Your Customer's AI Agent Is Coming to Buy a Car. Who Answers for the Dealership?
Personal AI agents stopped being a demo this fall. Meta launched Muse in September, a consumer agent that can browse, fill out forms, book and buy on someone's behalf, and other assistants are heading the same direction. The obvious question for our industry: what happens when one of those agents shows up to buy a car?
Right now, most of them would do it the hard way. Sierra, one of the companies behind the new standard, points out that a personal agent often has to sign in as the customer and navigate pages built for people. At a dealership that means clicking through VDPs, filling out a lead form and hoping someone answers. The CRM logs it as a regular lead, and nobody at the store knows an agent was on the other end.
What Poppy is
In October, Sierra and Meta introduced the Personal Agent Protocol, nicknamed Poppy. It's an open standard for how a personal agent finds a business, says who it is, gets the customer's permission and gets work done. Launch partners included Shopify, Stripe and Walmart. A few days later Sierra published the draft and added 35 more design partners, from Bank of America and Mastercard to OpenAI, PayPal and Visa. The spec site now lists version 1.0 as published, and the full spec is open on GitHub.
The mechanics are simpler than they sound:
- Discovery. A business publishes one file at /.well-known/poppy.json that tells agents how to work with it, how customers sign in and what an agent is allowed to do.
- Identified sessions. The agent says it's an agent and who it's acting for. The business always knows.
- Customer sign-in. When a task needs an account, the customer signs in on the business's own page through OAuth and approves exactly what the agent can access.
- One session everywhere. The same session works across the business's website, its APIs and its own agent, so the customer doesn't start over.
- The business picks the channel. Agents can work through the website, through APIs built on standards like OpenAPI and MCP, or by talking to the company's own agent.
That last part is the one dealers should pay attention to. Poppy lets a business put its own agent on the other side of the conversation. Sierra's launch demo was a mortgage pre-approval: Meta's agent found Rocket's agent through Poppy, the customer signed in to share credit and income details, and the two agents worked through price, down payment and rate.
Why car buying breaks the easy examples
Most of the examples floating around are fixed-price retail. Return a jacket. Rebook a flight. The price is the price, and the agent just needs permission to act.
Buying a car doesn't work that way, and anyone who has worked a desk knows it:
- Price is negotiated. The number depends on the unit, where the store is against its objective, the incentives in play and how the deal is structured.
- There's usually a trade. Its value depends on condition nobody can verify from a chat window.
- Financing comes with rules. Payments, rates and F&I products carry disclosure requirements. An agent can't just throw out a monthly payment.
- Inventory has to be real. A unit that sold yesterday and still shows online is an annoyance for a person. For an agent shopping ten stores at once, it's a dead end and a reason to move on.
- The deal ends in person. Test drive, signing, delivery. Whatever two agents agree on has to land in front of a salesperson who already knows about it.
None of that is a reason to wait. It's the reason a dealership should set its own rules before agents start making assumptions for it.
Where the dealer's agent belongs
There's a growing argument that website providers should build the dealership's AI agent, since the website is where shoppers show up. We see it differently. A website shows the inventory and the advertised price. The information that actually puts a deal together, like pricing rules, trade values, which incentives apply and who's already working the customer, lives in the store's own systems and with its managers.
An agent that can only see the website can answer questions. An agent that works from the dealership's data, inside guardrails the managers set, can work a deal. That's the one a shopper's agent needs to reach, and it's the one that should hand the customer to the store when it's time for a person.
What a dealership can do now
You don't need to publish a poppy.json this month. You can get ready for when you do:
- Find out whether AI can read your site today. Many AI crawlers read pages without running scripts, so a site that hides inventory behind heavy JavaScript is close to invisible to them. Our AI readiness rankings show how dealer websites score by provider and brand, updated nightly.
- Clean up the data an agent will rely on. Accurate inventory, current pricing, correct hours and listings. A fast, stable site matters too. Check yours with our free dealership website speed test.
- Write down your rules. What can be quoted without a manager? Which incentives can be stated? When does a person have to step in? If you can't answer those for your own BDC, an agent can't follow them either.
- Make sure your CRM can tell the difference. A lead an agent negotiated shouldn't land in the same queue, with the same follow-up cadence, as a form fill from a person.
- Be upfront about AI on your side. In a Poppy conversation, both sides identify whether a person or an AI is speaking. Shoppers should get the same honesty from the dealership.
What we're building
At Dealer Data One we're building the dealer side of this: a dealership's own agent that answers shoppers and their AI agents, works from the store's data in Control Center, stays inside the pricing and approval rules the store sets, and hands off to the team when a person should take over. We're building it to support Poppy, and we've applied to join the Poppy design partner group so automotive retail has a voice in where the standard goes next.
You can see an early version running for a fictional store on our dealer agent demo. It runs on Control Center today and is built to feed the Unified Dealer System when it launches in 2027.
For a broader look at where AI actually helps a store, read our guide to AI for car dealerships or our ranking of AI agents for car dealerships. If you want to talk through what your store's rules for agents should look like, reach out.





