Programmatic advertising spent two decades converging on one negotiation model: real-time bidding, a single auction happening in milliseconds. That convergence is now splitting in two. As buyer-side and seller-side AI agents start transacting media directly, ad tech is building two parallel rulebooks for how those agents talk to each other, and neither has won yet.
The clearest evidence sits inside Prebid.org, the open-source header-bidding nonprofit that effectively standardized the last programmatic transition. Under new chairman Joel Meyer, the OpenX chief technology officer who has sat on Prebid’s board since 2023, the organization has formed an Agentic Working Group and taken ownership of the seller-agent component of AdCP, one of the emerging protocols for agent-to-agent advertising transactions. At the same time, IAB Tech Lab is developing its own agentic advertising standards. Prebid says it wants to stay interoperable with both rather than pick a side, a neutrality that is itself a signal of how unsettled the space still is.
Why bid-by-bid auctions do not fit agentic buying
Real-time bidding was built for a specific problem: matching one impression to one advertiser through a fast, standardized auction. It is not going away. But it was never designed for the kind of negotiation an AI buying agent and an AI selling agent can now conduct: multi-turn, outcome-driven conversations about audience, price and packaging that look more like a negotiation than an auction.
John Rosendahl, director of product management at identity and data clean room company Optable, and Olha Paramonova, VP of ad tech at Sigma Software, both framed the immediate challenge as a communication problem before it is a business problem: the priority, they said, is standardizing how buyer and seller agents communicate and creating a shared “language” for agent interactions. Without that shared language, every platform’s agent ends up speaking its own dialect, and interoperability breaks down before a single deal closes.
What buyers and sellers actually want from the protocol war
The two sides of the trade are not asking for the same thing, which is part of why a single protocol has not emerged. Ingmar Zach, president of product, data and intelligence at WPP, described the buy side’s core need as better access to structured, high-quality supply signals and open frameworks that allow flexibility, the kind of clean, machine-readable inventory data an agent needs to make a negotiation decision without a human checking its work.
Brian O’Kelley, founder and CEO of Scope3 and the engineer credited with building the first modern programmatic ad exchange, put the sell side’s stake in blunter terms: misalignment between buyers and sellers is still a major inefficiency in programmatic, in his view, and agentic protocols are a chance to fix a structural problem that RTB never solved rather than just automate the old one.
That gap between “give me clean signals” and “fix the underlying misalignment” is why Prebid, IAB Tech Lab and the various agent-protocol efforts are not yet converging on one spec. Each is solving for a slightly different failure of the current system.
Prebid’s own history with this protocol predates the chairman change. In January 2026 the organization launched the Prebid Sales Agent, an open-source tool built on AdCP, the Ad Context Protocol governed by AgenticAdvertising.org, designed to let publishers of any size plug into agentic buying without custom engineering. Mike Racic, Prebid’s president, said the goal was to make “agentic advertising practical for publishers of any size,” so that instead of requiring heavy internal engineering or bespoke integrations, publishers could access agentic capabilities through Prebid and participate in more efficient, automated direct buying workflows.
Adoption is real but still small
None of this is hypothetical anymore. Prebid describes initial agentic spend as being in the low six figures and ramping, moving from that base toward what the organization describes as potentially massive scale as more buyers and sellers plug in. That is a genuine, if early, signal, not a pilot that never left the lab.
The constraint is not demand. It is readiness. Publishers are not structurally set up for this: their systems, Prebid’s own writeup notes, were built for human sales teams, not machine-to-machine communication. Mature publisher tooling, a standardized way to package inventory for agents, and widely adopted workflows are all still missing pieces. That is a familiar shape for anyone who watched header bidding’s own adoption curve, and it is why Prebid, which stewarded that transition, is positioning itself as the connective layer for this one too.
What it means for the marketing leader
For a CMO or head of programmatic, the practical question is not which protocol to bet on today. It is whether your DSP, SSP and identity partners can even tell you which agentic protocols they support, and whether their agents are built to negotiate on the buy side’s terms (clean supply signals, structured data) or the sell side’s (fixing misalignment, protecting yield). A vendor that cannot answer that question yet is not behind, since almost nobody has this fully solved. But a vendor that has not started thinking about it is a warning sign, because publisher-side readiness, not protocol elegance, is what determined how fast the last transaction-model shift moved.
The practical move now is evaluation, not commitment: ask current programmatic partners which agent protocols they are building for, treat “we support AdCP and IAB Tech Lab’s framework” as a stronger answer than a bet on either alone, and budget for the low-six-figure test spend Prebid itself describes as where agentic buying actually sits today, not the headline-scale number the category will eventually reach.
The pattern echoes direct deals becoming native to the ad server and the first live agentic sports ad buy: infrastructure is arriving in pieces, ahead of any single standard, and the winners will be the platforms that can speak whichever protocol the other side of the deal happens to be using.
Source: Prebid.org
