PALM SPRINGS, CALIF. – Data clean rooms were never meant to be a silver bullet. But, as AI agents begin reaching across organizational boundaries to hoover up everything they can find, the humble clean room is getting a second look — this time, as a critical guardrail.
That tension, between AI’s appetite for data and the enterprise’s need to control it, sits at the heart of Snowflake’s data infrastructure play.
“When you bring AI into the mix, it’s a whole different ball game,” said Dennis Buchheim, global head of marketing technology, media and entertainment at Snowflake, in this video interview with Beet.TV.
“AI wants to be able to connect all the data that it possibly can and ingest that and build models and learn so that it can answer the questions and provide the insights you want about your data. And that’s where clean rooms and other controls and governance around data collaboration, we just think are so incredibly important.”
Clean rooms as connective tissue
Buchheim said clean rooms are not a standalone product category but one tool among several for enabling secure data collaboration. His example of Disney illustrated the internal use case: a company with a parks business, a streaming business, and other divisions might want to link data across those silos without giving every team unfettered access to everything.
Snowflake has been building out its clean room capabilities since acquiring Samooha, a cross-cloud data collaboration startup, in late 2023. The acquisition fed directly into the launch of Snowflake Data Clean Rooms, designed to let enterprises share and analyze data across cloud environments under privacy and governance constraints.
More recently, NIQ launched a global data clean room on Snowflake in October 2025, enabling marketers to enrich first-party data and measure ad effectiveness within that controlled environment.
The agent-to-agent frontier
If clean rooms represent the governance layer, AI agents represent the activation layer.
Buchheim distinguished between two modes:
- Conversational AI, where a user queries their own data.
- Agent-to-agent communication, where autonomous systems interact with each other and potentially take action.
Snowflake has invested in a product called Snowflake Intelligence to address the former. The latter is where things get genuinely speculative.
Buchheim described a scenario in which a marketer’s agent queries a publisher’s agent about available inventory and audience segments — and then, in a more advanced version, actually executes the media buy autonomously. “We’re at the former state and that’s still exploratory,” he said. “We’re certainly not quite at the latter state of agents being autonomous and executing.”
The infrastructure underpinning agent-to-agent communication is itself unsettled. Buchheim pointed to emerging and competing protocols – citing “AdCP” as one example in the advertising context – as evidence that the industry has not yet converged on standards for how agents should talk to each other, what they are permitted to do, or what controls should govern their interactions.
It is a standards race that could shape the architecture of programmatic advertising for years to come.
You’re watching Beet.TV coverage from IAB ALM 2026. For more videos from this series, please visit this page.
