GitHub Copilot Agent Finder — Launched June 17, 2026, implementing the ARD spec to let Copilot discover and invoke external agents and tools dynamically .
• Microsoft's ARD investment — Microsoft is developing ARD as part of a secure common layer for publishing and indexing AI capabilities across its ecosystem .
• Cisco/AGNTCY — Cisco's Outshift division launched the AGNTCY open-source project in March 2025 for agent interoperability, which predates ARD and shares overlapping goals of an "Internet of Agents." By July 2025, over 75 companies had joined AGNTCY, which was donated to the Linux Foundation with Cisco, Dell Technologies, Google Cloud, Oracle, and Red Hat as formative members .
• GoDaddy — Emphasized that trusted discovery requires verified publisher identity, a key design consideration in ARD .
• Hugging Face — Positioning ARD as a foundational layer for an open agent ecosystem .

Related Discovery Protocols

ARD is not the only protocol addressing agent discovery. The Agent-to-Agent (A2A) protocol uses Agent Cards as a first step toward discovery, but intentionally leaves dynamic, scalable discovery infrastructure "up to you" . An Agent Registration and Discovery Protocol (ARDP) has also been submitted as an IETF Internet-Draft, though it has no formal standards standing .

What This Means for the Agentic Web

ARD represents a significant attempt to create a decentralized, open discovery layer for AI agents — analogous to what DNS did for websites or what search engines did for web content. By standardizing how agents advertise and find each other, it could reduce the manual configuration burden that currently limits agent interoperability.

However, the specification is still a draft, and adoption is effectively zero as of its launch date. Its success will depend on whether the backing companies — including fierce competitors like Microsoft, Google, and Snowflake — follow through with real implementations.

For Snowflake, ARD is a natural fit. The company has been rapidly building out its agentic capabilities throughout 2026, and ARD provides the missing piece: a way for all those agents, tools, and data sources to find each other across organizational boundaries. If ARD gains traction, it could become the plumbing that connects the agentic enterprise.