A French media buyer and a TV ad sales company have run a campaign through AI agents instead of phone calls, emails and Excel sheets, a first in France. The agents negotiated, configured and activated the campaign end to end. They did it over the Ad Context Protocol (AdCP), a standard that is still little known but could become foundational for ad tech. Here is what actually happened under the hood, step by step.
The demo was run by Olivier Lavecot, digital and innovation director at French media agency Values.media, and Mohamed Laaouissi, co-founder of Teknalab.ai. It shows the first exchange in France between a media buyer and an ad sales company, Adtlas, through AI agents running on the AdCP protocol. Teknalab.ai is an AI-driven platform that brings market analysis and media activation into one place. Here, it supplied the agent on the sell side.
AdCP is an exchange standard. It lets AI agents on the buy side and the sell side talk to each other directly to negotiate, create, activate and track ad campaigns in an automated, interoperable way. The standard keeps changing, since its first version only came out in October 2025.
The test was not meant to simulate a use case. The goal was to launch and track a real campaign, from brief to reporting. It started from a simple brief: a healthcare advertiser wanted to run ads from Dec. 8 to Dec. 31, with a set budget.
The buyer typed these details into a generative AI interface, in this case Anthropic's Claude. From that prompt, the AI agent was authorized to query the ad sales company's AI agent over AdCP.
The agent first pulled what the sell side had available: inventory, channels on offer, rate cards and list CPMs. From there, it returned a first structured recommendation. For a health campaign, it picked health channel Top Santé as the best fit, along with other channels it judged less relevant.
Once the buyer shared the budget (€2,000), the agent went back to the ad sales company to estimate the volume available over the period. It then produced a projection of total estimated impressions, broken down by day and by daypart. The buyer could see from it how much ad pressure to expect and how delivery would be paced.
The buyer then asked which programs the ads would run around. The agent pulled the list of shows and returned it. That added a qualitative view of the editorial context to the pricing and volume data.
Once the buyer approved these settings, the agent prepared the campaign. It set out the key elements: campaign name, channel, budget, dates and the other required parameters. It then created the campaign with a unique ID. At that point, the campaign sat in draft status, waiting for the ad sales company to approve it.
On the sell side, the campaign showed up directly in the ad sales company's dashboard. It stayed inactive until the seller signed off. Once accepted, it switched to active and could start airing.
The dashboard then tracked the main metrics: budget spent, impressions, number of airings, channels used and day-by-day progress.
With the campaign live, the buyer could go back to the AI agent with the campaign ID. The agent then returned a delivery status with aggregated performance data, detailed tables and charts breaking down delivery by program, format or daypart.
The demo deliberately kept a narrow scope: one ad sales company, list prices, a single campaign. Still, it shows how AdCP works in practice on both the buy side and the sell side. It also shows how AI agents can run the whole workflow, from campaign creation to analysis, without a break between systems.
On a scope this narrow, the media buyer may not save much time. "What is certain is that the day we apply this system to a dozen ad sales partners, we'll save quite a lot," Lavecot said. Values.media hopes to bring most TV, out-of-home and radio ad sales companies into the project.
"We think AI agents can help direct-deal traders execute faster, and also get a better real-time view of all the inventory available," said Emmanuel Crego, managing director of Values.media.
Editor's note: In the video, Lavecot talks from Claude directly to the agent representing the ad sales company. To set that up, he only had to open the tool's settings, go to "Connectors" and add an AdCP connector by pasting the URL the seller provided. To work with other ad sales companies this way, he would have to repeat the step for each one, so his agent can talk to each new seller's agent.
Once the AdCP box is checked on the home screen, the tool figures out on its own when it needs to query the seller. In this example, Claude's agent talks to the agent from Technalab.ai, which makes it a generic agent. But nothing stops Values.media from building its own agent, with more media planning expertise built in.
One last detail: two small "bugs" appear on screen, and neither affected how the campaign ran. The year shows as 2024 and the price appears in dollars. Rest assured, the campaign aired in late 2025 and was bought in euros.
