Orchestration
IETF Opens the Door to a Standard for Agent-to-Agent Communication
At IETF 126 in Vienna last week, the Internet Engineering Task Force held its first Birds-of-a-Feather session dedicated squarely to how AI agents talk to each other across organizational boundaries. The session, called agentproto (Agent Communication Protocols), ran Thursday, July 23, and was one of three BoFs at the meeting touching AI agents directly, out of five scheduled overall.
A BoF isn't a standard and isn't a vote. It's the IETF's mechanism for testing whether a problem has enough community interest and rough consensus to justify chartering a Working Group. As one write-up of the session put it, the significant thing isn't that a spec got ratified — nothing did — but that agent-to-agent communication is now being treated as an internet-infrastructure problem, not just a vendor feature. That's a real shift in framing. Until now, the protocols agents use to find and hand off work to each other — things like Anthropic's Model Context Protocol and Google's Agent2Agent (A2A) — have been developed and iterated by individual companies or foundations, not standards bodies with the IETF's cross-vendor mandate and formal consensus process.
The timing lines up with a busy stretch for these protocols outside the IETF too. A2A shipped its first stable v1.0.0 specification back in March, tightening security by dropping legacy OAuth flows in favor of Device Code and PKCE support, adding mTLS, and formalizing signature verification for Agent Cards using JWS and JSON Canonicalization. MCP has its own release candidate in progress. What the agentproto BoF adds is the possibility — still years off, by the IETF's own admission — of a vendor-neutral RFC governing how agents route tasks and hand off context to one another, rather than each platform defining its own dialect and hoping others adopt it.
For anyone building multi-agent systems today, the practical takeaway is narrow but real: there's now an open, public venue where the wire-level questions of agent orchestration — discovery of capabilities, task handoff, session semantics across vendors — are being argued in front of the broader internet engineering community instead of inside a single company's spec repo. Whether that produces a chartered Working Group, and eventually an RFC, is still an open question. But the fact that IETF put agent communication on the agenda at all, alongside its established Working Groups, is itself the news.