AdCP Agentic Ad Protocol Not Yet in Production at Scale

Full analysis

Brian O'Kelley, who wrote the AdCP agentic advertising protocol and co-founded AppNexus, went on the record with a three-word verdict that undercuts a lot of sales decks: the signed record "exists yes, works yes, active in production at scale: not yet." ADOTAT ran that quote as evidence that vendors pitching "agentic" as live are selling a roadmap with a demo attached.

What's actually being decided: whether ad-tech buyers should write agentic-protocol requirements into RFPs and roadmaps now, or treat "agentic" claims as sandbox-grade until someone shows production volume.

Reversibility: Type 2. Nobody is signing a decade-long contract off this. A buyer can pause an AdCP-dependent activation and lose nothing but a quarter. The cost of waiting is low; the cost of building your Q4 plan on a demo is higher.

Forcing function: Q3 and Q4 procurement conversations, where "agentic" is already in the pitch and buyers have to decide what it's worth.

The Skeptic

O'Kelley's "not yet at scale" quote is being used by ADOTAT to manufacture a crisis out of a draft spec, and that is a coverage choice. Every real protocol in this industry lived here for a year or two. OpenRTB was a draft before it was the plumbing under the whole open market. The question that matters isn't whether AdCP is in production today. It's whether AI agents will autonomously move meaningful budget inside two years. Today's advertiser "agents" are mostly bid-strategy wrappers with a chat box. If a human still authorizes the spend, the signed-record gap is a footnote. In plain terms: a new standard being early is normal, not scandalous.

The Operator

Here's what breaks Tuesday morning. Your DSP and SSP account teams are already demoing "agentic" to buyers, and their own engineering orgs can't support it at bid-stream volume. A buyer who writes AdCP compatibility into an RFP this quarter finds out at launch that the integration is a sandbox. The break point is trafficking and reconciliation. No production-scale signed record means no auditable chain of custody, which means your brand safety and measurement teams can't close the loop on where money went. In plain terms: the machine can't yet prove what it bought, so pause any AdCP-dependent activation until you see third-party volume logs instead of a slide.

The Strategist

The gap between spec and scale is useful signal: this market is pre-chasm. The interesting move is what happens when someone operationalizes the signed-record layer first. Whoever owns the trust primitive for machine-to-machine ad buying gates every AI agent's ability to spend. That's a moat, potentially a bigger one than cookie-era identity, because it sits under every transaction rather than beside it. O'Kelley's authorship gives Scope3 credibility. Credibility isn't the winner's medal. A scaled SSP or a DSP with existing identity infrastructure could absorb the spec and out-execute on the plumbing. In plain terms: the person who wrote the rules of the road rarely ends up owning the road.

The CFO

Strip the narrative and look at what you're paying for. An "agentic" line item today buys you a spec, a sandbox, and a promise. That's a roadmap priced as a product. The real cost is the engineering hours your team spends integrating against a standard that isn't moving real money yet, plus the credibility you spend telling your own CMO the capability is live when it isn't. The payback on waiting two quarters is nearly free, because early-stage protocols get cheaper and more stable to integrate the longer you wait. In plain terms: you don't pay premium prices to be someone's first production reference.

Where they part ways

The real disagreement is whether "not in production at scale" is a normal waypoint or a warning. The Skeptic says every protocol lives here and the ADOTAT piece manufactures a crisis out of a draft spec. The Operator says that's fine for the protocol and dangerous for the buyer, because account teams are already selling it as finished. Both can be right at once: healthy standard, oversold pitch.

The second split is about who profits from the gap. The Strategist bets the signed-record layer becomes a moat worth racing for. The CFO says stop financing other people's R&D and let the race resolve before you buy. That tension is the whole decision for an operator: lean in to shape the standard, or wait and integrate it once it's cheap and proven.

What it hinges on

Two beliefs. First, whether autonomous agents will authorize real budget without a human in the loop inside two years. If humans keep signing off, the signed-record problem is academic and there's no rush. Second, whether the trust layer under machine-to-machine buying becomes a durable moat or a commodity that gets absorbed into existing SSP and identity plumbing.

The council leans one way for buyers and another for builders. For buyers: treat "agentic" and "live" as separate claims, demand production volume logs, and pause anything that depends on a signed record nobody can yet audit. For builders and investors: the category valuation ceiling drops every time O'Kelley's quote lands in a diligence memo, and the incumbents who can afford to wait are the ones best positioned to absorb the spec once it actually scales.

What to verify before committing: ask any vendor pitching AdCP for third-party production volume, not a demo. If they can't show money moving at real bid-stream scale with an auditable signed record, it's a roadmap.

Prediction: No agentic ad-buying protocol, AdCP included, will move a documented, third-party-auditable share of open-market programmatic spend at bid-stream scale before the 2027 upfront season (May 2027), and vendor "agentic" pitches will get a visible credibility discount in Q4 2026 procurement.

Confidence: Medium. Protocol timelines are slow and the author already conceded the gap.

Why: The protocol's own author put the production-at-scale question at "not yet," and every comparable ad-tech standard, OpenRTB being the clearest case, took well over a year to go from working spec to real volume. The signed-record layer that gates machine-to-machine trust is exactly the hard part, because it requires auditable chain of custody across parties who don't yet share one, and that doesn't get built in two quarters. The opposite outcome, a sudden jump to auditable scale, would require multiple large DSPs and SSPs to ship and reconcile production integrations against a standard whose author says it isn't there yet, which is not how this industry ships plumbing.

Revisit by 2027-05-15: We're right if no vendor can point to a third-party-verified figure for agentic protocol spend moving real open-market budget at bid-stream volume by the May 2027 upfronts, and buyers are visibly harder on "agentic" claims in RFPs. We're wrong if a scaled DSP or SSP publishes auditable production volume moving through AdCP or a rival agentic protocol before then.

The buyer's move costs nothing. Ask for the volume logs. If the answer is a demo, you have your answer.