Google, alongside Microsoft, Cisco, GitHub, Hugging Face, and eight other companies, has published the Agentic Resource Discovery (ARD) specification, an open standard for how AI agents find, verify, and connect to tools, APIs, and other agents across the web. The mechanics are deliberately narrow: organizations publish a machine-readable ai-catalog.json file at a well-known path on their domain, and registries crawl and index those catalogs so agents can search by task intent instead of relying on hardcoded integrations. The spec is explicit that it sits before invocation, not during it. It finds the right resource; the resource is then called through its own native protocol, whether that's MCP, A2A, or a plain API. Google's own Agent Registry, part of its Gemini Enterprise Agent Platform, builds on this by handling authenticated publisher onboarding, namespaced URNs, egress policy enforcement, and a cryptographic 'trust manifest' via Agent Identity meant to prove agent authenticity for compliance regimes like HIPAA. What's more interesting than the plumbing is who the spec quietly excludes. As one analysis of the launch put it, ARD is for publishers of callable capabilities — APIs, MCP servers, agents — and a company that exposes tools gets a clear path to being discovered and trusted by agents. A typical content website, by contrast, has no equivalent action to take today. That's a real fork in the road for how the open web adapts to agentic traffic: sites built for human browsing don't obviously benefit from a catalog file describing invocable services, because they don't have any. If ARD becomes the de facto discovery layer the way robots.txt became the de facto crawler protocol, the asymmetry between 'agent-native' businesses and everyone else could harden fast. Coverage of the launch also flags that the value of the whole effort is still debated — this is a draft spec with reference implementations, not a settled standard, and Google's own native support for it inside Agent Platform isn't live yet, just promised for 'the coming months.' The working group includes Databricks, GoDaddy, NVIDIA, Salesforce, ServiceNow, and Snowflake, which gives it more cross-vendor weight than most agent protocols get at launch, but adoption outside that circle is the real test. A discovery layer only matters once enough of the web actually publishes a catalog, and right now that's a commitment, not a default. For agent builders, the practical takeaway is narrower than the branding suggests: ARD doesn't replace MCP or A2A, and it doesn't make an agent trustworthy by itself — it makes an agent findable, with a verification hook bolted on. Whether 'findable and cryptographically checkable' converts into 'actually trusted' will depend on how registries score and rank the catalogs that show up, and that ranking logic isn't part of the open spec yet.