The Signal · No. 007 · Biweekly · Filed May 13, 2026
I have been reading agentic buying as a tooling story. It is a structural one. The bid logic is leaving the buyer’s seat, and the dashboard is no longer the screen where the decision lives.
I. The Admission
I have been reading agentic buying as a productivity story.
For about a year I have read every announcement — the IAB survey, the PubMatic pilot disclosures, the trade-press explainers — as a story about the buyer’s tooling getting better. The brief that used to take a planner three days now takes the agent ninety seconds. The execution that used to require a trafficker can be assembled from a prompt. Faster cycles, less manual labor, a smarter brief-to-campaign loop. A productivity story.
That reading was wrong, and I missed what was actually moving.
Agentic buying is reshaping where the bid logic sits. The agent is no longer standing next to the planner. The agent is between the planner and the bid request, and the decision is composed in a place the buyer’s team no longer owns.
I was watching the wrong screen.
II. The Reframe
There are two kinds of bid logic.
There is the kind the buyer can witness. Every audience selection, every inventory choice, every floor adjustment is visible inside an interface the buyer’s team manages and the buyer’s auditors can read. Call it witnessed bid logic. And there is the kind the buyer delegates to an external agent whose decisions are summarized in outcome reports but never exposed as a stepwise log the buyer’s team owns. Call it delegated bid logic.
A year ago, almost every CTV programmatic campaign run by a holding company sat closer to the witnessed end of that range. A meaningful share now sits at the delegated end.
The dashboard the buyer watches still works. The dashboard is no longer where the decision is.
III. The Application
Yesterday, May 12, 2026, Cocie AI launched a product called CTV Autopilot. The launch description is precise about what the product replaces. A single buyer brief converts into a live, show-level targeted CTV campaign within minutes through a multi-agent system covering audience intelligence, inventory selection, and campaign activation. The audience, the inventory, and the activation are decided by the agent. The buyer’s brief is the input. The campaign result is the output. The reasoning in between is the part the buyer no longer composes.
That is a new place where the decision lives. The faster-planner reading misses it.
PubMatic has disclosed that its AgenticOS platform has run more than thirty fully autonomous, end-to-end agentic campaigns in addition to a thousand directly placed publisher deals. The framing matters. In addition to. The autonomous campaigns are a separate category, not a tier of the manual workflow. The buyer that ran them did not press a button thirty times. The buyer initiated a brief and the agent ran the campaign.
NBCUniversal and FreeWheel completed the first AI-agent-led programmatic guaranteed deal in the first quarter of 2026, with the first major campaign running during NFL playoff games. Programmatic guaranteed is the category traditionally managed by hand: the bookings, the inventory holds, the pod-position calls. In the NBCU pilot, those decisions were held by an agent.
The IAB ran a survey between November 2025 and January 2026. Sixty-six percent of US ad buyers said they were planning to focus more on agentic ad buying in 2026. That is a delegation number, surfaced as a tooling number. The buyers reporting the focus shift are not asking for better dashboards. They are committing to stop composing the decision themselves.
What is moving in each of these cases is not the quality of the buy. The quality, by most reports, is comparable to the manual version and sometimes better. What is moving is who and what holds the bid logic. The seat that used to host the decision is being vacated. The decision is migrating downstream of the brief and upstream of the bid request, into the agent’s reasoning, where the buyer no longer composes it line by line.
IV. The Reversal
The admission costs me something, and I want to name what it costs.
For a small or mid-market advertiser that has never had a programmatic CTV planner on staff, the agent is not converting witnessed bid logic into delegated bid logic. There was no witnessed bid logic to begin with. The brief-to-campaign loop is the first version of bid logic that buyer has ever had. Cocie AI and PubMatic AgenticOS are giving that buyer a campaign that previously did not exist, run with a decision structure that previously did not exist for them. The delegation framing does not apply, because there was no audit chain to delegate from.
So the post’s anxiety is partly a holding-company anxiety. The seat that is being vacated is the seat the agencies built. The buyer who never had the seat is gaining capability. That has to make room inside the frame, and it does.
There is a smaller admission inside the first one. I have used the word witnessed this whole piece as if it described a current condition at every holding-company desk. It described a recent one. The audit chain at most large agency programmatic teams has been partial for years: the DSP reasoning is a black box on the bid side, the curation choices are visible only to the desk that made them, the floors are negotiated in rooms the auditor was never in. The movement from witnessed to delegated is not a fall from full visibility into no visibility. It is a further reduction in a chain that was already shorter than the dashboard suggested. The post owes the reader that calibration, and it is owed to me too.
What survives the reversal is the holding-company case. A buyer with the audit chain they still have is, this year, choosing whether to convert what is left of that visibility into delegation. The choice has not been named in the buyer’s own internal communications as the choice it is. The trade press is calling it a tooling decision. It is a seat decision.
V. The Practice
The practice change is small and specific.
I am going to stop asking the question I have been asking. The question I have been asking is: did the agent do the buy well? That question is about outcomes. The agent will, in most cases now, return a result that looks good. The result is downstream of the decision. The result is not the part that moved.
The question I am going to start asking is: what does the agent’s audit log look like, and who owns it? If the agent returns a campaign result but not a stepwise log of the decisions made between the brief and the bid request, the buyer has delegated bid logic and should know they have. If the log exists but is owned by the agent’s vendor and not the buyer’s team, the buyer has delegated bid logic and given up the audit chain in the same step.
Witnessed and delegated are different positions inside the buy. A buyer that knows which position they are running is in better shape than a buyer who watches the dashboard and assumes the screen they are watching is still the one where the decisions live.
I have been watching the wrong screen. I am going to start watching the audit log.
Filed from inside the auction. The Signal ✦ By Albert · ElementalTV
Leave a Reply