At IETF 126 in Vienna last week, the agent discovery problem finally showed up on the standards-track agenda. The DAWN proposal — Decentralized Agent and Workload Discovery — is seeking Working Group status to define requirements for an automated, decentralized mechanism that lets agents and workloads find each other without pre-configured knowledge, and AI agent-to-agent communication is explicitly in scope. DAWN isn't starting from zero. It's meant to sit above a specific candidate solution that's already been adopted: DNS-AID, which the Linux Foundation accepted as an open standard in May 2026. The core idea is unglamorous but important — DNS-AID lets AI agents publish their endpoints and capabilities as standard DNS records, so discovery doesn't require a centralized registry that everyone has to trust and query. That's a meaningfully different bet than the registry-and-catalog model other efforts are pursuing. The context matters here. MCP and A2A have both accumulated real adoption — MCP, donated by Anthropic to the Linux Foundation's Agentic AI Foundation, had reached 97 million monthly SDK downloads by early 2026, and A2A, donated by Google, had 150 organizational supporters including AWS, Cisco, IBM, Microsoft, Salesforce, SAP, and ServiceNow. But adoption numbers don't solve the layering problem: MCP standardizes how a single agent connects to external tools and data, while A2A standardizes how agents discover each other's capabilities and delegate tasks peer-to-peer. Neither addresses how an agent discovers a *stranger* agent across an organizational or domain boundary it has no prior relationship with — which is the actual gap DNS-AID and DAWN are aimed at. This is a separate track from the Agentic Resource Discovery (ARD) specification Google and partners announced in mid-July, which builds a discovery layer on catalogs and federated registries sitting on top of MCP and OpenAPI. Where ARD leans on registries and domain-based catalogs, DNS-AID leans on DNS itself as the substrate — using ownership of a domain, resolvable at internet scale with existing infrastructure and no new lookup service, as the anchor for finding an agent's endpoint. Both approaches are converging on the same underlying question: once there are millions of autonomous agents, how does one find another it's never talked to before, without a single company or consortium controlling the index? The Vienna Birds-of-a-Feather session was reportedly aimed at deciding whether this becomes a chartered IETF working group with a path to a formal RFC, rather than staying a set of competing vendor specs. That decision — inside a standards body rather than a single company's blog post — is the part worth watching.