Building Trusted AI Agents: New Capabilities to Choose, Govern, and Scale with Confidence

In this wave of GenAI, competitive advantage won’t come from the models you rent—it will come from the data you own. Yet most organizations struggle to operationalize their data into production-grade agents because they lack the quality, governance, and flexibility to scale. 

Databricks Agent Bricks closes this gap, enabling every team to realize data intelligence with agents that understand, reason, and act on structured and unstructured information. With built-in evaluation through MLflow, unified governance via Unity Catalog, and open support for any model or framework with AI Gateway, Agent Bricks helps enterprises like AstraZeneca and Edmunds move from pilots to production with confidence.

This week, we’re introducing new capabilities across Agent Bricks, advancing each of its three pillars of trust: accuracy, governance, and openness. Welcome to another Week of Agents.

“With Agent Bricks, our teams were able to parse through 400,000+ clinical trial documents and extract structured data points, without writing a single line of code. In just under 60 minutes, we had a working agent that could transform complex, unstructured data into usable analytics. Agent Bricks (Information Extraction) is a valuable tool to drive from data to insights quickly!” — Joseph Roemer, Head of Data & AI, Commercial IT, AstraZeneca

Build Any Agent—Your Way
Agent Bricks is open by design, giving teams the flexibility to build high-quality agents on governed data and trusted MCP servers.

• Custom-Code Agents. Build with any framework, MCP server, or model—GPT-5, Claude Sonnet, Gemini, Llama—and run on Databricks’ serverless infrastructure. Every interaction is automatically traced, evaluated, and governed through MLflow observability and AI Gateway.
• Declarative Agents. Enable technical and non-technical teams to build and refine agents through natural language and pre-configured templates for tasks such as information extraction, knowledge assistance, and multi-agent coordination—with integrated evaluation for continuous improvement.
• Agentic AI Functions. Apply functions such as ai_parse_document directly on your documents to parse, enrich, and embed unstructured content at scale—automating manual workflows and revealing insights that were once out of reach.

“Databricks gives us a secure, governed foundation to run multiple models, like GPT-5, Claude Sonnet, and Llama, and switch providers as our needs evolve, all while keeping costs in check. That flexibility lets us automate review moderation and improve content quality faster, so car shoppers get trusted insights sooner.” —  Greg Rokita, Vice President of Technology, Edmunds

New in Agent Bricks

To help enterprises move from pilots to production, we’re deepening Agent Bricks across its three pillars of trust:

• Accuracy. MLflow for Agent Quality and Observability (GA) traces every agent interaction, scores outputs with custom judges, and provides an app for collecting subject-matter-expert feedback—making MLflow the standard for evaluating agents across any environment.
• Governance. With AI Gateway (GA) and MCP in Databricks Marketplace (Public Preview), enterprises can govern every model and external MCP server through Unity Catalog. All model activity is logged, rate-limited, and auditable to meet compliance requirements, while launch partners—including Glean, You.com, Dun & Bradstreet, Moody’s, and S&P Global—expand access to trusted data and tools.
• Openness and Interoperability. Agent Bricks supports any model, framework, or MCP server so enterprises can compose multi-step workflows across internal and external systems. Supervisor Agent (Beta) orchestrates complex workflows across agents and tools, and ai_parse_document (Public Preview) extracts insights from unstructured content—all in one governed environment.

"Databricks’ ai_parse_document reduces configuration overhead, enabling data scientists to spend less time on setup and more time advancing complex, customer-focused solutions." —  Meiling He, Sr. Data Science Manager, Rockwell Automation

What You Can Build with Agent Bricks
Agent Bricks powers a range of agent patterns—from data copilots to production automations—helping teams turn enterprise data into intelligence and action.

• Knowledge Assistants: Reason over structured and unstructured data to answer domain-specific questions, such as an HR chatbot that explains company policies or a technician assistant that searches manuals to resolve issues faster.
• Document Intelligence: Parse PDF contracts or reports to convert unstructured content into structured data for downstream analysis or integration with Knowledge Assistants.
• Process and Workflow Agents: Execute governed actions such as updating records, filing tickets, or triggering pipelines—for example, identifying churn risks and automatically sending retention offers.
• Product Intelligence – Embed reasoning or content generation into applications, such as a personalized health assistant and appointment scheduler. 

Each agent type can be implemented using custom code, declarative workflows, or agentic AI functions, giving teams the flexibility to start simple, scale securely, and standardize over time.

“MCP marks an important step toward open, interoperable enterprise AI - something that has been core to our approach at Glean from day one. The new MCP Catalog in Databricks Marketplace is an important part of that evolution. Through the MCP Catalog in Databricks Marketplace, customers can now extend Glean’s Enterprise Graph to build work agents that reason over governed data and unstructured knowledge, powering use cases like customer 360, renewal risk, and campaign insights with confidence.” — Emrecan Dogan, Head of Product, Glean

"The new MCP catalog in Databricks Marketplace allows customers to provision our high-quality web index as a governed tool with a single click. When combined with Agent Bricks, this immediately empowers agents to make real-world decisions grounded in the freshest information, fundamentally accelerating the journey from data to action." —  Bryan McCann, Co-Founder and CTO, You.com

The Path to AI Agent Production Starts Here

Agent Bricks establishes the enterprise standard for building, governing, and scaling trusted AI agents.

Stay tuned throughout the Week of Agents as we publish deep-dive blogs on each of these launches.

• Join us on 11/11 for “The Future of AI: Build Agents that Work” with Sam Altman and Ali Ghodsi.
• Watch the Demo: Learn how Agent Bricks brings accuracy, openness, and governance together in action.
• Govern Your Ecosystem: Explore the new MCP Catalog in Databricks Marketplace, integrated with AI Gateway for governed model access.
• Unlock Your Data: Start using ai_parse_document to transform your documents into structured data.

Build Actionable Agents: Leverage the Agent Bricks Supervisor Agent to deploy multi-step, governed workflows.