Agentic AI Could Lock Publishers Out Before the Auction

By Elena Vasquez

54% of 142 brand and agency professionals said in January that their companies do not use agentic AI in their workflows at all, according to Digiday research. That is the buy side reporting on itself.

The agentic skeptics have the better argument for now. Anthony Katsur, who runs IAB Tech Lab, expects “several false starts” and years of experimentation and alignment across platforms, agencies and publishers. Razorfish called the mood at CES in January inevitable but incremental.

After 20 years watching this industry call a shift imminent about once a year, I remain skeptical. Mobile was called inevitable around 2007, but it did not pass desktop in US digital ad spend until 2015. Cookie deprecation was announced in January 2020 and abandoned in July 2024.

Anyone telling publishers they are already behind on agentic AI is selling something. Earlier shifts required someone to build and pay for new integrations, while the parties funding them disagreed about what to build. That left sunk integration cost as the variable. With agentic AI, buyers have already paid that bill.

Fully autonomous media buying may still be years away. AI agents can already evaluate inventory, inform bidding decisions and optimize spend through existing programmatic infrastructure, using integrations the buy side built and paid for. The infrastructure exists. The logic is changing.

Where is the money already moving?

On April 8, 2026, Microsoft ended a relationship with Dentsu that ran for more than a decade and moved its global media planning and buying to Publicis. The trade press puts the mandate between $700 million and $1.2 billion.

The move came inside a larger plan to embed agentic AI across the marketing workflow: agents on Microsoft Fabric, informed by Epsilon, decide which segments matter and deploy media against them.

Asked why Publicis keeps winning without pitches, Arthur Sadoun said clients now understand: “it’s about identity.”

Publicis also agreed in May 2026 to acquire LiveRamp, a deal the FTC cleared in October subject to a five-year neutrality order. Once it closes, one of the two largest buyers of publisher inventory will own more of the identity layer its own agents reason over.

Those agents buy media. Media clears in auctions. Publishers get paid in those auctions.

Money and identity data are moving now, not after some autonomous future.

Why could publishers miss the auction?

The industry’s experience with bid shading shows how new buying logic can reshape auctions. Exchanges began moving to first-price auctions in 2017. Google announced Ad Manager’s move in March 2019 and completed it four months later.

Bid shading, which avoids overpaying in first-price auctions, spread across supply-side platforms and then demand-side platforms within months. It used existing integrations, with no new infrastructure or industrywide standards.

Publishers learned about its effects through clearing-price reports and adjusted their pricing floors.

Call bid shading machine learning rather than agentic if you prefer. Shading optimized the price of a bid already being made. Identity-informed agents decide what inventory is worth a bid at all.

The exposure is not whether agents pay a premium for inventory; it is whether that inventory survives evaluation. Fail that test, and a publisher will not be outbid in some future auction. It will not be in the running at all.

The danger comes before price: inventory judged unworthy may never enter the running.

How autonomous is agentic buying today?

Among 360 digital video buyers surveyed in February and March, 96% see a role for agentic AI in programmatic and real-time bidding, while 63% expect that role to improve bidding as an optimization layer.

IAB research says current adoption centers on decision support rather than purchase decisions. The rate for AI-based deal execution and negotiation in video campaigns sits at 28%.

Forty percent of buyers require human approval before an agent acts. Agents run freely where no money is committed and get gated where money is involved.

Katsur’s false starts describe the execution half of the market, and he has a point. The decision-making half is already gaining traction.

Among video buyers spending $50 million or more a year, 63% are using or planning agents for inventory discovery and evaluation, compared with 30% of smaller buyers. The largest budgets are already pointing agents toward deciding which inventory deserves consideration.

Decision support is scaling now; autonomous spending remains behind a human gate.

What can publishers do now?

Publishers do not need to wait for agentic standards to mature. Most preparation can improve inventory evaluation and match rates in today’s auctions, regardless of how quickly autonomous buying develops.

• Create stable identifiers across your properties.
• Structure first-party attributes so bid requests carry clear context.
• Authorize data access explicitly and get the terms in writing. Bid-request signals can be visible even to bidders that do not win, making authorization a contract question rather than a technical one.

This is the same publisher work as before, and it pays now. Clear context helps an agent value inventory, while stable identifiers and explicit permissions make the transaction legible and enforceable.

In 2019, publishers that understood their own floors before auctions changed did well. Those that learned about the shift from revenue reports did not.

With agentic AI, waiting for fully autonomous transactions could mean missing the more consequential shift already underway. The practical work starts now, without waiting for an agent standard.