By Seb Joseph • May 22, 2026 •
Ivy Liu
WPP’s chief AI officer has a theory about agentic AI. It is, Dr Daniel Hulm told attendees at the IAB U.K. ‘s AI growth summit in the “teenage sex phase” — everyone thinks everyone else is doing it but when they actually look, they’re not.
It’s a crude metaphor. It’s also probably accurate.
The past 12 months have produced a remarkable volume of agentic AI announcements from across the industry. Holdcos have unveiled AI operating systems, ad tech vendors have launched agentic campaign creation tools. Consultancies have published frameworks, CMOs have given keynote speeches about the autonomous future of marketing. And yet, when the people who are supposed to be deploying the tech — the agency traders, the media planners ad the campaign managers — are pressed, the honest answer is usually a version of: “we’re exploring it, we’re in pilot or we’re building the business case.”
Agentic AI is not the first technology to get caught in this particular gap, and it won’t be the last. Programmatic was going to revolutionize media buying — eventually it did, about a decade after the press releases said it would. Data clean rooms were going to helo fix third-party addressability in the open web — they’re still mostly unlocking it. CTV measurement was going to bring digital accountability to television — ask anyone trying to reconcile a cross-platform report how that’s going. Every generation of advertising technology gets its teenage sex moment, where the industry consensus outpaces the operational reality by enough to be embarrassing in hindsight.
What’s different about agentic AI is the texture of the gap. It’s not that companies are lying. It’s that almost no one is being honest about what deployment at scale actually requires.
“The reality is that companies will deploy an army of [agents] across the organization, and forgive the technical term, but it’s going to be a shit show, because most of those agents are not going to be capable of doing their job,” said Hulme. “At least 80% of the energy that you need to build agents is testing,”
The reason that’s especially important in marketing comes down to what Hulme called the second-order gap. Train an agent on historical campaign data, deploy it to make autonomous decisions and it immediately starts changing the very behavior it was built to predict. Inevitably, consumers respond differently, competitors react and media prices shift. The model, without the testing, risks becoming obsolete the moment it starts running. Marketers are not testing whether the agent does what it’s told so much as they’re testing whether what it’s told to do still makes sense after it’s been doing it for a week — against a market that has already moved.
“The problem is that buying and selling didn’t exist in the past.” said Hulme. “I’ve now changed the behavior, and marketing is the same thing. If I build a magic oracle and predict human behavior, I haven’t changed that behavior. Now I predict the models are out. You cannot predict the future based on the past.”
None of which means the tech isn’t coming. The more important part of Hulme’s argument made that all too clear. The industry, he suggested, is working with the weakest possible version of AI — think getting computers to do what humans can already do — when the genuinely disruptive version is systems that make decisions, observe outcomes and adapt. By that standard, the advertising industry is not yet doing AI in any meaningful sense. It is doing very fast, very sophisticated rule following.
“The reality is that quick wins and low hanging fruit can be solved by a third party at a fraction of the cost,” said Hulme. “You need to be focused on the problems and differentiate business.”
More in Media Buying
Read More
Seb Joseph