Blog · Agentic

Agentic buyers can drop your inventory before OpenRTB fires, and fill rate will not show that filter.

November 9, 2026 · RTBlint

On November 9, 2026, Jonathon Shaevitz of GrowthCode argued in AdExchanger's Sell Sider that publishers should worry less about agentic bid shading and more about whether their inventory survives evaluation before a bid request is ever sent. The column is opinion, but it rests on numbers from IAB's February and March 2026 digital video buyer survey: 96 percent of buyers see a role for agentic AI in programmatic, while only 28 percent are using agents for deal execution and negotiation on video. The gap is the story on the wire. One half of agentic adoption decides what is worth bidding on; the other half, still human-gated, decides how much to pay once the auction starts.

Fill rate cannot see a filter that happens upstream

Publisher revenue tools count impressions offered, bids received, and clearing prices. They were built for an era when the main mystery was why demand bid low or timed out inside the exchange. Shaevitz draws a sharper line from the first-price and bid-shading transition: shading changed how much winners paid, not whether they entered. Agents trained on identity and segment fit can refuse to call the SSP for inventory that fails a discovery step.

That refusal is invisible in the same CSV export that made bid shading legible years later. You will not get a BidResponse.nbr code when no request was composed. You will not get a loss reason from a bidder that never received your imp object. The symptom looks like soft demand or a bad floor, not a structured pass on context fields your integration never populated.

The CTV capacity gap post made the same diagnostic mistake on a different axis: treating unfilled pods as one bucket when demand may have dropped the request for parse reasons. Agentic evaluation adds a third bucket above parsing: semantic rejection before the bidstream handshake.

What IAB measured in the discovery versus execution split

IAB's Digital Video Ad Spend and Strategy Report, Part One, surveyed 360 digital video buyers. Among respondents, 96 percent see a role for agentic AI in programmatic and RTB, and 63 percent expect that role to be an optimization layer on bidding. Current adoption, in IAB's wording, centers on decision support rather than autonomous purchase decisions. AI-based deal execution and negotiation for video campaigns sits at 28 percent, and 40 percent of buyers require human approval before an agent acts.

The spend split matters for who you are negotiating with on Monday. Among buyers spending $50 million or more per year on digital video, 63 percent are using or planning to use agents for inventory discovery and evaluation, compared with 30 percent of smaller buyers. The largest budgets are pointing automation at the question Shaevitz highlights: is this impression worth sending to a bidder at all?

Anthony Katsur's public comments about false starts on agentic standards describe the execution half, where money commits and integrations break. Shaevitz agrees that fully autonomous buying may be years away. He also notes that buyers already paid for programmatic pipes. Agents do not need a new protocol to read your existing OpenRTB fields and decide whether to participate.

Bid shading is the wrong mental model for the new failure

Google Ad Manager's move to first-price auctions in 2019 did not wait for publishers to adopt a new spec. Bid shading spread across SSPs and DSPs through integrations that already existed. Publishers who learned about shading only from clearing-price dashboards adjusted floors after revenue moved. Those who understood the mechanism earlier treated shading as a buyer strategy on bids they were already making.

Shaevitz asks publishers to apply that lesson forward. The risk is not whether an agent will shade your winning price. The risk is whether your inventory is in the candidate set when the agent runs. Microsoft's April 2026 media appointment to Publicis, and the accompanying agentic marketing workflow announcement, is macro context: large buyers are wiring identity-informed agents into planning systems that eventually touch live auctions.

None of that changes the JSON shape on a single impression today. It changes the prior that sits above the DSP adapter. When that prior says skip, your OpenRTB payload never gets a structural validation error. It simply does not run.

What the bid request must carry for evaluation, not just auction

Shaevitz lists three publisher steps that do not depend on which agentic standard wins: stable identifiers across properties, first-party attributes structured so bid requests carry clear context, and explicit authorization for data access because signals can be visible even to losing bidders. Those are OpenRTB hygiene statements, not AAMP checklist items.

On the wire, evaluation agents still read the same objects your exchange already forwards:site.domain and app.bundle for identity, content and AdCOM fields for placement semantics, user.eids and permitted seller-defined audiences where contracts allow, and regs.gpp when consent gates personalization. Missing context is not a syntax error. It is a reason to never call the bidder.

{
  "id": "eval-pass-01",
  "imp": [{
    "id": "1",
    "video": {
      "mimes": ["video/mp4"],
      "minduration": 15,
      "maxduration": 30,
      "plcmt": 1
    }
  }],
  "site": {
    "domain": "news.example",
    "cat": ["IAB12"],
    "content": {
      "cat": ["IAB12-3"],
      "language": "en"
    }
  },
  "regs": { "gpp": "DBABLA~...", "gpp_sid": [7] }
}

The sample is valid enough to enter many exchanges. An evaluation agent asked to match finance vertical inventory may still drop it if your production traffic omits content.cat, ships a stale bundle, or leaves audience extensions empty while your sales deck promises curated segments. RTBlint flags inconsistent or malformed payloads. It does not simulate a buyer's discovery model. The point is to stop blaming auction mechanics for a pass that happened earlier.

The agentic version negotiation gap post covered another upstream problem: agents and exchanges agreeing on OpenRTB snapshots out of band. Evaluation failure can happen even when versions match, if the semantic content of the request does not support the buyer's brief.

What to do this week

Export a week of traffic where fill softened and separate impressions that received zero bid requests at the exchange from those that received bids but lost. If your SSP only shows one column, ask for adapter logs on the buy-side path for a named deal or PMP.

For your top open-market and PMP lines, capture one production request per surface and verify content, identity, and consent fields against what your data sales team promises in one-pagers. Mismatches belong in commercial review, not only in engineering backlogs.

When product teams ask whether you are behind on agentic standards, answer with field readiness on the requests you already send. Standards will keep shipping false starts on execution. Discovery agents do not need to wait for them.

RTBlint is independent and not affiliated with GrowthCode, IAB, AdExchanger, or Microsoft. Paste production JSON into the bid request tester when you need a structural second opinion before you renegotiate floors for demand that may never have seen the auction.