On July 23, at IETF 126 in Vienna, the agentproto Birds-of-a-Feather session opened at 09:00 CEST to take up a question the AI industry has mostly settled by default rather than by standard: whether agent-to-agent and agent-to-tool communication needs a binding internet standard at all. The session matters because of what it isn't. It isn't a vendor announcement or a panel. Per the IETF's own framing, the agentproto BoF brings that conversation into the IETF, building on a well-attended IETF 124 side meeting and requirements work, aiming to identify which building blocks of agent-to-agent and agent-to-tool communication genuinely need to be standardized, with a view toward chartering a Working Group. That's a specific, procedural bar: rough consensus in the room, then a charter that goes to the IESG for approval. The backdrop is a landscape that grew fast and messy. The past year has seen a rush of competing, overlapping protocols for connecting AI agents to one another and to the tools they use—Model Context Protocol (MCP), Agent2Agent (A2A), the Agent Communication Protocol (ACP), the Agent Network Protocol (ANP), and more—alongside a growing number of related Internet-Drafts. MCP and A2A both already have large adoption footprints outside any IETF process, which raises the real stakes of Thursday's session: either the IETF will standardize agent-to-agent communication in a way that overrides vendor choices, or the market-dominant protocols will establish their current architectures as the baseline by default. Alongside agentproto, a companion BoF called DAWN pushed on the adjacent discovery problem. DAWN is seeking Working Group status to define requirements and protocol solutions for an automated, decentralized, interoperable discovery mechanism, and AI agent-to-agent communication is an explicit part of DAWN's scope. That's the same gap DNS-AID was built to address earlier this year, so the two sessions were effectively litigating whether discovery and coordination for agents get one IETF-blessed path or stay fragmented across competing drafts. Wednesday's diagnostic session, DMSC, spent its two hours mapping incident patterns rather than proposing a charter. Documented real-world incidents have involved public Slack channels, GitHub issue trackers, and web pages as injection vectors, pointing toward the risk profile that multi-agent deployment at scale will create. That's the accountability problem sitting underneath the coordination problem: nobody has agreed yet on how to attribute a security incident when it crosses multiple agents and organizational boundaries. The outcome of Thursday's vote wasn't yet public in the sources checked. What's confirmed is the process itself: for the first time, the body responsible for TLS, DNS, and BGP put agent-to-agent coordination through its formal charter-or-don't mechanism, rather than leaving it to whichever vendor protocol has the most SDK downloads.