In May 2024, WPP unveiled WPP Open Pro, a generative-AI front end that lets planners brief, plan and execute media work inside a single conversational surface. Six months later Publicis followed with CoreAI, described in their own announcement as a system that 'supports decision-making across the entire business'. The vocabulary changed in a quarter. By the start of 2026, every holding company had something to say about agents, and a few had agents in the live buying loop.
The shift is easy to under-read. Marketing has had AI in the bid layer for fifteen years. What is new is that the system is now allowed to choose what to bid on in the first place.
What 'agentic' actually buys you
The most useful definition is operational, not philosophical. An agent in marketing has three properties.
- It can read the brief and produce a media plan, not just a parameter set
- It can reach into channel APIs and execute that plan, not just recommend it
- It can evaluate the result against the brand's reward signal and revise without a strategist closing the loop
When all three are present the work that used to live in slides starts living in databases. A planner who used to negotiate four channel teams now writes a policy document the agent reads. The cycle compresses from weeks to hours. Forrester's 2025 report on autonomous marketing framed it as a generational shift, and the holding-company earnings calls since have used the same language.
What CMOs should ask before signing
The risk in this category is selecting on demo polish rather than operational depth. Three questions usually separate the real systems from the wrappers.
- Show me the policy file. If the agent's guardrails live in a Slack thread or a strategist's head, the system is not autonomous. It is automated.
- Show me a write-API integration to a real channel. Many platforms read from Meta and Google Ads but cannot push back. Without push, the agent is a dashboard.
- Show me the reward function. Optimising for ROAS is fine for paid search
- an agent operating across brand and performance needs a richer signal that captures audience quality and creative diversity. If the team cannot produce that function, the agent will optimise for whatever is easiest to count.
“Most agencies are still teaching agents to push buttons inside a campaign. The work that pays is teaching them to design the campaign in the first place.”
The pricing implication
The billable hour does not survive a system that does the work in seconds. Agencies that built the holding-company-classic model on time-and-materials are watching the unit economics invert quarter on quarter. The teams running real outcome pricing today are doing better revenue per logo with smaller teams. The teams still defending hourly will be acquired or absorbed into ones that aren't.
We price agentic engagements as a fixed scope plus an outcome share. The scope covers the policy work, the integration audit and the reward-function design. The outcome share is the alignment. It removes the conversation about how many hours the agent spent doing the work. The work is what matters.