Databricks Launches Agent Bricks With Supervisor Agent GA, Putting Unity Catalog Governance Between Agents and Enterprise Data

Databricks announced Agent Bricks on April 14, an enterprise agent platform built around a core premise: the hard part of production agents is not the reasoning loop. It is governing what agents can access, under whose identity, with what audit trail. The platform unifies model routing, data access, and governance through Unity Catalog, the same governance layer Databricks customers already use for their data operations, according to Databricks’ blog.

Three components reached general availability: Supervisor Agent (multi-agent orchestration), Document Intelligence (structured extraction from unstructured documents like contracts and invoices), and Custom Agents (build and deploy with any model or framework on serverless compute).

Governance That Covers the Full Stack

Most agent platforms govern the agent itself: what tools it can call and what permissions it holds. Agent Bricks governs the agent and everything it interacts with in a single system. Through Unity Catalog and the new AI Gateway, access to data, models, and external MCP-connected tools is managed in one place with identity enforced end to end, according to Databricks.

Agents inherit user identity through on-behalf-of token passing, meaning they can only access what the requesting user is authorized to use. The same permissions, auditing, and routing apply whether an agent is querying the lakehouse or calling an external API.

“With Agent Bricks, we’re not building one-off AI projects, we’re building an enterprise AI fabric,” said William Acosta, Head of Agentic AI Engineering at EchoStar, in Databricks’ announcement. “Interoperability, identity-first security and governance were designed from day one, so our agents behave like any other mission-critical system, not a science experiment.”

Multi-Model, Multi-Framework

Agent Bricks natively supports frontier models and coding agents including Cursor, OpenAI Codex, and Claude Code through a single API with built-in routing, fallback, and cost optimization. It also supports LangGraph and OpenAI Agents SDK for building and deploying agents, according to Databricks.

63% of Agent Bricks customers currently route tasks across two or more model families. Managed OAuth MCP Connectors handle secure connections to external services like GitHub, Atlassian, and Glean with centrally managed credentials, keeping secrets out of agent code.

Accuracy From Business Context

Databricks claims Agent Bricks achieves 70% higher accuracy than standard RAG and a 30% improvement in multi-step workflows by embedding Unity Catalog metadata (schema, business definitions, lineage, permissions, and data quality signals) directly into retrieval and planning, according to the company’s blog post.

Genie Spaces let agents reason over the semantic layer using business definitions rather than raw column names. Alvaro Martin, Senior Data Engineer at Zapier, said the platform “gives us a structured way to coordinate multiple data intelligence endpoints in a single system,” according to Databricks.

The AI Gateway

The new AI Gateway provides a unified layer for managing access to models, coding agents, and MCP-connected tools. It includes guardrails for detecting and mitigating PII exposure, unsafe content, prompt injection, data exfiltration, and hallucinations. The gateway enforces identity, permissions, and observability across every interaction, according to Databricks.

Additional releases include Agent Mode in Genie Spaces for multi-step reasoning over structured data, Knowledge Assistant for automatic document ingestion, and the CLEARS Framework for agent quality evaluation in MLflow.

Production Scale Across Verticals

Workday, Virgin Atlantic, Zapier, EchoStar, and AstraZeneca are among thousands of organizations running production agents on the platform across financial services, retail, healthcare, and technology. Teams are building agents for continuous market analysis, supply chain orchestration, automated employee service requests, and marketing campaign anomaly detection, according to Databricks.

The integration partner ecosystem includes Accenture, Deloitte, EY, Infosys, LlamaIndex, and more than a dozen others extending Agent Bricks across enterprise tooling.

The Data Company’s Agent Bet

Agent Bricks is Databricks’ answer to a question every enterprise data team is asking: who governs what happens when agents touch production data? Unlike Microsoft’s agent offerings (which start from Copilot and M365) or Salesforce’s (which start from CRM), Databricks starts from the data layer. For organizations whose agent use cases are fundamentally data-intensive, from analytics to reporting to transactional workflows, that starting point determines whether governance is bolted on or built in. Databricks built it in.