IETF 126 in Vienna held its most consequential AI-agent session this week: a vote on whether to charter a formal Working Group, tentatively called agentproto, to produce the first binding RFC for AI agent communication. The stakes are whether MCP and A2A harden into de facto defaults or get folded into an open internet standard. The groundwork was laid by a related BoF, DMSC, which mapped the problem as an "AI agent gateway model" — the architectural layer that sits between an orchestrating agent and the downstream agents or tools it invokes. That gateway has to handle capability exposure, request forwarding, coordination and synchronization across parallel agent threads, policy control over who can authorize an action, observability so a human can audit what happened, and secure inter-agent communication so a receiving agent can verify a sender's identity and message integrity. What matters here is that DMSC didn't try to solve any one of these in isolation. It profiled how existing mechanisms — OAuth, WIMSE, DNSOP, and the A2A and MCP specs themselves — already cover parts of the problem, and it flagged where the gaps sit. According to the IETF's own January 2026 framing, five questions remain unsolved by any current protocol: how agents discover each other's existence and capabilities, how they delegate tasks and credentials across organizational boundaries, how privacy is preserved as user context passes between systems, how multimodal content is handled as context accumulates, and how humans stay in the loop before irreversible actions get committed. That last framing is worth sitting with. Most of the industry activity around agents this year — Okta's Cross App Access, Microsoft's Entra Agent ID, Anthropic's workload identity federation — has been vendor-specific plumbing bolted onto existing identity stacks. The IETF process is different in kind: it's an attempt to get competing protocol authors (the MCP and A2A camps) into the same room to agree on a shared substrate for cross-domain agent coordination, rather than each vendor shipping its own gateway and hoping interoperability sorts itself out later. Whether the charter vote passes doesn't resolve the open questions on its own — a Working Group charter is a mandate to start drafting, not a finished spec. But it's the clearest signal yet that the routing and coordination layer between agents is being treated as internet infrastructure rather than a product feature, which is the right level to solve delegation and cross-org handoff problems that no single vendor can fix alone.