Your agency took three weeks on a competitive pitch. Fable, the AI tool Brian O'Kelley name-checked at Advertising Week New York in October, produces something close to the same output in twenty seconds. That is not a productivity gap. It is a business model gap. The generalist deck that has been the agency's calling card for two decades just got replaced by a free prompt, and the agencies that survive the next twenty-four months will be the ones that stopped trying to be everything to everyone before the industry caught up.
O'Kelley, who built Scope3 and in September renamed it Apostra after a one-eighty pivot from carbon measurement to agentic media buying, said it plainly on the AdWeek NY stage. The pitch where the agency shows up with a deck built on broad experience and lots of consultant hours is now competing against a model that has read everything, does not sleep, and does not invoice at FTE rates. "If your job is to make PowerPoints, I'm not really sure that's going to be super valuable," O'Kelley said.
What is actually happening on agency rosters right now
The holdcos are not pretending this is a branding problem. They are restructuring the operating model. WPP is selling WPP Open, its AI platform, as a client product. Publicis spent 2.2 billion dollars on LiveRamp to bolt data collaboration onto the Epsilon stack it bought in 2019. Omnicom, fresh off the IPG merger, is pushing Omni, the Acxiom-fueled data and AI platform, as the centerpiece of what a post-IPG Omnicom actually sells. These are not agency improvements. They are technology businesses wearing agency clothing, and the difference matters because clients are starting to buy the technology directly.
Horizon Media, sitting next to O'Kelley on the same stage, is taking a different cut. Dane Kunkel, Horizon's SVP of performance and transformation, said his entire department was built to help clients transform -- deploying agents not just for marketing but across finance and operations. As AI eats the individual tasks, Kunkel argued, someone has to be the overarching brain that connects planning, buying, data, and measurement. Horizon is now pitching itself as a growth partner rather than a media agency, and explicitly moving away from FTE-based pricing because FTE-based work is the part the AI does now.
Why "forward deployed engineer" is the wrong new title
InMobi CEO Abhay Singhal proposed a different answer on the same panel. If he were running an agency, he said, he would start by calling every media planner a "forward deployed engineer." It is a reasonable instinct. Forward deployed engineers are the embedded builders the AI software companies use to translate customer problems into working systems inside the customer's stack. It sounds like a great new job description. It is also a trap, because the real forward deployed engineers in 2026 are software architects with deep expertise in agent systems, security, and scale. You cannot go to a media planner and tell them to build database integrations with internal systems and think about security and scale and everything else.
The agencies that are responding to the title pressure by relabeling planners are going to lose twice. First, because the new title does not change what the planner can actually do. Second, because the cost of a real forward deployed engineer is north of what the agencies can pay. Meta is spending billions of dollars a year to attract and retain top AI talent. Publicis can pay for some of it through the LiveRamp acquisition. The mid-tier independent agency cannot. O'Kelley called it out: "Can't compete with that." Most of the engineering work at the holdcos is being outsourced to consultancies in India. That is not the same as having forward deployed engineers in the room with the client.
What the agencies that survive will actually sell
O'Kelley's prescription was simple and uncomfortable for generalists: specialize, verticalize, pick a lane. Be the consultancy that is damned good at pharma. Be the agency that crushes it in linear TV. Be the creative shop that knows how to run an AI creative loop better than anyone else. Whatever the lane is, go deep enough that the AI cannot replicate it from a generic prompt. The math is straightforward. AI is "pretty damn smart about almost everything," O'Kelley said. The agency that tries to be good at everything will be beaten by the AI at almost everything. The agency that is the best at one specific thing, with agents that help it execute, will own that thing.
This is not new advice, but it has become urgent because the cost of being a generalist just collapsed. Twenty seconds of Fable has wiped out the analytical premium the generalist pitch was built on. The agency's response cannot be to add a faster deck workflow on top of the existing one. The deck workflow is the part that went away. The agency's response has to be a different kind of work, priced and staffed differently, with a clear answer to the question every CMO will be asking by Q1 of next year: what is this agency doing that the AI in my own stack cannot do.
What to do this quarter if you are running an agency
Pick one vertical. Not one practice area, not one channel. One vertical where you will commit to being the best in the category. Hire or partner for the engineering depth that vertical actually needs, and be honest about the scope. You will not beat Meta on AI talent salaries. You will not replace your media planning team with software architects overnight. What you can do is build the combined human-plus-agent workflow that lives inside one vertical better than any generalist agency can.
Stop selling decks. Decks are what Fable replaced. Sell the work the AI cannot do alone -- the proprietary dataset on the vertical, the regulator relationship, the embedded agent that runs against the client's stack and only the client's stack. Price on outcomes, not hours. FTE-based retainers are how the generalist model got monetized, and they are how the generalist model will get automated out.
Rebrand the people, not just the titles. If your planners are still working on plans the AI can write in a prompt, your planners are not planners anymore. They are prompt operators, and the role needs a different job description, a different comp plan, and a different seat at the table with the client. The agencies that win the next twenty-four months will not be the ones that renamed the most planners fastest. They will be the ones that rebuilt the role around what planners will actually be doing when the AI does the rest.