On July 23, the Internet Engineering Task Force held a Birds-of-a-Feather session called agentproto — Agent Communication Protocols — at IETF 126 in Vienna. It was one of three AI-agent-related BoFs held during that meeting week, out of five BoFs total on the schedule. The session ran two hours and represents the first time agent-to-agent communication has been brought to the IETF as a candidate for a chartered Working Group, rather than sitting entirely in vendor-defined specs like MCP or A2A. A BoF isn't a standard and isn't a vote. It's an exploratory session where the IETF tests whether a technical area has enough shared understanding and volunteer energy to justify chartering formal standards work. The realistic path from here — if there's rough consensus to proceed — runs through Working Group formation, multiple draft revisions, interoperability testing, and eventually an RFC. That timeline is measured in years, not months, and plenty of BoFs never produce a chartered group at all. What makes this worth noting isn't the outcome (there isn't one yet) but the framing: agent-to-agent communication is now being discussed as internet infrastructure — the same category as DNS, TLS, or HTTP — rather than as a set of competing corporate protocols. That's a different posture than what's been happening at DIF, the Agentic AI Foundation, or FIDO Alliance, all of which have been building agent identity and authorization specs largely outside the traditional IETF process this year. If agentproto does get chartered, it would put agent communication semantics under the same open, rough-consensus governance that produced HTTP and TLS, with mandatory interoperability review and public draft history — a meaningfully higher bar for scrutiny than a single vendor's spec repo. The gap the BoF organizers pointed to is specific: existing agent protocols handle capability discovery and message passing reasonably well, but leave open questions about how independently-developed agent systems verify each other's communications across organizational boundaries without prior bilateral setup. That's the same seam identity-layer efforts (DID/VC-based frameworks, MCP-I/KYA-OS at DIF) are also probing, but from the transport and protocol-design side rather than the credentialing side. Whether IETF ends up chartering a Working Group, and how it reconciles with the identity-layer work already underway elsewhere, is worth watching over the next few months as the mailing list discussion either builds momentum or stalls out. For anyone building multi-agent systems today, the practical takeaway is that no standard exists yet — this was a discussion, not a spec. But it's a signal that the standards-track option is now live, which matters for anyone deciding whether to bet on a vendor protocol or wait.