What happens when an AI buyer's agent shows up at my dealership presenting a Mandate?

When an AI buyer's agent shows up presenting a Mandate, what has arrived is not a tire-kicker — it is a credentialed, ready-to-transact customer acting for a specific human buyer, carrying an instruction that already names what to buy, the ceiling on price, and who approves. The desk does not negotiate the way it would with a walk-in, because the agent cannot be charmed, worn down, or up-sold in the chair; it reads what the unit's address publishes and closes against stated numbers. The store's job is no longer to control the conversation. It is to be answerable: to have the unit resolve, decode, and price at its own address so that when the agent reads it, there is something there to read.

That is the shift, and it is worth being precise about it, because it inverts the muscle a good desk has spent years building. The next buyer for a growing share of inventory is a consumer's agent, and the agent's behavior is the opposite of a shopper's.

What a Mandate is, and why it is not a lead

A Mandate is the instruction the buyer's agent carries: what to buy, the ceiling on price, and who approves. It is not a lead in the BDC sense — nobody bought it, it was not resold, and it does not need to be worked. It is presented at the unit's address and read against what the address states. If the unit answers the Mandate's terms — right vehicle, price at or under the ceiling — it makes the agent's shortlist. If the unit's address returns nothing, the agent does not call to ask; it simply moves on, and the shortlist is built from the units that answered.

This is the mechanical reason a store cannot treat agent commerce as a marketing channel to buy into. There is no feed to rank in and no placement to purchase. The agent resolves the VIN, reads the price at values.vin, pulls the record at vhr.vin, and decides. Presence at the address is the entire game.

How the agent reads a unit

Every unit answers in three representations from one address: its page for a human, the same facts as typed JSON at that same address through content negotiation, and one MCP server that an agent connects to directly. The agent does not walk the lot and does not call the desk. It reads the price, reads the record behind the price, reads the history and the recalls and the value band, and matches all of it against the Mandate. A unit with nothing published returns nothing — not a maybe, not a call for details, just an empty read that drops it from consideration.

The store becomes answerable through three steps, the same three that open the rails for a human sale. Connect: one account at auth.vin opens every address for the store. Post: the units resolve, decode, and price at their own addresses. Transact: appraise, wholesale, and move on the same rails the buyer reads. There is no separate agent-facing build to maintain; the representation the agent reads is a face of the representation a person reads.

Closing against stated numbers, both sides on one record

The reason this closes faster rather than merely differently is that both chairs read one shared record. The record a consumer's agent reads is the record the desk prices from — the history behind the unit, the spread behind the price, the math behind the close. When the numbers are stated and sourced on both sides, the deal is not a contest over whose figure is real; it is a match against a ceiling that either clears or does not. The distrust tax — the reflex where every close starts from suspicion and the straight store subsidizes the crooked one — has nothing to grip, because there is no room to move a figure between boxes when the figure is stated at its address with its source.

That last point connects to how the deal is actually structured line by line. When the agent and the desk agree, the structure — trade, taxes, rate, and every F&I attach — is built as a grid where each line stands on its own, and the customer's side reads the same grid the desk reads. The mechanics of building that deal live at desking.vin, where the four-square gives way to a grid with nothing hidden inside another line. An agent, unable to be confused by a payment moved from one square to another, is precisely the customer that grid was built to serve.

Who signs, and how the sale is legally real

A store may reasonably ask what makes an agent-driven sale a real transaction rather than an inquiry. The transacting layer is made legally real by the dealer-of-record rail: cloud.motors.vin is Vin's licensed virtual dealership, the entity that can sign as dealer of record through an API so that a counterparty — a buyer, or the agent acting for one — can transact bringing only funds and credit. For a licensed rooftop, that rail is what lets an agent-presented Mandate settle end to end without the deal falling back on a lot visit and a wet signature.

The honest limit is the same one that runs through every rail here: dealers.vin does not sell the customer to you. There is no pay-to-rank feed and no bought placement. The agent arrives because the unit answered at its address, and it closes because the numbers were stated and matched the Mandate. What the store controls is not the agent's attention — it is whether its inventory is answerable at all.

To be answerable is a build, not a purchase. The store connects once, the units post at their addresses, and the same record closes deals from both chairs. When the agent and the desk agree, the deal structures line by line at desking.vin; when it settles, cloud.motors.vin is the licensed rail that makes it real.