Testing of AI agents is becoming more widespread, but questions about differentiation and the economic model are fueling reflections behind the scenes at Cannes Lions.
During Lions, Cannes is the capital of brilliant marketing operations, but also an excellent thermometer of technological, operational and business changes in adtech. A year ago, everyone was talking about the arrival of AI in media buying. Now it’s done.
AI agents on all floors
The arrival of agents to manage the negotiation, planning, and even the set-up, the purchase order and the optimization takes place on all links in the chain: DSP, SSP, media quality verification tools, start-upagencies, publishers.
An emblematic example is that of Amazon, which has integrated its AI agents for all tasks and on all inventories (sponsored links and DSP) and of which we learned this Monday, June 22 in Cannes that they are now partly available in France. From brief to measurement, including set-up and activation, the advertiser can, if he wishes, entrust everything to Ads Agent for his operations across the entire conversion funnel, including via his own agent by integrating via MCP.
“The launch (in October 2025, editor’s note) of AdCP served as a trigger for all of us to make concrete progress on agentics. Since then, MCP has developed and other forms of integration with AI have become clearer, such as direct connection via API. What is certain is that agentics will help to promote and reveal the full potential of publishers’ inventories”, analyzes Edouard Schmidt, country manager for France and Belgium at Magnite.
Magnite is testing its seller and buyer agents and its orchestration tool for the two links in the chain, including agents external to the platform, on over-the-counter operations on CTV in the United States. In France, the first tests should take place no later than the fall of this year, our interlocutor tells us in his yacht moored on the pier, Albert Edouard.
Magnite is a founding member of adCP, but it is not the only one to have this analysis on the role of the launch of this protocol in the acceleration of agentics in media buying. Laure Malergue, CEO of Displayce, thinks so too. Displayce, which launched its suite of agents available on MCP on June 25 in Cannes for proposing media plans and analyzing DOOH campaigns.
All interoperable
In this race against time, AI agents are also arriving at media quality and performance verification tools. Like at DoubleVerify, which announced in Cannes on Monday June 22 the global launch of “Neura”. “Neura is a suite of AI agents that allows advertisers to obtain recommendations in natural language and (soon) to act at their request in optimizing campaigns on all channels and platforms based on the information obtained,” explains to JDN Alex Valle, chief product officer at DV, met during his stay on the pier.
Here again, the logic of interoperability is at work: the advertiser can also, if he wishes, interrogate Neura with his own AI agent, via an MCP connection or see other standards such as AdCP and AAMP from the IAB Tech Lab. “Ultimately we will also interface with the platforms (open web and Gafam, editor’s note) via MCP. This is already the case with Pubmatic and it will also be the case with Amazon in a few weeks and soon DV 360 and TikTok”, he adds. The advertiser’s agent can let Neura act or take the recommendations of the DV agent to optimize the campaign directly on the advertising platform via its MCP connection.
The generalization is such that the question of differentiation now arises. “The mere fact of having an agent will soon no longer be a differentiating factor,” says Laure Malergue. “What will count is the quality of the agent, which depends on two things: the data to which it has access and the workflow, i.e. the business process on which it is trained.”
The economic model in question
It is not just differentiation that gives rise to reflections among the professionals encountered in the vicinity of the Croisette: the viability of the all-agentic economic model is already at the center of reflections. In most cases, the proposed AI agents are not charged to customers, given that they are just released, either to encourage broad adoption or as a retention tool. “The question of the economic model still remains open and it will be key for the future, because agents are expensive,” agrees Edouard Schmidt.
A point also noted by Maxime Woussen, DGA of Making Science in France. “Adtech today is in the process of testing everything. But it will have to balance the choice of the power of the models deployed according to the use cases and the incremental benefits that AI will bring. Furthermore, even if the costs with AI can quickly add up, they are normally offset by the growth that these tools make it possible to obtain, which is what we are observing.”
The fact remains that all these agent-based tools will necessarily have standard logic. Advertisers who ask their adtech service providers for ultra-personalized configurations will undoubtedly be charged for the performance increment they obtain.
Craze for the ChatGPT offer
Even the general media noticed the long-awaited presentation of OpenAI on Monday June 22 at Lions, whose presence for the first time at the now iconic adtech show symbolizes its ambitions. Beyond the pitch on the progress of its stack and its offer, it is the enthusiasm it arouses which gives the “the” of future trends. “All our clients want to try advertising on ChatGPT as soon as it is possible in France,” attests Maxime Woussen. Same thing for Google: in Europe, in all countries where AI Overviews and AI Mode are deployed, the publicity begins to flow freely in the responsessometimes even without the advertiser being aware of it, via tools such as AI Max for Search.
Among adtech media quality verification companies, it’s the same song. “As advertising grows on conversational engines, advertisers need third-party tools to ensure the quality of these environments. DV is currently talking to all the major players in the sector and tests are already underway,” Alex Valle tells us.
