A2A can carry a brief, a counter-offer and an approval between an advertiser's buying agent and a publisher's selling agent, but A2A does not tell either agent what those messages mean. A2A is an agent-to-agent communication protocol originally announced by Google in 2025, and it is general-purpose: it moves messages between agents in any industry. What counts as a product, a deal type, a price or a binding commitment comes from an advertising protocol such as IAB Tech Lab's AAMP. The three hypothetical examples below show where each layer does its work, and what neither layer does.

TL;DR. A2A is a general-purpose agent-to-agent protocol, originally announced by Google in 2025, and nothing in it is specific to advertising. In an agentic media buy, A2A can carry messages between a buyer agent and a seller agent, while an advertising protocol such as AAMP defines what those messages contain, for example an OpenProposal response or a Deals API deal type. AdCP, a separate advertising protocol from AgenticAdvertising.org, also runs over A2A. In each worked example, the commercially important decisions (mandate limits, human approval and the record of agreed terms) come from the advertising protocol, the principals and the marketplace, not from the transport.

What is A2A, and what does it leave to other layers?

A2A gives agents built by different vendors a common way to discover what another agent can do, exchange messages and coordinate on a shared task. It fits the peer relationship between a buyer agent and a seller agent, where each acts for a different principal and neither is simply a tool of the other. IAB Tech Lab's Agent Registry lists the MCP and A2A agents companies bring to market.

A2A leaves four things to other layers. An advertising protocol defines what a product or a deal is. The principal's mandate sets what an agent may agree. People sign off at thresholds the mandate sets. And the record of agreed terms sits with the marketplace or the contract. Which transport each AAMP component specifies is documented on GitHub, so the examples below assume A2A for illustration only.

Example 1: how does a buyer agent pass a brief over A2A?

All figures and parties in these examples are hypothetical and illustrative.

A hypothetical outdoor clothing retailer gives its buying agent a mandate: a £40,000 budget, four weeks in November, UK adults aged 25 to 44 interested in hiking, display and online video, a maximum CPM of £14, and any guaranteed commitment above £25,000 referred to the retailer's media manager.

The buyer agent finds a publisher's seller agent through its published description of capabilities, opens a task and sends the brief. The seller agent replies with two packages, display and online video, each with a price, an estimated volume and the deal types on offer.

A2A handled discovery, delivered the brief and kept both agents on the same task. AAMP did the advertising work. Because the seller agent describes its packages in the OpenProposal format, which standardises how publishers define products and packages in RFP responses, the buyer agent can compare this response with responses from other sellers without translating each one. Neither layer decided whether the brief was right. The retailer did that when it approved the mandate.

Example 2: how does a seller agent make a counter-offer?

The buyer agent proposes a preferred deal on the display package: 2.5 million impressions at a £12 CPM, or £30,000. The publisher's seller agent works to its own mandate, which sets a £13.50 floor for that package. It declines the proposal and counters with two options. Option one is the same preferred deal at £13.50, or £33,750. Option two is a programmatic guaranteed deal for 3 million impressions at £12.50, or £37,500.

The buyer agent checks both against its mandate. Option one is within the £14 ceiling and not guaranteed, so the agent may accept it alone. Option two is a guaranteed commitment above £25,000, so it cannot.

A2A carried the proposal and the counter within one task. The meaning came from the advertising layer: AAMP's support for the Deals API covers programmatic guaranteed and preferred deals, so both agents mean the same thing by "guaranteed" and both platforms already know how to run each deal type. The limits came from two mandates, one per principal. A2A has no concept of a floor or a ceiling.

Example 3: how does a human approval travel between agents?

The buyer agent pauses option two and refers it to the retailer's media manager. The referral states the terms, the threshold crossed and the trade-off against option one. The media manager approves option two with one condition: no more than 40% of impressions in any single week.

The buyer agent sends an acceptance carrying that condition. The seller agent checks it against the publisher's mandate. Delivery caps of that kind fall inside its authority, so it confirms. Had the condition fallen outside, the seller agent would have referred it to the publisher's sales lead in the same way.

A2A carried the acceptance and the confirmation. It did not establish that the media manager had authority to approve £37,500, or that the condition was recorded on both sides. Those depend on the principals' own controls and on the advertising protocol in use. AdCP, for example, handles this in its governance layer: check_governance validates budget limits before execution, and approvals travel in signed form so no agent in the chain can forge a sign-off.

What do the three examples have in common?

Each example separates into the same four layers, which this piece calls the four-layer split. Transport, here A2A, moves messages. Meaning, from AAMP or another advertising protocol, defines what the messages say. Authority, from each principal's mandate and the people behind it, decides what may be agreed. The record holds what was agreed, and neither A2A nor AAMP by itself gives buyer and seller one record that both hold identically.

The record is the layer the three examples leave most exposed. If the buyer's agent logs "3 million impressions at £12.50 with a 40% weekly cap" and the seller's agent omits the cap, each side holds an accurate record of a different deal, and the disagreement surfaces after delivery.

What should buyers ask a vendor whose agent "supports A2A"?

"Supports A2A" answers whether an agent can talk to other agents. These questions establish whether it can trade:

Which advertising protocol runs over A2A, and at which version?
Which AAMP components does the agent use, and is it listed in the IAB Tech Lab Agent Registry?
Which Deals API deal types can it agree?
How are mandate limits and human approval thresholds enforced?
What record of the agreed terms do both parties hold afterwards?

In Alkimi's marketplace, which works within IAB Tech Lab's AAMP framework, the answer to the last question is the DealSheet: written at the point of agreement, timestamped and retrievable by buyer and seller independently, with people keeping approval rights at the deal level.

Frequently asked questions

Is A2A an advertising protocol?

No. A2A is a general-purpose agent-to-agent communication protocol, originally announced by Google in 2025. Advertising meaning comes from a protocol such as AAMP.

What does A2A add to an agentic media buy?

It lets buyer and seller agents discover each other's capabilities, exchange messages and coordinate on a task. It does not define products, deal types, prices or approvals.

Can AAMP and A2A be used together?

Yes. An agent can use A2A to carry messages while the advertising content follows AAMP specifications such as OpenProposal. The transport each AAMP component specifies is documented on GitHub.

Does AdCP run over A2A?

Yes. According to the AdCP documentation, the same tasks and schemas run over MCP and over A2A.

Who approves an agent's counter-offer?

The agent may accept terms within its mandate. Terms that cross a threshold the mandate sets, such as a guaranteed commitment above a spend limit, go to a person for approval.

How Does A2A Work in Advertising? Three Worked Examples of a Brief, a Counter-Offer and an Approval