Most explanations of AdCP stay abstract: it's a protocol, it standardises agent-to-agent communication, it covers discovery and proposals. True, and also not that useful if you've never actually seen what one of these exchanges looks like. So here's one, read the way a person would, with the bits that matter and the bits that still need a human flagged as we go.

The request

A buying agent sends a structured request: budget of £15,000, CTV inventory, US audience, brand safety tier "standard," minimum viewability 70%, flight dates two weeks out. In plain terms, that's a media buyer's brief, just written in a format a machine can parse without a human retyping it into a different system.

The response

A publisher's agent replies with three inventory packages that fit the criteria, each with price, format, estimated reach, and a viewability guarantee. One package sits at exactly 70% viewability and consumes the full budget. A second offers 78% viewability for 90% of the budget. A third is cheaper again at a lower viewability guarantee.

Read as a human would read this: the second option is probably the right call, slightly less inventory for a meaningfully safer viewability number, and that's a judgement, not something the protocol tells either agent to do. AdCP delivered three comparable, structured options. It did not tell the buying agent which one to pick, and it shouldn't, that's the actual job.

The counter

Say the buying agent counters on the second package, asking for the full budget at the 78% guarantee instead of 90%. The publisher's agent can accept, reject, or counter again. This is the part that looks most like "negotiation" and it's also the part our Concourse benchmark exists to test, because an agent can run this exchange fluently and still cave on terms a human negotiator would have held, or push so hard it kills a deal that was fine as offered.

Where the structured exchange ends and the open questions start

Say terms land: 90% of budget, 78% viewability, two-week flight. The AdCP exchange is now finished and both systems have a clean, machine-readable record of what was proposed and accepted. What it doesn't do: confirm anyone on the buyer's side was actually authorised to commit £13,500 to this specific publisher. Record the agreement anywhere both parties have a shared, tamper-evident copy to point back to. Flag it for a human to glance at before it goes live. Or check, two weeks later, that delivered viewability actually hit 78% rather than the number both sides agreed to on paper.

Why this walk-through matters more than the summary

Reading the whole exchange step by step instead of skimming "AdCP standardises discovery and proposals" makes the boundary obvious in a way the summary doesn't. The protocol did real work here, three comparable options instead of three phone calls, one clean counter instead of an email thread. It also stopped exactly at the point where judgement, authority, and accountability start to matter, because that was never the job it was built for.