The demos are confident. The keynotes are bullish. The press releases arrive weekly. Ask the people building the standards, and a different picture emerges.
Brian O'Kelley built the Ad Context Protocol. Asked, in writing, on the record, whether the binding signed record of an agentic transaction exists and works in production today, his answer was three sentences. The third was the one that mattered: "Active in production at scale - not yet." That sentence, from the architect of AdCP, summarises what every agentic advertising pitch this year has been careful not to say.
TL;DR. Both organisations most invested in building agentic advertising standards say the defining capability -- a binding, shared record of what was agreed -- is not yet running at scale. The only operator willing to give a number is moving roughly $3,000 a day in gross media spend. A founding member of the Agentic Advertising Organization has said on the record that most category announcements describe the old programmatic pipes with a new word painted on them. Those three facts, held together, describe the actual state of the category in August 2026.
What does "not at scale" actually mean?
The scale question has a number attached to it, because one operator was willing to give one.
Andrew Mole is chief executive of pubX, a founding member of the Agentic Advertising Organization. He runs live agent-bought, agent-sold volume. Asked what the daily gross media spend was, he answered with an integer: two to three thousand dollars a day. He said so in writing, on the record, in an investigation published by ADOTAT in August 2026.
For context, that is roughly what a mid-market brand spends on search before lunch. The volume is not continuous; there was a break between campaigns, with the level returning the week the interview ran. It covers a handful of publishers across several sites.
Mole is not embarrassed by the number, and he should not be. He is the only person in this category who answered the scale question with a figure rather than a roadmap. He has also said, on the record, that a number of companies are running the same programmatic infrastructure they already own with a new word painted over it. Asked whether he stood behind that assessment, he wrote: "yes I do stand by that and I think it's harmful."
The harm he is describing is specific: a term that should mark a genuine structural shift is being used to label incremental automation, which makes it harder for buyers to know what they are actually evaluating.
Why do both standards bodies say the main thing is still missing?
The Ad Context Protocol and the IAB Tech Lab's Agentic Advertising Marketplace Protocol are competing standards. They disagree on architecture, on approach, on which body should lead. They agree on one thing.
When 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."
Shailley Singh of IAB Tech Lab, answering separately through his press team for the same ADOTAT investigation, was more specific. He said that an agentic transaction "cannot simply rely on a model saying that a transaction occurred." The architecture needs, in his words, "a deterministic, auditable record of what was proposed, approved, and executed."
Read the implication. Every demo this year in which an agent narrates a media buy in conversational prose is, by the explicit standard of the people writing the protocols, not a transaction. It is a description of one.
Singh also named the failure mode he is designing against: not fraud, not latency, not fill rate. His concern was "differences between LLMs degrading interpretation of the transactional context." Two agents, each speaking their own version of the same words, closing a deal and walking away with different accounts of what was agreed.
Why does an agent's word not count as a receipt?
The structural explanation is straightforward. When one agent acts and a human checks the outcome, one record matters: the human's. When two agents agree and both walk away, each holds its own version of events, and nothing forces those versions to match.
Research published with WPP modelled this problem across 90,202 simulated agent-to-agent transactions. Under separate record-keeping, where each agent maintained its own account of the deal, the two records disagreed on at least one term in 95.3% of cases. Under a single shared record that both agents wrote to, disagreement fell to 0.19%. The methodology and codebase are available from WPP Research.
The gap does not announce itself. 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. Both sides paid or invoiced against their own numbers. Neither system flagged an error, because from inside each system, nothing looked wrong.
What is the right question to ask a vendor?
O'Kelley and Singh, from opposite sides of the standards debate, named the same precondition independently. Singh put it plainly: agentic advertising needs "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."
That is a checklist in plain language. When a vendor shows two agents closing a deal, the question that matters is not how fast they closed. It is where the deal is written down, who can read it, and whether a buyer can prove, six weeks later across four hundred deals, what the agent was actually told to do.
Most vendor answers go quiet at that point. The handshake was always going to be the easy part. The shared record is what the category needs to produce before it earns the present tense.