Databricks presents a reference implementation of agentic media buying and selling built entirely on their platform. The system uses multi-agent crews (built with CrewAI) to automate the traditional manual ad-buying workflow — budget planning, audience mapping, publisher discovery, price negotiation, and deal booking. It leverages IAB Tech Lab's AAMP open standards and SDK for interoperability, Lakebase (serverless Postgres) for transactional state, Unity Catalog for data governance and lineage, OAuth and service principals for identity/trust tiers, and MLflow tracing for observability. The accelerator deploys with a single CLI command and serves as a working blueprint for autonomous buyer-seller agent systems in the advertising industry.
Table of contents
The bottleneck in media buying today isn't talent, it's coordinationWhy this is finally solvable and why standards matterWhat goes into an agentic media buyA complete ecosystem for buyer and seller agentsWhat does the right architecture look like?Databricks Model Serving: the agent crewLakebase: transactional state for agentsGovernance: Lakebase today, Unity Catalog nextIdentity and trustObservabilitySee the agentic buying app in actionWhat’s NextDeploy it yourself96 Impressions