The story. Is agentic buying actually changing which inventory gets bought, or just how fast the bid gets set? Brand-side adoption of AI-agent campaign management — Google’s Performance Max and AI Max, Meta’s Advantage+, and holding-company copilots like Publicis CoreAI and WPP Open — has crossed from pilot to default setting for a growing share of mid-market advertisers this year, and the honest answer from the buy side is: both, and publishers are better positioned to benefit from the second than the first.
01What happened
- Budget share, not headline adoption, is the real signal. Multiple agency holding companies now report that a majority of new campaign launches for mid-market clients default into an AI-managed bidding and creative-rotation mode rather than manual setup — a shift from “available option” to “default path” inside the last two buying cycles.
- The agents optimize inside a black box the advertiser doesn’t fully see. Performance Max and Advantage+ both pool inventory decisions — search, display, video, in-app — behind a single reported outcome, and neither platform discloses the per-channel allocation logic; agencies running both report inconsistent channel mixes for similar campaigns with no visible explanation.
- IAB Europe’s 2026 AI report (covered elsewhere in this issue) found the same governance gap from the institutional side: 58% of buyers expect agentic buying at scale within a year, but oversight and audit tooling for what an agent actually decided lags well behind adoption.
- Agencies are building their own guardrails faster than platforms are. WPP Open and Publicis CoreAI both now ship “explainability” layers that log an agent’s channel and creative decisions independently of the ad platform’s own reporting — an agency-side response to clients asking why an AI-managed budget moved.
02Why marketers should care
- An agent optimizing to a platform’s own reported outcome will always favor that platform’s own inventory when the choice is close. Performance Max and Advantage+ report a blended result; neither is incentivized to show a marketer the counterfactual case for spending outside its own ecosystem.
- “Agentic” doesn’t mean “audited.” The explainability tooling agencies are building exists precisely because the platforms’ own agents don’t produce an audit trail a marketer can independently verify — until that changes, a channel-mix shift attributed to “AI optimization” is a claim, not a receipt.
- The inventory an agent chooses is shaped by what it can measure, not what performs best in a vacuum. A publisher whose signals are hard for an agent to read — thin first-party data, inconsistent conversion tagging — will get starved of budget by the agent regardless of the inventory’s actual quality.
03What marketers should do
04The bottom line
Agentic buying is real and accelerating faster than the governance tooling meant to check it — which means the platforms setting the defaults have more influence over where budget lands than either the agencies or the publishers competing for it fully realize yet. The marketers moving fastest here aren’t the ones handing the most control to an agent; they’re the ones building an independent way to see what the agent actually did.
05Offer tie-in
Publishers whose own signal quality determines whether an agentic buyer can credit their inventory correctly are exactly who APH’s Cortex yield program works with directly — clean, well-tagged first-party signal is table stakes for winning an agent’s allocation call, not just a nice-to-have.