Agentic AI and the future of Advertising Agencies
What happens when the agency org chart can be replaced with a directed graph of agents — and which parts of the work refuse to flatten.
For twenty years I watched the agency model bend. It bent around digital, around social, around programmatic. Each time, people predicted a collapse that didn’t come — the holding companies restructured, hired differently, charged new rates.
Agentic AI feels different. Not because of the technology itself, but because of what it replaces. The previous disruptions all automated the execution layer. Agentic AI is going after the coordination layer — the account manager, the traffic manager, the project manager, the junior strategist assembling the brief.
What an agent actually does
An agent, in the technical sense, is a model that can take actions, observe outcomes, and loop. Give it a goal — “build a campaign brief for this product launch targeting urban women 25–40 in India” — and it doesn’t just return text. It searches, reads, synthesises, drafts, reviews against a rubric, and returns something that used to take a team three days to produce.
That’s not a parlour trick. That’s an org chart question.
The parts that flatten, and the parts that don’t
The flattening happens where the work is throughput: brief assembly, research decks, copy variations, legal reviews, scheduling. These are coordination tasks dressed up as creative work. Most agencies quietly know this.
What doesn’t flatten: genuine taste, earned trust, and the ability to hold a client’s anxiety at 11pm the week before a launch. Those are human-scale problems. An agent can draft the deck. It can’t sit across the table and make someone feel like they made the right bet.
The agencies that survive will figure out what they’re really selling. The ones that don’t will keep optimising for headcount efficiency until there’s no headcount left to optimise.
What this means if you’re building
The interesting opportunity isn’t “replace the agency.” It’s building the layer that sits between the agent infrastructure and the client’s marketing stack. The opinionated workflow. The trained taste. The accountability layer.
That’s what I’m working on. More on that soon.