Databricks Agent Bricks

Agent Bricks is a Databricks platform to build and govern production AI agents that continuously improve, unifying development, management and governance of enterprise AI agents. It uses enterprise context — schemas, business definitions and custom semantics — to decide which tools and tables to use, join data correctly and produce accurate, consistent answers. It provides access to AI models from OpenAI, Anthropic, Google and open source through a single platform, letting users switch models instantly to optimize cost, quality and performance without re-architecting their stack. Databricks governs the full stack — from data to AI models — in a single system of record, tracking every agent, MCP server, model and tool with clear ownership and end-to-end permissions. Agents can be built with preferred IDEs and frameworks such as LangChain, LangGraph or LlamaIndex, or with no-code managed agent builders including Supervisor Agent and Knowledge Assistant, and validated in AI Playground before going live. Agents connect directly to the lakehouse for AI-ready enterprise data, with built-in context, retrieval and memory via Lakebase, and natively support the Model Context Protocol (MCP) for secure, governed access to APIs, databases and SaaS applications. Agents deploy to serverless compute via Databricks Apps with automatic scaling, and quality is continuously evaluated using LLM-as-judge, custom metrics and human feedback loops.