Butler/Till just wrapped its first real-world test of an “agentic” programmatic buy for Geloso Beverage Group, working with PubMatic—and the numbers are hard to ignore. Across two flights (Dec. 2025–Jan.) running on CTV, online, and mobile app inventory from names like Samsung, Paramount, Vizio, and Tubi, the setup cut supply chain costs by 82% (with DSP tech fees taking the biggest hit). Those savings translated into 40% more impressions and a 30% drop in CPMs, while still landing solid quality metrics: 98% video completion rate, <1% MFA per Jounce, and 80% of inventory scoring above DoubleVerify’s typical benchmark.
The mechanics were more “agent-to-agent messaging” than sci-fi: Butler/Till used an Anthropic Claude-based agent to translate a human-written brief into targeting instructions sent into PubMatic’s AgenticOS. PubMatic’s system surfaced curated inventory, humans approved it, and then PubMatic’s existing pipes handled the actual purchase. Butler/Till’s Gina Whelehan framed it mainly as a feasibility test—can this run and scale in-market?—with fee reduction a welcome side effect rather than the original bullseye.
Bigger picture: everyone’s building agents, but they’re not all betting on the same rails. While Butler/Till’s test sits in the AdCP-style world of agentic execution, FreeWheel just introduced an MCP server aimed at giving agencies an infrastructure layer for planning agents (PMG has used it since December to speed CTV/streaming campaign setup—not to execute buys). PubMatic says it has “upwards of 10” agency tests underway and claims a 98% cut in campaign setup time in its own experiments—great for anyone who’s ever lost a day to dropdown menus.
The looming question isn’t whether middlemen and manual hours get squeezed—it’s who still earns a seat when it happens, with Forrester noting survivors will need to win on transparency, explainability, and trust, not just position in the chain.
Read more at Digiday.