Software is starting to buy advertising the way a person would. An agent takes a goal, finds inventory, asks a publisher's agent what it has, compares offers, books a campaign, and checks how it ran. None of that is hypothetical. PubMatic shipped a buy-side stack of agents in January 2026, and independent agencies have run live campaigns through it. The problem underneath the demos is older and more boring than the demos themselves: when an advertiser's agent talks to a publisher's agent, what language do they speak?
In the last six months two answers have appeared, and their backers do not like each other. One camp wants a new protocol designed for agents from scratch. The other wants to bolt an agentic layer onto the standards that already run programmatic. The acronyms are AdCP and AAMP, the names are easy to confuse, and the industry body behind one has called the other "deeply flawed." This post lays out who backs each, what layer each actually controls, and why a split here would cost the open web more than a split anywhere else in ad tech.
Origin: why agents need a shared language at all
Programmatic advertising has a precedent for exactly this problem. In the late 2000s, every demand-side platform and every exchange spoke its own dialect, and connecting a buyer to a seller meant a custom integration each time. OpenRTB, the real-time bidding standard the IAB shepherded from 2010 onward, fixed that. It defined a common bid request and bid response so any DSP could bid into any exchange. That single agreement is a large part of why programmatic scaled to handle billions of auctions a day.
Agents recreate the old mess at a higher level. A modern agent does not just submit a bid. It holds a conversation: it discovers what a publisher has, negotiates terms, requests creative, activates an audience, and reads back performance. Those exchanges happen in seconds or days, not the hundred milliseconds of an auction, and they often sit outside the bidstream entirely, in direct deals and curated marketplaces. OpenRTB was never built for that kind of back-and-forth. So the same question that produced OpenRTB fifteen years ago has returned in a new form, and the technical groundwork for an answer already exists. Anthropic's Model Context Protocol, released in late 2024, gives any AI model a standard way to call external tools and data. A second pattern, agent-to-agent messaging, lets two agents talk directly. Both camps in this fight build on those foundations. They disagree about what to build on top.
Present: two camps, two philosophies
The Ad Context Protocol
The Ad Context Protocol, AdCP, launched on 15 October 2025. Its six founding members are Scope3, Yahoo, PubMatic, Swivel, Triton Digital, and Optable, each of which put in engineering time plus a 10,000 dollar annual budget, and the supporter list has since grown past twenty companies including Magnite and Kargo. Brian O'Kelley, the founder of AppNexus and now chief executive of Scope3, has been the most visible figure behind it.
AdCP's pitch is that agents need a purpose-built language. It defines a set of domain-specific tasks, things like discovering products, creating a media buy, generating creative, and activating an audience, that an agent calls over MCP or agent-to-agent transport. The protocol is organised into modules covering signals, media buy, and creative, with a curation protocol due in 2026. It is governed by a new non-profit, AgenticAdvertising.org, also called the Agentic Advertising Organization, whose interim board includes former IAB chief executive Randall Rothenberg, O'Kelley, and Ebiquity's Ruben Schreurs. The group has said it is modelling its community governance on Prebid, the open-source header bidding project, and that no single company should control the protocol's evolution. AdCP reached general availability and a 3.0 release by April 2026.
Crucially, AdCP positions itself as a parallel to OpenRTB, not a replacement. In its own documentation the project frames the split plainly: OpenRTB handles the impression layer, the millisecond auction, while AdCP handles the campaign layer above it, the agent workflow. One AdCP instruction to create a media buy can spawn thousands of impression-level events underneath it.
AAMP, the IAB Tech Lab's answer
The IAB Tech Lab, the standards arm of the Interactive Advertising Bureau, took the opposite philosophical route. Rather than write a new language, it is extending the ones the industry already runs: OpenRTB, OpenDirect, AdCOM, and its data taxonomies, each getting an agentic layer. On 26 February 2026, in a post by chief executive Anthony Katsur, the Tech Lab gave that whole body of work a name: AAMP, the Agentic Advertising Management Protocols. The naming itself was a fix. Katsur said the market kept conflating two specific Tech Lab components with the entire program, and a single umbrella term was meant to end that confusion.
AAMP is structured in three layers. At the base sits Agentic Foundations, built around the Agentic Real Time Framework for fast, deterministic agent operation in the auction. In the middle sit the Agentic Protocols, a set of components with names like Agentic Direct, Agentic Audiences, Agentic Mobile, and Agentic Ad Object, each mapped onto an existing standard, plus reference implementations for buyer and seller agents. At the top sits Trust and Transparency, anchored by an Agent Registry. That registry went live on 1 March 2026 as a neutral, industry-governed directory where agents verify their identity, capabilities, and compliance status. By 11 March it listed ten agents, including ones from Amazon, PubMatic, Equativ, and Optable.
The Tech Lab's argument for its approach is institutional knowledge. Its standards, it says, encode lessons refined through billions of real transactions, and throwing that away to start fresh is wasteful. AAMP also has early proof of life in the wild. On 25 March 2026, Kochava opened its StationOne platform to public beta as an open-source AAMP workspace, letting teams test standards-compliant agent workflows, with 19 skills running through the Tech Lab's reference server, before risking real money.
The fight nobody is being coy about
Standards bodies usually feud in polite abstractions. This one has not. Katsur told The Current that "what we don't need is another industry trade group," and in the same coverage AdCP was characterised, from the Tech Lab's side, as deeply flawed and resource-consuming. The Tech Lab's consistent position is that foundational work belongs in a neutral, established forum, meaning its own, rather than in a separately governed organisation that appeared weeks earlier.
The friction is sharpened by an awkward overlap. Several companies sit in both camps. PubMatic is an AdCP founding member and an early AAMP roadmap participant. Optable appears in the AdCP founding six and in the Tech Lab's Agent Registry. Yahoo and others span both. The companies actually building agents are hedging, because they cannot afford to bet the integration on the losing standard.
Read past the rhetoric and the two are not, on paper, trying to occupy the same square. AdCP's own FAQ offers the cleanest summary anyone has put forward: AAMP is agentic bidding, AdCP is agentic buying. AAMP's work concentrates on impression-level concerns, how an agent is discovered and behaves inside the programmatic auction. AdCP concentrates on the campaign workflow above it, how a buyer agent and a seller agent negotiate and transact. In that reading the layers compose rather than collide, and an AdCP buy could sit on top of AAMP-governed bidding underneath. Commentators writing in April 2026 described exactly that stacked relationship.
So why the heat? Partly governance. Whoever owns the registry, the reference implementations, and the compliance rules holds real leverage over a market that could intermediate enormous spend, and two non-profits now claim that role. Partly it is timing and pride. The Tech Lab has run advertising standards for years and a well-funded outsider arriving first, with a former IAB chief executive on its board, reads as a challenge whatever the layer diagram says. The risk is that a turf question gets resolved by the industry picking sides, and a layering that should have been complementary calcifies into two stacks that need translation between them.
Future and impact: the real cost of a split
The honest forecast has three parts.
The first is that some convergence is the most-quoted hope and the least-evidenced outcome so far. The Tech Lab's framing is that any structured protocol for discovery, comparison, negotiation, and campaign management can feed into AAMP, and that agents from projects like AdCP should simply register in its Agent Registry and commit to transparency. That is an invitation to fold AdCP's work in under the Tech Lab's umbrella, on the Tech Lab's terms. AdCP's separate non-profit, its own funding, and its own board mean it is not obviously going to accept that. As of spring 2026 there is no joint technical roadmap between the two, only parallel ones and warm words about interoperability.
The second is that the practitioners who would have to live with two standards are not impressed by either yet. Skeptics quoted across trade press have made pointed objections. Permutive's co-founder argued that no publisher is forecasting a single dollar of revenue through AdCP, that it distracts from unsolved performance basics. The longtime fraud researcher Augustine Fou warned that more automation tends to mean less transparency, and that agents do not remove bad actors from the supply chain. Ozone's representative put the core fear directly: the biggest risk is agents decreasing transparency and control rather than increasing it. That is the same charge once aimed at programmatic itself, and it is the reason header bidding and supply-path optimisation exist. An agentic layer that hides which intermediaries it used, and what data it touched, behind a clean conversational interface would undo years of that cleanup work.
The third is the part specific to a standards war, and it is the one that should worry anyone who cares about the open web. Fragmentation here does not just slow adoption. It hands the outcome to the walled gardens. Google, Meta, and Amazon already control most programmatic spend, and each can ship a fully agentic buying experience inside its own platform with no need for any shared protocol at all, because both ends of the conversation are theirs. The open web cannot do that. Its only path to competitive agentic buying is a common standard that lets thousands of independent buyers and sellers interoperate. If AdCP and AAMP spend two years arguing over the shape of the rails, the open web does not get a worse agentic standard. It risks getting none that matters, while agentic spend quietly consolidates inside the closed platforms that never needed a standards committee. One analyst drew the parallel to the in-game advertising standards fight of the late 2000s, where rival camps argued over technical definitions until the category stalled and the moment passed. The technology worked. The industry just failed to agree in time.
For a marketer the practical posture is patience with preparation. Do not re-platform around either protocol while the layer boundaries are still being drawn, but do treat agent-readiness as real work that pays off under either outcome. Clean, structured, machine-readable product and audience data helps an agent whichever standard wins. Pushing every vendor to say plainly which protocols it supports, and to register its agents somewhere auditable, keeps the transparency question alive while the standards settle. This is where an implementation partner such as Perform Digital is useful: wiring an agent layer onto an existing stack without surrendering the auditability that makes the stack trustworthy in the first place, and staying deliberately protocol-agnostic until the picture clears. The right question to carry into every agentic pitch is the oldest one in ad tech, and it has not changed: which side are you on, and what is your cut.
Council summary
This post argues that AdCP and AAMP are not really competing for the same job, AdCP owns the campaign workflow and AAMP owns the auction layer, and that the real danger is not a worse standard but a stalled one that hands agentic spend to the walled gardens. Review confirmed the protocol names, backers, and launch dates: AdCP launched 15 October 2025 with six founding members, AAMP was named by Anthony Katsur on 26 February 2026, and the IAB Tech Lab Agent Registry opened on 1 March 2026. Two corrections were made: the AdCP founding commitment is a 10,000 dollar annual budget plus engineering time, and the Agentic Protocols components are Agentic Direct, Agentic Audiences, Agentic Mobile, and Agentic Ad Object, not the names the draft first listed. The quotes from Katsur, Permutive, Augustine Fou, and Ozone all check out against trade coverage. The takeaway for a marketer is to stay protocol-agnostic, get product and audience data agent-ready, and press every vendor to register its agents somewhere auditable.
Comments