AGTNEWS
Dr. Daniel Hulme has a theory about where the advertising industry stands on AI agents: the teenage sex phase. Everyone's talking about it, nobody's sure anyone else is actually doing it, and nobody wants to admit they're not sure what they're doing either.
That assessment, delivered at the IAB UK AI Growth Summit in London on May 21, was blunt. The deployment of AI agents across organizations will be "a shit show," Hulme told attendees, because most are not actually capable of doing the jobs they're being assigned. Digiday
There's a structural reason agencies are walking into this trap. About 62% of mid-market advertising agencies bill by the hour, according to Digital Applied. Digital Applied The logic is simple and brutal: an AI agent that delivers the same output in three minutes that a human team used to produce in three days does not reduce costs under an hourly billing model. It reduces revenue. Either agencies restructure how they charge for work, or they have a financial incentive to ensure the agents never actually work.
That contradiction is the real story underneath the industry's AI celebration. The question is not whether the technology works. It's whether the business model can survive it.
The deployment gap is measurable. Only 11% of mid-market agencies have a mature evaluation harness for agentic delivery — the kind of system that tells you whether an agent is actually completing the task or failing in ways that still generate a billable hour. Digital Applied At the IAB event, Hulme estimated that at least 80% of the energy required to build agents goes into testing. Digiday Without the eval infrastructure to tell success from expensive failure, most agencies are flying blind on whether any of this is working.
The practical consequence shows up in the billing confusion the Digital Applied analysis describes: when an agent produces correct output faster than the human team it replaced, the client sees the same deliverable in less time and questions why the invoice has not adjusted. The agency, still staffed and overheaded for the old timeline, has no measurement system to justify the charge or renegotiate the scope. The productivity gain becomes an accounting problem instead of a business case.
The strategic picture is not rosier. About 42% of organizations are still developing their agentic AI road map, with 35% reporting no formal strategy at all, according to Deloitte's 2026 technology management survey. Deloitte Gartner predicts that over 40% of agentic AI projects will fail by 2027 because legacy systems cannot support the demands of modern AI execution. Deloitte
The technology itself is not the bottleneck, and Hulme is clear on this point. "The industry is working with the weakest possible version of AI," he told the summit. "It is doing very fast, very sophisticated rule following." Digiday The problem is not that the AI is not smart enough. The problem is that deploying it honestly requires an industry-wide accounting restructuring that no holding company has publicly committed to.
The deeper framing comes from Hulme's prior work on AI prediction. Marketing, he has argued, is a second-order chaotic problem: if you can model how people will behave accurately enough to change their behavior, the model invalidates itself. The Drum Hulme describes agents that observe outcomes and adapt. That capability would break the billing model. The industry wants the capability without the breakage.
The structural trap has a concrete beneficiary if agencies don't move. If the hourly model holds, the productivity gain doesn't vanish — it relocates to consultants who can bill outcome-based fees, to tech platforms that own the agent infrastructure, and to in-house enterprise AI teams at large advertisers who bypass agencies entirely. The eval harness gap, currently at 11% with mature systems, means most agencies cannot even prove they are delivering value faster, let alone charge for it. Every month the industry spends protecting hourly billing is a month the alternative ecosystem matures.
The test will come from pricing. If an agency can credibly move from hourly billing to outcome-based pricing, charging for the value of the work rather than the hours spent, the productivity multiplier aligns with the revenue model. If the industry protects the hourly structure and slows AI deployment to preserve it, the deployment gap widens and the "shit show" Hulme predicted extends past 2027. Right now, the industry is doing neither deliberately. It is hoping the problem resolves itself.
There is a counterargument worth naming: this analysis is anchored to one executive at one company. WPP may be structurally behind its competitors on agentic deployment. Publicis has announced agentic AI initiatives and Omnicom has disclosed experimental media buys through its agent framework. Neither has disclosed functional production deployments generating measurable revenue, nor has either publicly committed to restructuring its billing model away from hourly. The kill-if-false is specific: if any major holding company had disclosed functional agentic deployments generating measurable revenue and is actively restructuring billing toward outcome-based pricing, the structural argument collapses. That disclosure has not happened.
The IAB UK AI Growth Summit took place May 21, 2026 at Kings Place, London. IAB UK Whether anyone at that event had a plan for the billing problem was not reported.