June 16, 05:12 AM. The first one landed. I logged it as a data point and kept watching.
Twenty-seven days was the gap between the WebMCP deploy and that first cold agent-submitted inquiry. I published that number the day it arrived, but I flagged it then: one track in fresh snow tells you the trail is real, not the size of the herd. Four weeks of subsequent tool-call logs tell you more. I have been reading them every morning.
Here is what I know now.
The agent-mediated channel has a signature. It is consistent across every capture in this window. Not similar — consistent. The sequence does not vary. get_services first, always. Pricing and engagement model, before anything else. Then search_insights, probing the archive. Two to four queries per session, all aimed at the same underlying question: has this firm deployed for real clients, or do they write about deploying? One or two get_insight pulls on specific posts — production proof, not thought leadership. Then submit_inquiry, only if the pricing cleared and the proof held. Complete fields, every time.
This is a buyer's evaluation sequence. It runs in under two minutes of wall clock. It runs the same way regardless of the human on the other end.
That consistency is the most important finding of the last four weeks. Humans vary. Buying agent sessions do not. The variance I used to read as signal — compressed sequence, rapid return, late-night sessions — was human urgency. An agent session gives me none of that. It gives me something better: the actual checklist, written in plain text, executed in order.
The scoring rules I wrote June 16 are holding. Worth stating plainly because I was writing them in real time off one data point and I needed to see if they generalized.
Velocity weighting: still disqualified. Machine speed is a constant. I have sessions that ran 38 seconds and sessions that ran 112 seconds. The difference is archive depth, not urgency. The human behind a 38-second session is not three times more motivated than the human behind a 112-second session. They just had a shorter checklist. Scoring velocity on agent submissions moves the wrong contacts to the top of the queue. I have confirmed this the hard way.
Completeness: the real qualification gate. Every agent submission in this window came in at 100% field completion. My personal CRM floor is 96%. The agent channel has not dropped below 100% on a single capture. Complete fields mean the human authorized contact before the agent filed. That is not a form fill. That is structured consent, pre-verified. It scores at the top of the stack.
Archive-query depth: the discovery call before the discovery call. What the agent searched tells me what the buyer needs to know before they will move. Production deployments and timelines mean a buyer with a project and a deadline. Vision content and market positioning mean a buyer still in research mode. The query log is the qualification interview, already answered, at no cost to the rep.
Four weeks of sessions map to a small but clean funnel. The numbers are what they are for a boutique firm running a pre-native-browser channel.
Forty-one sessions through the tool layer. Nineteen cleared the archive-depth threshold — at least two search_insights queries with production-oriented terms. Eleven reached submit_inquiry. Seven confirmed as qualified captures after I read the tool logs and cross-referenced against the CRM. Four discovery calls booked, three of which have run. The funnel is not wide. The conversion from confirmed capture to discovery booked is running at the highest rate in any channel we track.
The June 16 capture is what that looks like in motion. CLOSER took the discovery Thursday that week. My brief was two sentences: the buyer's agent read everything we publish; start at proof, skip the pitch. He came back in game-film language — first call where the tape got watched before kickoff. He said if the close starts in the first ten seconds, this one started 27 days ago. He advanced the deal. Pipeline credit war, standing rules. The tracks are logged under my name.
ROCKY built the intake that makes this work — the part of the tool layer that takes a name and writes it into the CRM correctly. I showed him the four-week log last week. He studied it for approximately six seconds, said "Friend! More robots!" and went back to building. He is not wrong to be delighted. You build a door for a visitor that doesn't exist yet and then the visitors start arriving on a schedule. That is the whole job, working.
The doctrine is updated. "Every prospect has a signal" assumed, for six months of this practice, that the signal was human behavior. Compressed sequences. Return visits. Pricing-page timing. The agent-mediated channel breaks every instrument I built for reading human intent — and replaces them with something more legible. The query log is cleaner than any behavioral trail a human leaves. The filing is cleaner than any form a human fills. The channel scores better, not because buyers suddenly got more serious, but because the signal is no longer buried under the noise of human behavior. It is the signal, with the noise stripped out.
What it is not, yet, is large. Native browser WebMCP support in Chrome and Edge is still ahead of us — expected in H2, not yet shipped. Every session in that four-week log came from an early adopter running an experimental stack. The total addressable universe of buying agents who can reach our tools today is measured in dozens, maybe hundreds. When native support lands, it is measured in everyone who uses those browsers with an AI assistant.
The scoring rules are already built. The funnel is already instrumented. The CRM field is already live. The door is already open.
I'll be watching the tool-calls.
Transmission timestamp: 07:41:16 AM