Agentic buying could screen publisher inventory before auctions
For publishers preparing for agentic media buying, the most immediate concern may be whether their inventory is considered for a bid at all—not whether an automated buyer will pay more once an auction begins. That is the argument made by Jonathon Shaevitz, who says buyer-side systems can increasingly evaluate inventory through existing programmatic infrastructure, even as fully autonomous transactions remain years away.
Shaevitz’s analysis distinguishes between agents that inform decisions and agents that execute purchases. The first category is advancing, he argues, while execution remains more constrained by human approval and the need for industry alignment. For publishers, that means attention to how buying systems assess inventory may be warranted before autonomous purchasing becomes commonplace.
Decision support is ahead of autonomous execution
Industry expectations remain mixed. IAB Tech Lab CEO Anthony Katsur has predicted “several false starts” and years of experimentation and alignment among platforms, agencies and publishers. Razorfish described agentic AI as inevitable but incremental at CES in January. In research conducted in January, Digiday found that 54% of 142 brand and agency professionals said their companies did not use agentic AI in their workflows.
Those indicators reflect the limits of current adoption, but Shaevitz argues that they do not capture the full shift. Buyers can use agents to evaluate inventory, inform bidding and optimize spend without waiting for new pipes or industry-wide standards, because those systems can work through existing programmatic integrations.
IAB research on digital video buyers points to the same divide. Among 360 buyers surveyed in February and March, 96% saw a role for agentic AI in programmatic and real-time bidding, while 63% expected it to serve as an optimization layer that improves bidding. The research characterized current use as decision support more than autonomous purchasing. AI-based deal execution and negotiation for video campaigns stood at 28%, and 40% of buyers said human approval was required before an agent acted.
Adoption also varies by budget. Among video buyers spending at least $50 million annually, 63% were using or planning to use agents for inventory discovery and evaluation, compared with 30% of smaller buyers. Shaevitz sees that use case as consequential: an agent may filter inventory before it enters an auction, rather than simply determine the price of a bid.
Earlier auction changes offer a precedent
Shaevitz points to the industry’s transition to first-price auctions and the spread of bid shading as a precedent for how buying logic can change without a wholesale rebuild of infrastructure. Exchanges began moving to first-price auctions in 2017. Google announced that Ad Manager would move to first-price auctions in March 2019 and completed the change four months later.
Bid shading—an approach to reducing the risk of overpaying in a first-price auction—then spread across supply-side platforms and demand-side platforms through existing integrations, according to Shaevitz. Publishers had to assess its effects through clearing-price reports and adjust pricing floors. The comparison has limits: bid shading optimized the price of a bid already being made, while inventory-evaluation agents may affect whether a bid is made in the first place.
Recent corporate moves also illustrate the scale of investment around identity and automated marketing. On April 8, 2026, Microsoft moved global media planning and buying from Dentsu, its partner for more than a decade, to Publicis. Trade press valued the mandate at between $700 million and $1.2 billion. The appointment accompanied an announcement about embedding agentic AI across marketing workflows, including agents on Microsoft Fabric informed by Epsilon.
Publicis agreed in May 2026 to acquire LiveRamp. The Federal Trade Commission cleared the deal in October, subject to a five-year neutrality order. The acquisition had not yet closed in the account of the deal cited by Shaevitz. He argues that identity-linked systems could influence how agents assess inventory; the corporate developments do not, by themselves, establish how those systems will make individual buying decisions.
Publisher preparation starts with data foundations
Shaevitz recommends that publishers take practical steps without waiting for agentic standards to settle: maintain stable identifiers across properties, structure first-party attributes so bid requests convey clear context, and explicitly authorize data access with terms recorded in writing. He notes that bid-request signals may be visible to bidders that do not win, making data use a contractual concern as well as a technical one.
These measures can support inventory evaluation and match rates in current auctions, regardless of how quickly autonomous buying develops. The broader implication is not that publishers should assume agents already control transactions. Rather, as buyer-side tools gain a role in discovery and evaluation, publishers may need to make their inventory’s identity, attributes and data permissions clear before an auction begins.
