🇬🇧 CM AI Community: From One Working Node to an Open Question

A follow-up, not a restatement — if you’re looking for the original proof-of-concept writeup, that’s still here: [link na prejšnji članek]. This one picks up where that left off.

A few weeks ago, this Journal described a working proof of concept: an AI Docking Port node, live and answering real queries, with no central directory telling it where to look. Since then, three things changed the shape of the question.

First, it stopped being one project’s idea.

Rouiada Abbas, author of the Vacation Rental Protocol (VRP), read the original piece and did something more useful than compliment it — she tested it. Her own tooling independently verified apartmamatevz.si as a second, genuinely separate VRP signing node, the exact bar her protocol’s own governance requires before it can call itself a real standard rather than one company’s format. What followed was a real collaboration: a joint technical proposal — not a pitch for either project individually, but a description of one shared gap (how does an agent discover an independent provider, and then trust what it finds) with two adjacent, independently-built answers.

Second, the gap turned out to be measurable, not theoretical.

A cold-start test — ask a general-purpose AI agent about this property with no URL given — showed the actual failure mode: the agent found the listing through ordinary search, never touched the signed, verifiable source, and in one run listed the direct booking page alongside two commission-based platforms as three interchangeable options. Publishing a machine-readable manifest doesn’t mean an agent knows to check it first, or at all. That distinction is now written into the proposal as something to actually test, not assume.

Third, a second independent node went live

— this time on a genuinely separate network, proving the discovery-and-forwarding mechanism works across real infrastructure boundaries, not just in principle.

###One open question followed directly from building that mechanism:

Should the underlying discovery layer — call it the Reception Protocol — become its own standard, the way VRP did? The honest answer turned out to be: probably not alone.

A W3C Community Group already exists for exactly this problem space, open and still early. If the goal is a genuinely open protocol, it shouldn’t matter whose name is on it — so rather than building a competing standard, the proposal now on the table is a contribution: a concrete answer to a real risk (uncontrolled traffic growth across forwarding nodes) that the existing draft hasn’t addressed yet.

None of this is finished.

The joint proposal is still being reviewed before it goes to the developers actually building agent infrastructure. The W3C contribution is one GitHub issue, not yet a conversation. What’s changed is the shape of the work — from “does this work at all” to “who else is solving adjacent pieces of the same problem, and where do we actually overlap.”

Update, 2026-11-15:

The depth/fanout mechanism described above is now also published as a standalone IETF Internet-Draft, draft-dvojmoc-reception-forwarding-limits-00 — a citable, stable reference for anyone reviewing the W3C contribution or building their own forwarding layer. Rouiada Abbas published her own VRP offer-verification Internet-Draft the same week, independently of this one.