• by Laurie Sullivan , Staff Writer, 9 hours ago

As industry conversation revolves around whether AI will actually run media buying, this real example will give advertisers and media buyers a glimpse into how it works.

Geloso Beverage Group in December 2025 ran what its digital marketing agency partner and ad-tech platform provider called "the industry’s first fully autonomous, end-to-end agentic campaign." 

Butler/Till, the agency, and PubMatic, the ad tech provider, began sharing the results on Tuesday. The campaign ran across premium connected TV (CTV), sports, online and mobile app publishers, and geo-targeted consumers 21 years of age and older.

AI agents on PubMatic’s AgenticOS platform planned and executed the entire campaign. No traditional demand-side platform (DSP) workflow and no manual optimization desk were used, and the campaign outperformed the original plan across every major metric.

AgenticOS, launched in January, is an operating system designed to orchestrate autonomous, agent-to-agent advertising -- allowing the agent to plan, transact, and optimize programmatic advertising.

The advertisers define the objectives, guardrails, brand-safety requirements, and creative parameters in their preferred LLM interface, such as Gemini or OpenAI. PubMatic’s platform takes it from there.

It should not be a surprise that the campaign reached performance expectations and outperformed planned baselines on every point with lower cost, higher quality and a greater number of impressions at higher completion rates.

There will be mistakes along the way, but as long as advertisers can follow guidelines, they should attain these kinds of results, because the agentic technology learns from mistakes and accomplishments based on historic and real-time data. A year from now, this will not be an anomaly.

Geloso used Anthropic Claude’s large language model (LLM) to interpret media briefs, generate strategy, and execute media buys through PubMatic’s AgenticOS.

The companies used Model Context Protocol (MCP) for agent communication and the Ad Context Protocol (AdCP) to standardize programmatic ad workflows.

To make the agentic media buy happen, the technology relied on technology that translates human intent into machine-executable actions using the AdCP framework.

Rather than just using a bid request, AdCP supports a multi-step conversation process. A brand AI agent can ask, "Find me women interested in bicycling from Huntington Beach to San Diego, California, along the coast,” and a publisher agent would respond with specific matching inventory.

When the agents and technology negotiate the deal, AdCP allows the transaction to be written directly into an ad server. It often bypasses traditional complex programmatic bid stream requests.

Early tests in January showed that campaign setup time could be reduced by 87% and issue resolution by 70%, according to PubMatic.

A couple of month later, the companies shared the results. The campaign achieved 5.5x greater supply-chain cost efficiency compared with the standard economics of traditional demand-side platforms (DSPs), according to the companies. The campaign delivered a 40% increase of total impressions compared with the plan running, on the same budget. Video performance campaigns reached a 98% video completion rate (VCR). This means that nearly every viewer watched the content to the end. That may be because of the AI’s ability to sharply target consumers.

Less than 1% of the ad inventory failed to meet DoubleVerify standards, a performance roughly 80% better than the current industry benchmark, the companies reported.