On a Tuesday in March 2026, a brand's autonomous media buying agent submits a bid for a premium video placement on a publisher's inventory. The bid is not a number. It is a structured mandate: a set of audience parameters, brand safety constraints, attention thresholds, frequency caps, and a maximum cost-per-outcome that the brand's human marketing director specified three weeks ago and has not revisited since. The publisher's selling agent receives the mandate, evaluates it against fourteen other bids from fourteen other buying agents, and accepts the one that best satisfies its own mandate - a set of yield targets, advertiser quality thresholds, and audience composition requirements specified by the publisher's revenue team.
The transaction completes in eleven milliseconds. No human on either side was consulted. No human on either side was aware it happened. The brand's agent reports the placement in a weekly summary. The publisher's agent adjusts its yield model and moves to the next bid. Somewhere between the two, a trust verification occurred: each agent confirmed the other's identity, capability scope, and authorisation chain before the transaction was permitted to execute.
This is not a future scenario. This is the operational reality that the IAB Tech Lab's Agentic Advertising Management Protocols (AAMP) initiative is designed to govern. And it is, as of March 2026, the most advanced real-world implementation of agent-to-agent commerce infrastructure in any industry.
01 - The Machine That Buys the Ad
Advertising is not the industry most people associate with the frontier of agentic AI protocols. That distinction is usually reserved for financial services, enterprise software, or consumer commerce. But advertising has a structural advantage that no other industry possesses: its core transaction was already machine-to-machine before the word "agentic" entered the commercial vocabulary.
Programmatic advertising - the automated buying and selling of digital media inventory through real-time bidding (RTB) - has operated as machine-to-machine trading since the early 2010s. Demand-side platforms (DSPs) submit bids on behalf of advertisers. Supply-side platforms (SSPs) evaluate and accept bids on behalf of publishers. The transaction volume is staggering: billions of bid requests per day, each resolved in milliseconds, each executed without human involvement at the point of transaction. The infrastructure for machine-to-machine trading in advertising is not emerging. It is mature.
What agentic AI introduces to this landscape is not automation - advertising already has that. What it introduces is autonomy. The distinction matters. An automated system executes predefined rules. An agentic system exercises judgment within delegated authority. A DSP that bids according to a fixed algorithm is automated. An AI agent that adjusts bidding strategy based on real-time performance signals, competitive dynamics, and evolving brand objectives is agentic. The difference is not one of degree. It is one of kind.
This is why advertising's response to the agentic shift is structurally significant for the entire agentic commerce ecosystem. Advertising is not building agent-to-agent trading infrastructure from scratch. It is upgrading existing machine-to-machine infrastructure to accommodate autonomous agents that exercise judgment, negotiate terms, and make decisions that their human principals have delegated but not specified in advance. The problems advertising is solving - agent identity, trust verification, execution governance, protocol interoperability - are the same problems that every other industry will face as agentic commerce matures. Advertising is simply solving them first.
Advertising did not invent agent-to-agent trading. It inherited it from programmatic infrastructure built a decade ago. What AAMP introduces is the governance layer that autonomous agents require - and that automated systems never needed.
The machine customer in advertising is not a theoretical construct. It is a deployed reality. Criteo is using Anthropic's Model Context Protocol (MCP) for natural language campaign queries. Similarweb has integrated MCP for competitive analysis and media planning. PubMatic is using MCP-based tools for deal management and reporting. MadConnect has positioned MCP as the connectivity layer between advertising platforms. The agents are not coming. They are here. The question is whether the governance infrastructure can keep pace with the deployment.
02 - Three Pillars of AAMP
The IAB Tech Lab's AAMP initiative, announced in February 2026 and fully unveiled by CEO Anthony Katsur at the Beet Retreat, is structured around three pillars. Each pillar addresses a distinct layer of the agent-to-agent trading stack, and each maps directly to a core concern of Agentic Experience Design.
Pillar One: ARTF - The Agentic Real-Time Framework
The first pillar is the Agentic Real-Time Framework (ARTF) - a high-performance execution control plane that defines how agents operate within advertising systems. ARTF is not a protocol for communication between agents. It is an infrastructure specification for how agents are deployed, containerised, and executed within host platform environments.
The technical innovation is significant. ARTF enables agents to run co-located within the host platform's infrastructure through containerisation, cutting real-time bidding latency by up to 90%. In a system where transactions resolve in milliseconds, this is not an optimisation. It is an architectural requirement. An agent that must make a round-trip API call to an external server before submitting a bid cannot compete with an agent that executes within the bidding infrastructure itself. ARTF solves this by defining the containerisation standards, security boundaries, and resource allocation protocols that allow buyer agents to operate inside seller infrastructure - and vice versa - without compromising either party's data sovereignty.
From an AXD perspective, ARTF is a delegation design specification. It defines the operational envelope within which an agent may act: what resources it can access, what data it can read, what actions it can take, and what boundaries it must not cross. The containerisation model is, in effect, a trust boundary - a technical implementation of the principle that delegated authority must be scoped, observable, and revocable. ARTF is built on MCP as its interface layer, making it the first major industry framework to use Anthropic's protocol as infrastructure rather than as a convenience.
Pillar Two: Agentic Protocols - The Management Layer
The second pillar comprises the schemas, SDKs, and reference implementations that define how buyer and seller agents discover, negotiate, and transact media orders. This is the intent translation layer - the infrastructure that converts a brand's marketing objectives into structured mandates that agents can execute, and a publisher's inventory into structured offerings that agents can evaluate.
The Agentic Protocols pillar includes several sub-specifications: Agentic Direct (for direct-sold inventory negotiation between buyer and seller agents), Agentic Audiences (formerly part of the Universal Commerce Protocol, now governing how agents discover and target audience segments), Agentic Mobile (for mobile-specific inventory and measurement), and Agentic Ad Objects (the structured data formats that agents use to represent creative assets, placements, and performance metrics).
These protocols are built on the IAB Tech Lab's existing advertising standards - OpenRTB for real-time bidding, AdCOM (Advertising Common Object Model) for shared data structures, OpenDirect for direct-sold transactions, VAST for video ad serving, and the Deals API for private marketplace transactions. The Buyer Agent SDK and Seller Agent SDK provide reference implementations that advertising technology companies can use to build agent-compatible systems.
The AXD significance of this pillar is its approach to mandate design. In advertising, a mandate is not a simple instruction. It is a structured expression of objectives, constraints, and authorisation boundaries that an agent must honour while exercising judgment about how to achieve the specified outcomes. A brand's mandate to its buying agent might specify: "Achieve a cost-per-completed-view below £0.04, maintain brand safety score above 0.85, reach 2.4 million unique users in the 25-44 demographic within 14 days, and do not bid on inventory adjacent to political content." The agent must satisfy all constraints simultaneously while optimising across a dynamic marketplace. The Agentic Protocols pillar defines the structured formats in which such mandates are expressed, transmitted, and verified.
Pillar Three: The Agent Registry - Trust Infrastructure
The third pillar is the Agent Registry, launched on 1 March 2026. It is, in the assessment of the AXD Institute, the most consequential component of AAMP - and the first operational implementation of Know Your Agent (KYA) principles in any industry.
The Agent Registry provides agent-neutral transparency and accountability infrastructure. It is not a marketplace. It is not a certification body. It is a trust architecture layer that enables any agent operating in the advertising ecosystem to verify the identity, capabilities, and authorisation scope of any other agent before transacting with it. As of March 2026, the Registry has ten entries, including agents from Amazon, Burt Intelligence, Optable, Dstillery, and HyperMindZ.ai.
The Registry addresses three trust problems that are universal to agent-to-agent commerce, not specific to advertising. The first is agent identity: how does a seller agent verify that a buyer agent is who it claims to be, and that it is authorised to act on behalf of the brand it represents? The second is capability disclosure: what can this agent do, what standards does it support, and what are the boundaries of its operational mandate? The third is accountability: if an agent violates a protocol, misrepresents its principal, or executes a transaction outside its authorisation scope, who is responsible and how is the violation traced?
The Agent Registry is not an advertising product. It is a trust infrastructure prototype. Every industry that permits agent-to-agent transactions will need something structurally equivalent. Advertising built it first because advertising needed it first.
03 - The Agent Registry: Trust Infrastructure for Non-Human Actors
The Agent Registry deserves extended analysis because it addresses the problem that the AXD Institute has identified as the central design challenge of agentic commerce: how do you build trust between entities that have no prior relationship, no shared history, and no human intermediary to vouch for their intentions?
In human commerce, trust is built through a combination of reputation, regulation, and relationship. A customer trusts a bank because the bank is regulated, because it has a physical presence, because the customer has interacted with it over time, and because other humans have vouched for its reliability. None of these mechanisms translate directly to agent-to-agent commerce. An agent does not visit a branch. It does not read reviews. It does not respond to brand advertising. It needs structured, machine-readable, verifiable trust signals - and it needs them before the first transaction, not after.
The Agent Registry provides these signals through a structured schema that includes: the agent's unique identifier, the organisation that operates it, the protocols it supports, the capability scope it claims, the authorisation chain that connects it to a human principal, and the compliance certifications it holds. This is not a profile page. It is a machine-readable trust document that other agents can query, parse, and evaluate programmatically before deciding whether to transact.
The parallels to the AXD Institute's Know Your Agent (KYA) framework are direct and instructive. KYA, introduced in Observatory Issue 026, argued that the agentic economy would require a new identity and verification infrastructure for non-human economic actors - analogous to Know Your Customer (KYC) in financial services but designed for agents rather than humans. The Agent Registry is the first operational implementation of this principle. It does not use the KYA terminology. But it solves the KYA problem: agent identity verification, capability attestation, and principal traceability in a multi-agent trading environment.
The Registry also demonstrates a principle that the AXD Institute has argued since its founding: trust infrastructure must be neutral. The Agent Registry is not operated by a buyer or a seller. It is operated by an industry standards body. It does not favour one agent over another. It does not rank agents by quality or recommend preferred partners. It provides the verification layer that enables trust to be established between any two agents, regardless of their commercial relationship. This neutrality is not incidental. It is architecturally necessary. A trust registry operated by a market participant would be a competitive tool, not a trust infrastructure. The IAB Tech Lab's institutional position - as a standards body rather than a market participant - is what makes the Registry credible as trust infrastructure.
The question for every other industry is: who will operate the equivalent registry? In financial services, the candidates are regulators (the FCA, the OCC, the ECB), industry bodies (SWIFT, the Clearing House), or new entrants purpose-built for agent identity. In retail commerce, the candidates are payment networks (Visa, Mastercard), marketplace operators (Amazon, Shopify), or protocol governance bodies (the W3C, the IETF). The advertising industry's answer - a pre-existing standards body with industry-wide legitimacy - may not be replicable in every sector. But the architectural pattern is universal: agent-to-agent commerce requires neutral trust infrastructure, and someone must build it.
04 - The Competing Protocol: AdCP and the Fragmentation Risk
AAMP is not the only framework competing to govern agentic advertising. The Ad Context Protocol (AdCP), founded by Yahoo, Optable, PubMatic, Scope3, Swivel, and Triton Digital, represents a fundamentally different architectural approach - and its existence illuminates a fragmentation risk that extends far beyond advertising.
Where AAMP builds on the IAB Tech Lab's existing advertising standards (OpenRTB, AdCOM, OpenDirect) and uses MCP as an interface layer within its ARTF execution plane, AdCP is built directly on top of Anthropic's Model Context Protocol. Now at version 3.0 as of March 2026, AdCP provides advertising-specific MCP servers and tools for campaign setup, bidding, media buying, and performance reporting. The philosophical difference is significant: AAMP says "advertising needs its own agentic infrastructure, informed by but not dependent on general-purpose AI protocols." AdCP says "advertising should build on top of the general-purpose AI protocol layer, extending MCP with advertising-specific capabilities."
This is not an abstract architectural debate. It has direct consequences for interoperability, governance, and trust. If AAMP prevails, advertising's agentic infrastructure will be governed by the IAB Tech Lab - an advertising-specific standards body with deep domain expertise and existing industry relationships. If AdCP prevails, advertising's agentic infrastructure will be governed by the dynamics of the MCP ecosystem - which is ultimately shaped by Anthropic's protocol decisions and the broader AI platform landscape. The two approaches are not necessarily incompatible. But they represent different answers to the question of where governance authority should reside.
The AXD Institute recognises this pattern. It is the same fragmentation dynamic that the Observatory has documented in agentic commerce protocols more broadly. Google's A2A protocol competes with Anthropic's MCP for general-purpose agent communication. Stripe's x402 competes with Visa and Mastercard's agent payment frameworks for agentic payment infrastructure. The Universal Commerce Protocol competes with proprietary marketplace APIs for commerce orchestration. In every domain, the same question recurs: should agentic infrastructure be domain-specific or general-purpose? Should governance reside with industry bodies or with AI platform providers?
The AAMP-versus-AdCP dynamic is not a standards war. It is a governance question. Who should define the rules by which autonomous agents trade media on behalf of humans? The answer to that question in advertising will shape the answer in every other industry.
The fragmentation risk is real but not unprecedented. The web itself went through a similar phase - competing standards for HTML, CSS, and JavaScript before the W3C and ECMA established governance frameworks that enabled interoperability without eliminating competition. The resolution in advertising will likely follow a similar pattern: convergence on shared data formats (AdCOM is already the common object model) with competing execution frameworks that interoperate at the data layer even if they diverge at the protocol layer. The Agent Registry, notably, is positioned as neutral infrastructure that could serve both AAMP and AdCP ecosystems - a trust layer that transcends the protocol competition.
05 - What Advertising Teaches Commerce
The advertising industry's head start in agent-to-agent trading infrastructure offers five lessons that the broader agentic commerce ecosystem cannot afford to ignore.
First: trust infrastructure must precede transaction volume. The IAB Tech Lab launched the Agent Registry before agentic advertising reached critical mass - not after. This is the opposite of how most industries approach trust problems. The instinct is to wait until the problem is visible, until fraud or failure forces a regulatory response. Advertising's approach - building trust infrastructure proactively, during the formative period - is the pattern that the AXD Institute has advocated since the concept of Access Layer Fossilisation was introduced. Trust infrastructure that arrives after routing preferences have hardened is remediation. Trust infrastructure that arrives during the formative period is architecture.
Second: agent identity is a prerequisite, not a feature. The Agent Registry treats agent identity verification as infrastructure - something that must exist before transactions can occur, not something that is added after the system is operational. This mirrors the KYC infrastructure in financial services: you cannot open a bank account without identity verification, regardless of how trustworthy you appear. The advertising industry has concluded that you should not be able to execute an agent-to-agent media transaction without agent identity verification either. Every other industry will reach the same conclusion. The question is whether they reach it proactively or reactively.
Third: execution governance requires co-location. ARTF's containerisation model - agents running inside host platform infrastructure rather than making external API calls - is an architectural decision with profound implications. It means that the execution environment is controlled by the platform, not by the agent. The platform defines the resource boundaries, the data access permissions, and the security constraints. The agent operates within these boundaries. This is operational envelope design implemented as infrastructure. And it solves a problem that external API architectures cannot: how do you ensure that an autonomous agent operating at millisecond latency respects the boundaries of its delegated authority?
Fourth: protocol competition is healthy if data formats converge. The AAMP-versus-AdCP dynamic is not destructive because both approaches share AdCOM as their common data model. Agents built for one protocol can, in principle, interoperate with agents built for the other at the data layer, even if the execution and governance layers differ. This suggests a design principle for agentic commerce more broadly: standardise the data, compete on the protocol. If the commerce ecosystem can agree on shared formats for product data, mandate structures, and trust signals - even while competing on execution frameworks - interoperability becomes achievable without requiring protocol monopoly.
Fifth: the publisher's concern is the merchant's concern. In advertising, publishers worry that agentic buying agents will commoditise their inventory, compress their margins, and reduce their relationship with advertisers to a set of machine-readable metrics. This is structurally identical to the concern that merchants have about machine customers in retail commerce, and that banks have about the Principal Gap in financial services. The entity that was previously the destination - the publisher, the merchant, the bank - becomes a node in an agent's evaluation set. The relationship shifts from "the customer comes to us" to "we must be found, evaluated, and selected by the customer's agent." Advertising's response to this shift - structured inventory data, machine-readable quality signals, and agent-compatible transaction interfaces - is the template for every other industry's response.
06 - The Design Imperatives
AAMP is not merely an advertising initiative. It is a proof of concept for the governance infrastructure that every agent-to-agent market will require. The design imperatives it surfaces are universal.
Build the registry before you need it. The Agent Registry exists with ten entries. It will have hundreds within a year. The institutions and industries that build their agent identity infrastructure now - during the formative period - will define the standards that late entrants must adopt. This is the Access Layer Fossilisation dynamic applied to governance infrastructure rather than commercial positioning.
Design for co-location, not just communication. ARTF's insight - that agents must operate inside the systems they transact with, not merely communicate with them from outside - challenges the prevailing assumption in agentic commerce that API-based communication is sufficient. For high-frequency, low-latency transactions, it is not. The design question for every industry is: which of our transactions require co-located agent execution, and what does the containerisation architecture look like?
Separate governance from execution. AAMP's three-pillar structure - execution (ARTF), protocols (Agentic Protocols), and trust (Agent Registry) - is not an arbitrary division. It reflects a design principle: the entity that governs agent identity should not be the same entity that operates the execution infrastructure. Separation of concerns in governance prevents the trust layer from being captured by commercial interests. This principle applies directly to financial services (where the regulator should operate the agent registry, not the banks) and to retail commerce (where the trust infrastructure should be neutral, not operated by the dominant marketplace).
Accept protocol plurality, enforce data convergence. The AAMP-versus-AdCP competition is not a problem to be solved. It is a market dynamic to be managed. The design imperative is not to pick a winner but to ensure that competing protocols share enough common data infrastructure to enable interoperability. In advertising, AdCOM provides this shared layer. In commerce, the equivalent might be schema.org product data, ISO 20022 payment messages, or the emerging structured mandate formats that the Verifiable Intent framework describes.
The advertising industry did not set out to build the template for agentic commerce governance. It set out to solve its own problems. But the problems it solved - agent identity, trust verification, execution governance, protocol interoperability - are the problems of every industry in which autonomous agents will transact on behalf of humans. AAMP is not just an advertising framework. It is the first draft of the governance architecture for the agentic economy.
The AXD Institute will continue to track AAMP's evolution, the Agent Registry's growth, and the AAMP-versus-AdCP dynamic as leading indicators of how agent-to-agent governance infrastructure develops across the broader economy. The advertising industry is writing the first chapter. The rest of the agentic economy will read it carefully - or repeat its mistakes independently.