The summer AI moved into the marketing stack

Ads went inside the answer. Agents went into production. And the gap between ambition and execution turned into the clearest opportunity of the year.

If the spring of 2026 was spent talking about agentic AI, the summer was the quarter it stopped being a conference topic and started shipping inside the tools marketers use every day. Three things happened in parallel: advertising moved into the chat interfaces where people now ask their questions, the big platforms released agents that carry out work rather than just suggest it, and the consultancies put hard numbers on the distance between what companies say they are doing with AI and what they are actually doing. Read together, they describe a single shift, and a window that is still open.

Advertising moved inside the answer

On 31 August, OpenAI reported that ChatGPT Ads had passed one billion dollars in annualised run rate in under 200 days, and opened self-service advertising through its Ads Manager across India, Europe, the Middle East and North Africa.

The mechanics will look familiar to anyone who has run paid media: cost-per-click and outcome-optimised bidding, product feeds, geo-targeting, custom audiences and conversion measurement. Ads are labelled, kept separate from the model's answers, and, according to OpenAI, do not influence what the assistant says.

The format is not the story. The surface is. With more than a billion weekly users, and advertising now available to European businesses, a genuinely new channel enters the mix: the place where people go to ask, compare and decide. That reshapes what “search” and “purchase intent” even mean. For anyone planning a channel mix for 2027, the question is no longer whether the assistant is a channel, but what a brand is supposed to look like once the answer, not the results page, is the destination.

Agents moved from demo to production

The second shift is quieter and more structural. The platforms are embedding agents that do the work, directly inside the software you already own.

In early September, Anthropic released Claude Commerce Agents, an open (Apache-2.0) blueprint with two agent types. A shopping agent handles the customer side: catalogue search, product comparison, cart building and service. A merchant agent supports the team behind the counter with inventory, pricing and, notably, marketing campaigns. Releasing it openly matters: it lowers the barrier for the whole industry to build production-grade agents rather than demos.

Google set the same direction earlier in the year at Google Marketing Live. Gemini now powers new ad formats inside Search and AI Mode, a unified “Ask Advisor” agent spans Ads, Analytics and Merchant Center, and Asset Studio gained multimodal capabilities that turn brand guidelines into finished creative.

The pattern is consistent across very different companies: the systems you already use are becoming capable of running their own agents inside your marketing production. The skill that matters shifts from “how do I operate this tool” to “what outcome do I want, and what guardrails do I set.” The job moves from making assets to directing the systems that make them.

The consultancies put a number on the gap

The same summer produced the data that puts all of this in perspective, and it is more sobering than the product launches suggest.

BCG's June report on agentic marketing (300 CMOs) found that while roughly half of marketing leaders now control the AI budget inside their own function, only 8% are running autonomous, multi-agent campaigns. Forty-two percent still use generative AI purely as an assistant for individual tasks. Ambition is everywhere; execution is rare.

McKinsey, publishing the same month, estimated that agentic AI could power around two-thirds of today's marketing activities and compress campaign timelines by ten to fifteen times. Its central warning is what it calls the “gen-AI paradox”: the technology can be found everywhere except on the bottom line. The value, McKinsey argues, does not come from automating isolated tasks. It comes from connecting insight, planning, production, testing and optimisation into one coordinated system.

What actually separates the winners

Put the launches and the data side by side and the lesson is not “buy more AI.” Nearly everyone already has the model. What almost nobody has yet is a system, and a point of view for that system to express.

When advertising, creative and campaign execution all become agentic and abundant, volume stops being a differentiator. Two things take its place. The first is context: how deeply the agent understands a specific brand, its voice, its best-performing work, its terminology, how it differs from competitors. Most horizontal, one-size-fits-all agents capture this only at the surface: a short description, a few keywords, a tone slider. The more companies switch on those built-in features, the more the output converges. Different senders, same voice.

The second is orchestration. The organisations pulling ahead are not the ones with the most access; they are the ones that have connected the pieces into a workflow and built the experience to run it. That is precisely why the 8% figure is not a warning to move slowly. It is a description of open space. The gap between what AI can do and what most teams actually do with it is not an argument to wait, it is the window, and McKinsey's ten-to-fifteen-times timeline is, quite literally, the size of the head start on offer.

Where to start

The move that separates the 20% from everyone else, across every study this summer, is unglamorous: deciding to actually use AI on real work instead of talking about it. In practice that means picking a few measurable activities, connecting the agent to your own structured data and brand context rather than generic settings, keeping a human on strategy and judgment, and treating the workflow, not any single tool, as the thing you are building.

Summer 2026 was not the quarter AI arrived in marketing. It was the quarter it stopped being an experiment on the side and became part of the infrastructure. The brands that treat it that way now are the ones who will own the answer when everyone else is still buying the tool.

Sources: OpenAI, Expanding access to AI with ChatGPT ads (31 Aug 2026). Anthropic, Claude Commerce Agents (3 Sep 2026). Google, Google Marketing Live 2026 (20 May 2026). BCG, Making the Agentic Marketing Transformation a Reality (29 Jun 2026). McKinsey, Re-inventing marketing workflows with agentic AI (Jun 2026).