Build and govern agents using enterprise data and shared controls
Databricks Agent Bricks is a platform for developing and operating AI agents within an enterprise data environment. It connects agent behavior with information such as schemas, business definitions and governed data sources. The product supports both developer-led work and managed builders for specific tasks. Its purpose is to make agent development, access management and deployment part of the same operating environment instead of leaving each team to assemble disconnected controls.
Agent Bricks supports multiple model providers and frameworks, allowing teams to select components for their use case. MCP connections give agents a route to approved tools and external systems, while Unity Catalog supplies shared governance. The platform also uses MLflow tracing and evaluation capabilities to inspect interactions and compare behavior. These facilities help teams understand what an agent called and how it reached a result. They do not eliminate the need to define acceptable behavior or to test an agent on realistic business cases before deployment.
The official product describes serverless deployment, REST interfaces and ongoing monitoring, with costs based on the underlying Databricks services used. A useful project begins with a bounded task, authorized data and explicit rules about permitted actions. Developers can evaluate sample requests, inspect tool calls and compare quality against latency and cost requirements. Before broader release, confirm workspace permissions, provider availability and billing for the chosen components. Agent Bricks is most relevant to organizations that want AI development connected to their existing data platform, with ownership and review continuing after an agent goes live.
