Building AI agents is one thing; making them reliably function within an enterprise is another. Databricks is tackling this challenge with the launch of Agent Bricks, its new governed enterprise agent platform. This initiative aims to bridge the gap between experimental AI agents and mission-critical business applications.
The core problem Databricks identifies isn't agent creation itself, but the operational complexities of integrating them with sensitive business data, ensuring proper permissions, and maintaining control. Agent Bricks seeks to unify data, models, and governance into a single, cohesive system.
The Enterprise Agent Challenge
Valuable AI agents are defined by their deep connection to an organization's specific data and context—customer records, internal policies, operational systems. Running these agents in production requires them to understand business context, operate under correct identities and permissions, and work across different models without vendor lock-in. This is where most current solutions fall short, often providing only fragmented pieces rather than a comprehensive platform.
Agent Bricks: A Unified Platform
Agent Bricks is designed as an end-to-end solution for building, deploying, and governing agents that operate on business data. It integrates model access, execution, governance, and context management, enabling reliable production deployments. Databricks reports that thousands of organizations already use the platform for diverse applications, from market analysis to supply chain orchestration.
Key Platform Pillars
The Agent Bricks platform is built on three foundational principles:
- Open and Multi-AI: Teams can leverage multiple model providers and frameworks through a single API, supporting frontier models and popular coding agents. This flexibility allows for routing, fallbacks, and cost optimization, with 63% of customers reportedly routing tasks across multiple model families.
- Unified Governance: Unlike systems that only govern the agent, Agent Bricks extends governance to all data, models, and external tools. Using Unity Catalog and AI Gateway, access is managed and observed centrally, with agents inheriting user identity for strict permission enforcement.
- Accuracy Through Context: Agent accuracy is enhanced by leveraging Unity Catalog metadata, including schemas, business definitions, lineage, and permissions. This contextual grounding reportedly leads to 70% higher accuracy than standard Retrieval-Augmented Generation (RAG) and a 30% improvement in multi-step workflows.
New Capabilities Announced
Databricks is also rolling out several new features to bolster the platform's capabilities:
- Custom Agents on Apps (GA): Allows building and deploying agent applications with any model or framework, featuring lifecycle support and serverless compute.
- Supervisor Agent (GA): Enables orchestration of multiple agents and tools into complex workflows.
- AI Gateway (Beta): Provides a unified layer for managing and governing access to models, endpoints, and external tools, enforcing identity and observability.
- Document Intelligence (GA): Extracts structured data from unstructured documents like contracts and reports, transforming them into queryable knowledge.
- Knowledge Assistant (GA): Ingests enterprise documents, making them accessible to agents with context-aware retrieval.
- Agent Mode in Genie Spaces: Enhances data analysis by enabling multi-step reasoning and planning over business data.
The company emphasizes that the challenge has shifted from building the agent loop to orchestrating the surrounding infrastructure: secure identity, credential management, flexible model routing, accurate business context, and comprehensive observability. Agent Bricks aims to consolidate these elements into a reliable, multi-AI, governed platform for enterprise data.
