Carbuki Insights

AI Agents Just Agreed on How to Talk to Each Other. Your Dealership Systems Still Haven't.

August 24, 2026

Organizations supporting the A2A protocol

Milestone counts reported by Google Cloud (June 2025) and the Linux Foundation (April 2026). Each figure is a 'more than' threshold announced at that date, not an audited census.

On August 17, Axios reported that Agent2Agent - the protocol Google built so AI agents from different vendors can talk to one another - is moving into the Agentic AI Foundation, the Linux Foundation body that already hosts Anthropic's Model Context Protocol. The story ran for about a day in the tech press. It did not reach automotive at all.

It is still one of the more useful technology stories of the year for a dealer, because it concerns the part of an AI purchase nobody demos: the wiring between the software you just bought and the systems you already run. That wiring is where most dealership AI projects actually stall, and it is the line item that quietly decides whether you can switch vendors in three years or not.

Myth vs. data
The assumption is that dealership AI is limited by the quality of the AI.
A2A went from first release to a production-ready 1.0 specification backed by more than 150 organizations in twelve months, per the Linux Foundation (April 2026).
The proposed standard for how a dealership's own systems expose inventory, leads, deals and service to an agent is still a draft - RFC v0.1, openly recruiting founding partners (Automotive MCP Standards Council, 2026).
The models standardized faster than the plumbing underneath them.

Two protocols, two different problems

The two names get used interchangeably in vendor decks. They are not the same thing, and the difference matters when you are reading a contract.

Model Context Protocol (MCP) governs the vertical connection: how one AI agent reaches tools and data inside your walls. Your DMS, your CRM, your inventory feed, your scheduler. It is the layer that decides whether an AI phone agent can actually see an open service slot on Thursday at 10, or is guessing.

Agent2Agent (A2A) governs the horizontal connection: how two independent agents, built by different companies, discover each other, negotiate a task, and hand work back and forth across an organizational boundary. Your voice agent and your CRM's follow-up agent are, today, two strangers who happen to share a customer.

Google donated A2A to the Linux Foundation in June 2025 alongside AWS, Cisco, Microsoft, Salesforce, SAP and ServiceNow, with more than 100 companies supporting it at the time. By its one-year mark in April 2026, the Linux Foundation put the count above 150, with the protocol embedded in Azure AI Foundry, Copilot Studio and Amazon Bedrock AgentCore, and a stable 1.0 release that added cryptographic identity verification for agents. The Agentic AI Foundation itself, launched in December 2025, went from fewer than 40 members to more than 250 by this month, according to Axios.

The stated point of all of it, in the Linux Foundation's own words, is letting agents coordinate without being locked into a single vendor's ecosystem.

Layer	What it standardizes	Where it stands in August 2026
MCP	How an agent reaches tools and data inside your systems	Linux Foundation project; the integration pattern most dealership AI vendors now reference
A2A	How agents from different vendors coordinate with each other	Version 1.0 stable; 150+ supporting organizations; native support in major cloud AI platforms
AP2	How agents carry payment authority	Announced with 60+ organizations across payments and financial services
Automotive MCP (proposed)	Dealership-specific vocabulary: inventory, leads, deals, service, parts, F&I, signals, consent	Draft RFC v0.1; unratified; certification program not yet open

Why a protocol story lands on the service drive

Because the cost of not having a standard is not abstract. It is the reason the same connector gets rebuilt, badly, at store after store.

The Automotive MCP Standards Council - a group proposing a dealership-specific profile on top of MCP - describes the current state as N-by-M: every AI tool hand-wired to every dealer system, each pairing taking weeks of engineering and, by the council's own estimate, tens of thousands of dollars, then requiring maintenance as each system's API drifts. That estimate comes from a party with an interest in the standard existing, so treat the dollar figure as directional. The structural claim is harder to argue with: nothing built that way is reusable, and every connector is one more thing that can break at 6 p.m. on a Saturday.

Fullpath, in an April 2026 technical guide written for dealership tech directors, lays out the three ways a store actually gets there today: native vendor support built into the source system, middleware sitting between the system and the agent, or a customer data platform that has already unified the data. Its own assessment of option one is the telling part - "many legacy DMS and CRM systems don't offer this yet." That is a vendor writing about a category it sells into, and it is still the least flattering sentence in the piece.

There are real implementations. AutoUnify's ServiceMCP, announced as the first AI-commerce tool for automotive service scheduling built on MCP, lets an AI assistant book a repair order across multiple DMS platforms. That is one domain, from one vendor. It is a proof of concept for the idea, not a finished layer.

The number that should frame the decision

The average franchised dealership spent $586,246 on advertising in 2025, up 7.8% over 2024 - $739 for every new vehicle sold - according to NADA DATA covering 16,997 reporting dealerships. Search marketing, third-party sites and SEO absorbed roughly 60% of it.

That is the demand-generation side of the ledger, and it is well measured. The demand-handling side - the software that answers, routes, books and follows up on what that $586,246 produced - is where the integration bill lives, and most stores cannot state theirs as a single number.

Meanwhile Cox Automotive's Q2 2026 Dealer Sentiment Index, based on 958 dealers surveyed April 21 through May 4, put the cost-of-doing-business index at 74 on a scale where anything above 50 means costs are growing. A third of dealers named expenses among the factors holding back their business. Structural cost pressure is the environment every one of these renewals happens in.

Four questions worth asking before the next renewal

Open standards do not help a store that never asks about them. Four questions a GM can ask without a technical background:

Do you support MCP, and which version? The Automotive MCP council notes dealers are already being coached to ask this - and that there is currently no canonical automotive spec to answer against. A vendor who answers precisely is telling you something. A vendor who answers "yes, we're AI-native" is telling you something else.
If we leave, what leaves with us? Not just the customer records. The call transcripts, the appointment outcomes, the tuned behavior your team spent a year correcting. Ask for the export format in writing.
Who pays for the integration, and who owns it? If your DMS or CRM charges for API access and your AI vendor charges to build against it, you are paying twice for one connection. Get both numbers on one page.
What happens when the system on the other end changes? Schema drift is the ordinary failure mode, not the exotic one. Ask who is on the hook for the fix and how fast.

What standards will not fix

Interoperability is not intelligence. A conformant connection to your DMS does not make an AI agent good at handling a service call, and none of these protocols say anything about whether the thing answering your phone is worth answering with. A perfectly standardized agent can still mishandle an upset customer, quote a number it should not, or hand off badly to a human.

Nor is any of this settled. A2A is stable and MCP is widely adopted, but the automotive profile on top of them is a draft written by three named individuals recruiting founding partners. It may ratify. It may be superseded by whatever the large DMS vendors decide to publish on their own. Both outcomes are live, and a dealer does not need to bet on either one this quarter.

The reasonable move for the next 90 days

Do not restructure a technology stack around a governance announcement. Do three smaller things instead.

Write down every AI or automation tool touching a customer, and for each one, name the systems it reads from and writes to. Most stores have never had that on a single page. Then find the renewal dates. Then, on the next vendor call, ask question two above - what leaves with us - and notice how long the pause is.

The industry-wide version of this problem will get solved by committees over the next few years. The store-level version gets solved by knowing, before you sign, which connections you own and which ones you are renting.

If the phone is the part of that stack you are least sure about, that is the layer Carbuki works on - AI voice agents for U.S. dealerships, and a straight answer about what they connect to. carbuki.com

Sources

Protocol adoption counts are the figures each organization published at the dates shown and are self-reported. Vendor sources are labeled as such. NADA figures describe franchised new-car dealerships in aggregate, not any individual store.

Carbuki builds AI voice agents for retail automotive — answering sales and service calls, following up on leads, and booking appointments 24/7 in multiple languages.

See how it works →

Share:X LinkedIn Facebook Reddit Email

← All articles