Audiences have always moved as lists and segment IDs: text that has to be matched, synced, and reconciled between systems. That is slow and leaks information. Agentic Audiences proposes a different representation, built for machine speed.
Agentic Audiences (formerly the User Context Protocol, contributed to the IAB Tech Lab by LiveRamp) exchanges audience signals between agents as vector embeddings rather than text. An embedding is a compact numeric representation of meaning. Comparing embeddings is fast and does not require passing around the raw segments, which makes matching quicker and more privacy-aware.
Key idea
The goal is the same one from Track D: get the right audience in front of the buyer so relevance, and price, go up. The difference is the mechanism. Instead of syncing segment lists, agents compare embeddings at machine speed.
It sits alongside the AdCP Signals module from Track H. Both are about activating audiences for agents; Agentic Audiences is the IAB Tech Lab’s fast, embedding-based route. Consent and privacy rules still apply, the representation just makes the exchange leaner.
Quick check
What is the main difference between Agentic Audiences and passing traditional segment lists, and what stays the same?