AI Surfaces Brands Spark Blog About Contact Us
← All Posts Industry

The Price Was Set Inside a Black Box. Agentic Buying Moves It Out.

Google and Amazon are in different courtrooms for different alleged conduct, and the cases share one root cause. In both, the price was determined inside a mechanism that only the party running the auction could see, and that party kept the difference. At Amazon that party was the seller. At Google it was an intermediary, and the publishers on the other side of the trade were as blind to it as the advertisers were. Behavioral remedies can regulate that mechanism. They cannot make it visible to the people on either end. The fix is to stop setting price inside a per-impression black box for inventory that never needed one, and to make price a term the seller commits to before the buy exists.

What the Google and Amazon cases have in common

Both cases turn on an auction operator telling the parties who depended on it one thing and running it another way.

The Google trial record already gave the industry a vocabulary for this: Reserve Price Optimization, Dynamic Revenue Share, Project Bernanke, Last Look. Each changed how the price was computed, was made by the party that kept the difference, and ran where neither advertiser nor publisher could inspect it. Google sat between seller and buyer; both trusted its math. Amazon sells its own inventory, so only the buyer was locked out. The names change. The structure does not.

Why behavioral remedies cannot fix an auction nobody outside can see

A behavioral remedy tells the auction operator which levers it may not pull. It gives no one outside a window into the machine. Neither Brinkema's order nor a win for the FTC changes the arrangement: the buyer submits a number, the operator's system produces a different number, and the only evidence of what happened in between is the operator's description of its own rules. When the operator is an intermediary, the publisher gets the same report.

That arrangement has now failed twice, under the two most sophisticated auction operators in the market, and a consent decree does not fix it at every other platform. The problem is that the auction, as the default pricing mechanism for every impression, requires the buyer to trust a computation it cannot observe, run by a party paid on the result.

Amazon's defense is an argument for buying outcomes at a stated price

Amazon's public response says advertiser contacts are about bids, budgets, and performance, not auction mechanics. Probably true, and it proves the wrong point for Amazon. If advertisers are optimizing to outcomes anyway, the auction mechanics deliver no value the buyer is using. The auction does one job, setting the price, in a room the buyer cannot enter.

The honest version of Amazon's argument is this: the buyer wants an outcome at an acceptable cost. Then sell the outcome at a stated cost. Put the price in the terms, not in the mechanism. A seller who believes its own defense should have no objection to committing to a price before the buy is created, since by its own account the buyer was never bidding on the mechanics. That is the outcome-driven model: the campaign is directed by the business result the advertiser wants, and the price is a term attached to that result, not the output of a computation the buyer cannot audit.

How an agentic buy sets price differently

In an agentic buy there is no per-impression auction on committed inventory, so there is no second bid to fake and no spread to hide. The sequence is:

The terms pass directly between the two agents, with no intermediary computation for either side to be blind to, which is the part of the Google case no behavioral remedy addresses. It also removes the three places manipulation has lived:

The buyer's agent also holds the terms and reconciles delivery against them, not against the seller's report on its own auction. This is how Adgentek ORCA approaches the buy side: it negotiates, commits, and checks delivery against the commitment across every seller it works with. Inconsistent pricing shows up in the buyer's own records, with no need to see inside the seller's stack.

The Agentic Ad Server is built on the same principle on the sell side. Its decisioning order is declared: direct commitments first, then programmatic, then performance demand. Pricing is expressed in the products and pricing options returned to the buyer, not computed after the fact. A seller running that model has nothing to explain to a regulator, because the buyer saw the terms before agreeing to them.

Where the auction still belongs

RTB is the right tool for long-tail inventory where price discovery is the point and no single buyer would commit in advance. Agentic buying does not replace it, and anyone claiming otherwise has not run an exchange. The mistake was making it the default for every dollar.

Premium CTV, sports, host-read podcast spots, DOOH, and large sponsorships were never auctions in any meaningful sense. A host read is priced on the host and the show. A DOOH flight is priced on the screen, the location, and the window. In each case the price was agreed and the auction was bolted on for delivery, which is exactly where a soft reserve or a revenue-share adjustment can sit unnoticed for years. Retail media's sponsored placements are query-driven and an auction has a real role there. Even so, the auction's terms, including whether and how a reserve applies, belong in the agreement the buyer accepts, not in help content updated after a complaint.

What sellers should do now

Publish your decisioning order and pricing terms in a form a buyer's agent can read before the buy is created. The Google and Amazon cases will bind two companies after years of appeals. Every other seller can wait for a rule or make disclosure a product feature.

Getting started

Adgentek builds both sides of this. The Agentic Ad Server gives sellers decisioning with a declared order and declared pricing. Adgentek ORCA gives agencies a buy-side judgment layer that negotiates committed terms across sellers and reconciles delivery against them. If you are working through how much of your spend or inventory should still clear in an auction, contact us at hello@adgentek.ai.

Ready to get started?

Contact Us
hello@adgentek.ai
NVIDIA Inception Program