TL;DR: Agentic advertising infrastructure is being built simultaneously at three layers: standards bodies defining how agents communicate, platforms providing the marketplace layer where deals are executed, and verification infrastructure providing independent evidence of what was agreed and delivered. Understanding which organisations are working at which layer, and how those layers must eventually connect, is the starting point for any serious evaluation of where the market is heading and which bets are durable.
The infrastructure question in agentic advertising is not which AI model is doing the buying. The model is replaceable. The infrastructure question is what governs the transaction: how agents establish shared terms, how those terms are recorded, how disputes are resolved, and how the whole system remains accountable to the humans who are responsible for the budgets.
In 2026, that infrastructure is being built simultaneously by multiple organisations at different layers of the stack. Some of those organisations are competing; some are complementary. Understanding the distinction requires a clear map of who is building what and where the gaps remain.
What are the three layers of agentic advertising infrastructure?
The infrastructure can be mapped across three distinct layers, each of which is a precondition for the layers above it.
The first layer is the communications protocol. For agents to negotiate a deal, they need a shared language: a specification that defines how a buy-side agent and a sell-side agent exchange offers, counter-offers, constraints, and confirmations. Without a common protocol, agent-to-agent negotiation requires bespoke integration for every pair of platforms, which does not scale.
The second layer is the marketplace layer. This is where the actual negotiation and deal execution occurs: the infrastructure that hosts the deal record, applies the approved mandate parameters, routes the negotiation between parties, and generates the bilateral record of what was agreed. This layer may or may not be operated by the same organisations building the protocol.
The third layer is the verification and evidence layer. This is independent infrastructure that records what was agreed, what was delivered, and what variances occurred, in a way that neither party can alter retroactively. It is the audit infrastructure, and it is where the accountability of the whole system rests.
Who is building at the protocol layer?
Two competing specifications have emerged as the primary candidates for the inter-agent communications protocol in advertising.
The IAB Tech Lab's Agent Communication Protocol, referred to in most industry documentation as AdCP, is the most widely referenced. The IAB Tech Lab published the initial AdCP specification in late 2025, covering the message format for deal initiation, parameter exchange, and execution confirmation. As of mid-2026, the specification is at version 1.2, with working groups active on the approval workflow extension, which governs how agents escalate decisions to human principals, and on the deal record schema, which defines the mandatory fields in a bilateral transaction record.
The Agentic Advertising Marketplace Protocol, published by an independent consortium that includes several demand-side and supply-side operators, takes a different architectural approach. Where AdCP defines a message format that works over existing programmatic infrastructure, AAMP defines a dedicated negotiation channel with its own state management. The practical difference is that AdCP is designed to be layered on top of existing SSP/DSP relationships, while AAMP requires a new integration.
In May 2026, Prebid.org announced that it would develop an AdCP adapter, which allows existing Prebid-integrated publishers to accept AdCP-compliant agent bids without a full platform rebuild. No equivalent adapter for AAMP has been announced by Prebid.org at the time of writing.
Neither specification is fully adopted. The majority of programmatic transactions in 2026 still use existing DSP/SSP infrastructure without any agent communication protocol. The protocol layer remains the most unresolved part of the infrastructure stack.
Who is building at the marketplace layer?
The marketplace layer is where the most commercial activity is concentrated, because it is where deals are executed and revenue is generated.
Several DSPs have added agentic buying interfaces, which allow buy-side agents to submit structured bids via API rather than through a human-operated interface. These interfaces are, in most cases, wrappers around existing auction infrastructure rather than genuine agent-to-agent negotiation environments. The buy-side agent submits a bid; the auction clears; the win notification comes back. The agent has automated the bid submission, but the deal has not been negotiated bilaterally.
Dedicated agentic marketplaces, which are built specifically for bilateral agent-to-agent negotiation rather than adapted from auction infrastructure, are a smaller category. These platforms generate a bilateral deal record at execution, hold mandate parameters separately from the negotiation itself, and provide both parties with access to the same authoritative record. The distinguishing technical feature is that the price and terms are set by negotiation between the two parties' agents, not by an auction that one party cannot observe.
Supply-side operators are at varying stages of readiness. Major publisher-side platforms have announced agent-readiness programmes, but the definition of "agent-ready" ranges from "accepts API-submitted orders without a human in the workflow" to "participates in bilateral agent-to-agent negotiation with a shared record."
The commercial implication for buyers is that "agentic" on a vendor's website may describe any point on this spectrum. The evaluation question is not whether the platform accepts agent input, but whether it generates a bilateral deal record that both parties can access and reconcile against.
Who is building at the verification layer?
The verification layer is the least developed and arguably the most important. Verification infrastructure in traditional programmatic is provided by independent measurement companies (DoubleVerify, IAS, MOAT before its acquisition) and by the MRC's accreditation framework. These organisations provide a third-party count of what was delivered and what met viewability and brand safety thresholds.
In agent-negotiated deals, verification needs to extend upstream of delivery. Verifying what was delivered does not answer the question of whether the deal terms were honoured. If the agent agreed to a specific placement, a specific audience parameter, and a specific price, and the delivery was technically on-spec but the deal record no longer reflects the original terms, delivery verification tells you nothing.
Several verification vendors have announced "agentic measurement" products in 2026. The definitions vary. Some are extending existing delivery measurement to API-submitted campaigns; others are building deal record verification infrastructure that audits the bilateral record against both the mandate and the delivery. The latter is the product that addresses the accountability gap; the former is conventional measurement relabelled.
The MRC has published a working document on agentic measurement standards, released for comment in Q2 2026. It sets out requirements for deal record access in accredited measurement products and specifies that the deal record must be accessible to the measurement provider independently of the platform that generated it. No accreditation under these standards has yet been awarded.
What does this map mean for buyers?
The immediate practical implication is that buyers evaluating agentic advertising vendors in 2026 are buying into a stack that is still being built. The protocol layer is unresolved; the marketplace layer varies significantly in what it actually provides; the verification layer is early.
That is not a reason to wait. The buyers who understand the infrastructure landscape are better positioned to ask the right evaluation questions, make the right contractual demands, and recognise when a vendor's "agentic" claim describes genuine infrastructure versus a rebranded workflow.
The durable bets in this landscape are on the layers that have the longest shelf life. The protocol specification will change; the mandate model will stay. The auction format will evolve; the requirement for a bilateral deal record will not. Infrastructure that is built around accountability, record integrity, and audit access will be useful regardless of which protocol wins the standards race.
The map is incomplete. That is not unusual for an infrastructure category at this stage of development. What is unusual is the pace at which buyers are being asked to commit production budgets before the map is finished. The appropriate response is to commit to the layer you can verify, and to build contract rights for the layers that are still forming.