The Ad Context Protocol and the IAB Tech Lab's Agentic Advertising Marketplace Protocol are competing standards built by organisations that would each prefer to set the direction. They disagree on architecture, governance, and approach. They agree, independently and without coordination, on the one thing the category still lacks.
TL;DR. Brian O'Kelley of AdCP and Shailley Singh of IAB Tech Lab were asked the same questions separately and arrived at the same answer: a shared, deterministic record of agentic transactions is a precondition for the category to function as a market, and it is not yet in production at scale. The convergence between two rival bodies on the same unsolved problem is the most useful signal the category has produced this year. It also gives buyers a checklist that holds regardless of which protocol eventually wins adoption.
What do both protocols say about shared records?
Both organisations were asked the same set of questions, in writing, on the record, for an investigation by ADOTAT published in August 2026.
O'Kelley was asked whether a shared, machine-readable semantic layer is a precondition for cross-company agent buying. His entire answer was one word: "Yes."
Singh was longer, and more revealing about why. "A core principle of AAMP is continuity," he said, describing a protocol built on standards the ecosystem already understands rather than a clean-sheet specification. The goal, in his words, is to ensure buyer and seller agents work from "consistent, machine-readable definitions, rather than allowing differences between LLMs to degrade interpretation of the transactional context."
Read that clause carefully. The failure mode Singh is designing against is not fraud. It is not latency. It is two AI models understanding the same words differently, each writing down a different version of the deal, and neither flagging an error, because from inside each system the account is internally consistent.
Why does a model's word not count as a transaction?
Both men were asked directly whether a real agentic transaction requires a deterministic, auditable record rather than a model asserting a buy occurred.
O'Kelley's answer: "Yes."
Singh went further. The architecture requires "structured transaction objects, explicit transaction states, and systems of record that minimise hallucination or misinterpretation of context and that both sides can reference and audit." He added that existing standards like OpenDirect provide a foundation for "defining canonical transaction states rather than relying on natural-language confirmation."
Natural-language confirmation is the polite technical term for an agent telling you in prose that it bought something. Singh's position is that this is not a transaction record. It is a claim. O'Kelley's answer means the same thing in one word.
The research supports the concern. Modelled across 90,202 simulated agent-to-agent transactions, WPP Research found that under separate record-keeping, where each agent maintained its own account, the two records disagreed on at least one term in 95.3% of cases. Under a single shared record both agents wrote to, disagreement fell to 0.19%. The full methodology is available from WPP Research.
In the same simulation, run to ninety days at holding-company scale, 677 million impressions settled with no agreed record of what had been delivered. Neither side flagged an error. The problem is not visible from inside either system. It only shows up when both sides are asked to produce their version of the same deal.
Does it matter which protocol wins?
For buyers and platforms making platform commitments now, the horse-race framing is less useful than the convergence finding.
Both protocols agree on the precondition. Whatever standard the ecosystem eventually adopts, it will need to produce structured transaction objects, explicit states, and a deterministic audit trail that both buyer and seller agents can query and receive the same answer. The specific protocol architecture is a secondary question. The primary question is whether a given vendor's implementation produces those outputs.
That checklist does not depend on which specification gets adopted. A platform that claims to be AdCP-compliant, or AAMP-aligned, or protocol-ready should be able to demonstrate the outputs both standards bodies have named. If the answer is a demo that narrates buys in confident prose but has no structured output behind it, that is a protocol claim without the substance either O'Kelley or Singh says is required.
What does the current state of each protocol tell you?
O'Kelley described the current state of AdCP plainly. Asked whether the binding signed record of an agentic transaction exists and works in production today, his answer was: "Exist yes. Work yes. Active in production at scale - not yet."
Singh's timeline runs to releases rather than a calendar date.
Neither answer is discouraging given where the category is. Both are honest. The two men are building competing products and have every commercial reason to describe their progress generously. That both chose not to is the finding that matters. The shared unsolved problem, named independently from opposite sides of a competitive standards debate, is a more reliable signal than any single organisation's roadmap.
The category will get there. The protocols are being built, the standards bodies are doing the work, and the operators running real volume are honest about current scale. What the convergence on shared records tells buyers is that the question to keep asking vendors is not which protocol they back. It is whether the record exists and whether both sides can read it.