Agentic Ad Buying Not At Scale, Standards Bodies Admit

Agentic Ad Buying: The Builders Admit It Isn't Running Yet

Two rival standards bodies just said the quiet part out loud, and the number that comes with it is brutal.

Brian O'Kelley, who runs Scope3's Ad Context Protocol (AdCP, a proposed standard that lets AI agents buy and sell ads autonomously), confirmed in writing that AdCP exists and works but is "not yet" running at scale in production. Shailley Singh at IAB Tech Lab, which backs the competing AAMP standard, agreed on the core technical point: an AI agent claiming a buy happened is not the same as a buy happening. You need a deterministic, auditable record of what was proposed, approved, and executed. Neither standard produces that today.

Andrew Mole, CEO of pubX (a publisher revenue optimization platform), put a number on the live reality: $2,000 to $3,000 a day in real agentic media spend. A mid-market brand burns more than that on search before lunch.

The valuation problem

The Trade Desk (the largest independent demand-side platform) trades partly on being the platform of the agentic future. If the two organizations building the rails are both on record saying "not at scale," that future is a 2027 story at the earliest. Anyone paying a premium for it today is paying for a demo.

AppLovin looks better by contrast every time an open-standards process admits it isn't production-ready. AppLovin never joined the open-standards effort. It runs a closed, proprietary machine-learning loop and reports the revenue.

The relabeling risk Mole raises is the real threat to DoubleVerify and IAS (the two dominant ad verification vendors). If "agentic" is just programmatic with a chatbot layered on top, existing verification tools already cover it, and there is no new premium category to sell.

For operators, the question is when to commit

Integrating to AdCP or AAMP in production this year is a coin flip on the wrong protocol, for volume that wouldn't cover the salary of the engineer on the project. The real cost is opportunity cost: every sprint on speculative agentic plumbing is a sprint not spent on curation, clean rooms, or CTV supply paths that move money this year. Fund it like a research line item.

The one date worth watching is Q1 2027, when locked-in annual programmatic commitments come up for renewal and budget could, in theory, get re-pointed at agentic pipes. Mole flagged that date himself, then called his own optimism generous.

The practical move: run a cheap, read-only integration to whichever protocol your top DSP partner leans toward. Learn the schema without betting on it. And press any vendor selling you "agentic" today for the deterministic audit trail Singh described. If they cannot produce the record of what was proposed, approved, and executed, they are selling you programmatic with a chatbot on top.

Our call: Through Q1 2027 upfront and renewal season, no top-five DSP or agency holding company will report agentic ad spend material enough to break out as its own figure. The builders themselves say it isn't at scale. Real transaction volume won't materialize while two rival standards fragment the buy side and nobody has the audit trail finance and brand safety teams will demand. Inertia, in this industry, is undefeated.

The more interesting version of being wrong: Amazon's DSP defines the de facto standard by sheer volume, and the AdCP-versus-AAMP race becomes irrelevant because everyone integrates to Amazon. Watch Amazon.

Two rival standards bodies just said the quiet part out loud. Brian O'Kelley, who runs the Ad Context Protocol (a proposed standard that lets AI agents buy and sell ads on their own), confirmed in writing that his protocol exists and works but is "not yet" running at scale in production. Shailley Singh at IAB Tech Lab, which backs a competing standard called AAMP, agreed on the core technical point: an AI agent saying a buy happened is not the same as a buy happening. You need a deterministic, auditable record of what was proposed, approved, and executed. Andrew Mole, CEO of pubX, put a number on the live reality: $2,000 to $3,000 a day in real agentic media spend. That's a rounding error.

What's being decided: Nobody in this story is deciding anything for you. The question for an operator is where to place your integration bet, how much of your "agentic" roadmap is real, and when the transition actually starts. This is a Type 2 call for now (easy to reverse, because there's almost nothing to commit to yet) that quietly becomes Type 1 the day you write production code to one protocol's schema. Forcing function: Mole's Q1 2027, when annual programmatic spend commitments expire and could, in theory, get re-pointed at agentic pipes.

---

The Market Analyst

The public-market read is that a premium is now exposed. The Trade Desk trades partly on being the platform of the agentic future. If the two people building the rails are both on record saying "not at scale," that future just slipped to a 2027 story at the earliest, and anyone paying up today for it is paying for a demo. In plain terms: investors have been valuing a thing that isn't running yet. AppLovin looks smarter by contrast every time an open-standards process admits it isn't production-ready, because AppLovin never joined the open-standards party. It runs a closed, proprietary machine-learning loop and reports the revenue. The relabeling charge Mole raises is the real threat to DoubleVerify and IAS. If "agentic" is programmatic with a chatbot on top, the existing verification tools already cover it, and there's no new premium to sell.

The Skeptic

Programmatic has been agentic since 2009. A machine decides, a machine bids, a machine buys, no human signs off on the individual impression. Calling OpenRTB "non-agentic" is a marketing choice, not a technical one. What's actually on offer is a new API layer on top of infrastructure that already works, wrapped in a buzzword. For this to matter, you have to believe AI agents will develop buying logic that's meaningfully different from algorithmic bidding, and that the difference shows up as measurable lift. No evidence for either today. In plain English: someone built a fancier front door for a house that already had one. The most useful line in the whole story is Mole calling his own 2027 forecast "optimistic."

The Operator

The $3K/day number is brutal, and it's the only real benchmark on the table. That's what a mid-market brand burns on search before lunch. No trading desk should build new workflows for that volume, and the two competing standards mean any integration you write today is a coin flip on the wrong protocol. What breaks first is internal credibility. Some VP already promised the CMO an "agentic buying pilot," and now needs a demo that looks live without being live. That is exactly the relabeling O'Kelley's own community is calling out. In plain terms: the pressure to look ahead of this is going to produce a lot of theater. At 90 days, brand safety and finance start asking for the deterministic audit trail Singh described, and discover nobody has one.

The CFO

Where's the payback? There isn't one to model yet, and that's the point. Spending engineering hours integrating to AdCP or AAMP today buys you optionality on a protocol that may lose, for volume that wouldn't cover the salary of the engineer on the project. The real cost is opportunity cost: every sprint on speculative agentic plumbing is a sprint not spent on curation, clean rooms, or CTV supply paths that move money this year. In plain English: this is a research line item, not a revenue line item, and it should be funded like one. The one date worth watching on the finance side is Q1 2027, because that's when locked-in programmatic commitments come up for renewal and budget could actually move.

---

Where they disagree

The Skeptic and the Market Analyst split on whether any of this is new. The Skeptic says it's OpenRTB with a chatbot and no proven lift, so the whole category is a solution hunting for a problem. The Analyst says it doesn't matter if it's new, it matters that the market has already paid for it, so the exposure is real either way. Both can be right: the tech can be incremental and the valuation premium can still deflate.

The second fault line is the standards race. The Analyst sees a clean winner-take-most infrastructure prize, like FIX Protocol for stock trading. The Operator sees a coin flip you shouldn't wager on yet. The resolution is who writes to whose schema first, and in programmatic the buy side has always driven the standard.

What this actually hinges on

Two beliefs. First, whether AI-driven buying produces lift that algorithmic bidding doesn't. No evidence today, and until there is, "agentic" is a label. Second, whether Q1 2027 renewals actually re-point spend into new pipes instead of rolling into legacy programmatic by default. That requires the buy side to hold firm through a nervous market, and inertia is undefeated in this industry.

The council leans skeptical on the near term and agnostic on the standard. Don't build to AdCP or AAMP in production this year. Do run a cheap read-only integration to whichever your top DSP partner leans toward, so you learn the schema without betting on it. And press any vendor selling you "agentic" today for the deterministic audit trail. If they can't produce the record of proposed, approved, executed, they're selling you programmatic with a chatbot.

---

Prediction: Through Q1 2027 upfront and renewal season, no top-five DSP or agency holdco will report agentic ad spend material enough to disclose as its own figure, and the "agentic" volume in market will remain a rounding error against total programmatic.

Confidence: Medium. The builders themselves say it isn't at scale, but a walled garden could force volume faster than expected.

Why: The two organizations building the rails, Scope3's AdCP and IAB Tech Lab's AAMP, both confirmed in writing the infrastructure isn't running at scale, and the only operator willing to give a number, pubX's Andrew Mole, put live volume at $2,000 to $3,000 a day and called his own Q1 2027 optimism generous. Real transaction volume doesn't materialize while two rival standards fragment the buy side and nobody has the deterministic audit trail finance and brand-safety teams will demand. The opposite outcome, a genuine agentic spend wave by early 2027, requires the buy side to abandon locked-in programmatic commitments for unproven pipes during a jittery market, and inertia plus a standards war argue against it.

Revisit by 2027-04-15: We're right if, after Q1 2027 renewals, agentic spend is still described in "pilot" and per-day terms and no major buyer breaks it out as real volume. We're wrong if a Tier-1 DSP or holdco reports agentic media spend at a scale that moves its numbers, or a walled garden's DSP drives standardized agentic volume into the billions.

The more interesting version of being wrong: Amazon's DSP defines the de facto standard by sheer volume and everyone integrates to it, which would settle the AdCP-versus-AAMP race by making it irrelevant. Watch Amazon. The standards bodies will follow.