Every major ad tech earnings call this year has mentioned agentic AI. What is harder to find is a straight answer to what these agents actually do once a real campaign is on the line. Over the past three weeks, five separate primary releases, two from a sell side platform, plus a broadcaster, an ad creative startup and Europe’s ad industry association itself, have quietly answered that question with more precision than any keynote has. The picture they draw together is narrower than the pitch decks suggest: agentic AI is live in production today, but only for specific, bounded tasks inside a workflow a human still owns.
The tasks agents are actually doing
Start with what shipped. On September 8, Magnite said it ran its first agentic campaign in EMEA with the French trading desk Amnet, part of Dentsu, on behalf of an automotive manufacturer. Amnet used natural language prompts to direct Magnite’s buyer agent to build and activate a video campaign through Magnite’s ClearLine product. The result, according to Magnite, was a campaign setup time cut by roughly 70 percent and a video view through rate of 95. Edouard Schmidt, Magnite’s commercial director for France, put the limiting factor plainly: “AI is only as valuable as the inventory and data it can access.”
That is a narrow, mechanical claim: an agent assembled a media plan faster than a human trafficker would have, against inventory Magnite already controlled. It is not a claim that the agent decided strategy, set the budget or chose the automotive brand’s target audience. Barbara Thuillier-Romeri, Amnet France’s ad tech manager, framed the test the same way, as an effort “to understand how agentic technology could deliver practical value” inside an existing workflow rather than replace it.
The pattern repeats across the supply side
FOX Corporation’s press release announcing the agentic build out of FOX AdStudio, first published June 17 and still the clearest public architecture document for a broadcaster’s agent stack, splits the work into three narrow lanes rather than one general purpose buying agent: audience planning and forecasting (built with WPP and Horizon Media), digital transaction and briefing workflows (with Comcast’s Universal Ads), and linear spot buying automation (with Simulmedia). “With the introduction of agentic capabilities, we’re taking the next step in the evolution of the FOX AdStudio,” said Jeff Collins, FOX’s president of advertising, marketing and brand partnerships. Three separate technology partners for three separate tasks is a tell: nobody in this market, including a company with FOX’s scale, is trying to ship one agent that plans, buys and reconciles a whole campaign. They are shipping several narrow agents stitched to a platform a human still supervises.
The same task-by-task pattern shows up on the creative side. AdAI, a New York startup incubated by Revolution Venture Studios that came out of stealth on September 11, builds finished digital ad creative from a brand’s existing assets and performance data rather than generating buying decisions. “As the digital ad landscape transforms from manual creation to agentic automation, ad volume has increased exponentially. Quality has never been so critical, and yet in such short supply,” said Kevin Bartley, AdAI’s co-founder and CEO. His co-founder, RVS managing partner Kyle Robertson, put the bet on the category rather than the company: “Digital advertising is one of the largest pools of allocated capital in the world. AI is changing how ads are created and consumed. This is the moment for change, and for a new kind of adtech company.” Whether that specific company wins is beside the point for a marketing leader. What matters is that venture capital is now funding narrowly scoped agentic tools at the production layer, not just at the buying layer, which means the fragmentation Magnite and FOX show in media buying is going to repeat in creative operations too.
The plumbing underneath: a shared language, not a shared agent
The reason none of these companies are shipping one do everything agent becomes clearer in a second Magnite release, published September 16, announcing what it calls Magnite Orchestration: a coordination layer meant to let AI buying systems from different vendors discover, evaluate, activate and optimize inventory across a common protocol rather than each building its own point-to-point integration. The list of named participants is the real story here, not the product name. ITN (local linear TV), DIRECTV Advertising, Samsung Ads, iHeartMedia, Newton Research, Databricks and the agency Butler/Till, plus the French broadcaster TF1 PUB, are all listed as building on or supporting the layer. That is a sell side platform, two data companies, an agency, a linear TV technology vendor and a European broadcaster, none of whom compete directly with each other, agreeing to speak the same protocol so that whichever agent a buyer happens to use can transact against whichever seller’s inventory without custom code on each side.
“Agentic AI is going to fundamentally transform how media is bought, sold, and experienced,” said Sean Buckley, Magnite’s president of revenue and market strategy, in the release. Manny Puentes, chief product officer at Butler/Till, described the gap the layer is meant to close: “This industry has needed a connective fabric for a long time, something that weaves platforms together.” DIRECTV Advertising’s chief advertising officer, Amy Leifer, made the more important point for a marketing leader trying to read past the hype: “AI is changing how media is planned and bought, but it doesn’t change what advertisers value most.” In other words, the agent is a faster way to execute a plan whose underlying goals, reach, brand safety, measured outcomes, have not changed. This is the same lesson the IAB Europe data below makes at the survey level: the technology is arriving as connective tissue between systems humans already run, not as a replacement decision maker.
What the data says about the gap between ambition and practice
The most useful primary of the four is not a product announcement at all. IAB Europe’s 2026 Impact of AI on Digital Advertising report, released September 24 after surveying advertising executives across 44 European markets, found that 58 percent of respondents expect agentic ad buying to reach operational use or scale within a year. That is a striking number until it is set against the report’s other finding: only 86 percent of organizations with more than 500 staff, and just 48 percent of smaller ones, report having reached what IAB Europe calls Level 2, advanced production systems with human agent collaboration. Expectation is running ahead of deployment, and the gap is widest exactly where most marketing budgets sit, inside mid-sized organizations that lack the systems integration teams FOX and Magnite used to ship their pilots.
Dimitris Beis, IAB Europe’s data and innovation strategist, named the gap directly: “Ambition remains ahead of practice in areas such as agentic trading, although the results suggest strong interest and continuing investment. For industry leadership, the practical questions are familiar: where is the value, have we established the conditions to capture it, and who is accountable when something goes wrong?” His colleague Katharina Schneeberg, who co-leads IAB Europe’s AI working group, located the actual blocker as organizational rather than technical: “AI adoption is no longer the key challenge. Organisations are now focused on scaling AI across their business, establishing governance and translating experimentation into tangible business impact.” Every one of the primaries behind this piece backs that claim in its own way: Magnite’s win required Amnet to already have clean inventory access, FOX split its stack across three named integration partners rather than build one internally, Magnite’s orchestration layer exists precisely because no single vendor’s agent can reach the whole market alone, and AdAI is selling into brands that still need “brand safety and legal compliance” built in, per its own launch materials, before an agent can touch regulated-industry creative.
Why the promise keeps outrunning the rollout
The pattern across all four releases is the same failure mode that has dogged every automation wave in ad tech since real-time bidding: the task that is easiest to hand to software, campaign setup and trafficking, is also the task that was already the least strategic part of the job. Cutting Amnet’s setup time by 70 percent is a real operational win, and marketers should take it. It is not evidence that an agent can choose media strategy, adjudicate brand safety trade-offs, or explain to a CFO why spend moved from one channel to another. None of the four organizations behind this reporting claim that it can. The gap between the 58 percent of European executives who expect scale within a year and the roughly half who have not yet reached advanced production deployment is the gap between believing agentic AI will eventually do the strategic work and having any agent that currently does.
What this means for the marketing leader
Three things follow for anyone evaluating an agentic pitch this quarter. First, ask what specific task the agent performs, not what category it belongs to. “Agentic buying platform” describes Magnite’s ClearLine, FOX AdStudio and Magnite’s own Orchestration layer equally, and all three do different, narrow things. Second, ask whether the vendor’s agent speaks to a shared protocol like Magnite Orchestration or only to its own walled stack, since the former can reach inventory the latter cannot. treat the accountability question IAB Europe’s Beis raised as a procurement requirement, not a hypothetical: before an agent touches live spend, know who signs off when it is wrong, and confirm that governance exists on paper, since only 48 percent of the report’s respondents had marketing-specific AI guidelines in place at all. Third, budget agentic tools for operational efficiency this year, not strategic judgment. The 70 percent setup time reduction Magnite reported is a real, bookable cost saving, similar in kind to what marketing execution is already showing as campaigns start running themselves. It is a different claim from an agent that can be trusted with strategy, which is closer to what this publication has argued is still missing from agentic RFP tools that move the gatekeeper rather than remove it.
None of that argues against adoption. It argues for matching the claim to the task. The vendors making the most credible claims this month, Magnite and FOX among them, are also the ones being the most specific about what their agents do not do. That specificity, more than any demo, is the signal worth buying on.
Source: IAB Europe
