Databricks tackles AI sprawl with Agent Bricks
Databricks has introduced a shared infrastructure framework, featuring Agent Bricks, to help enterprises scale autonomous AI agents without creating fragmented data and security setups.
Databricks is addressing the growing challenge of AI sprawl by offering a centralized infrastructure designed to manage, govern, and scale autonomous agentic applications. As enterprises transition from simple question-answering bots to active agents that execute complex workflows, they often build redundant integrations and inconsistent policies. To prevent this fragmentation, Databricks has introduced a foundation comprising Agent Bricks, Omnigent, and Unity Gateway to provide shared choice, context, and control across all enterprise agents.
For practitioners, this architecture simplifies how agents access governed enterprise data. Instead of rebuilding business definitions for every new application, developers can leverage a shared context layer. This layer utilizes Unity Catalog to govern access to data and AI assets, while the Genie Ontology provides a unified understanding of business concepts. Additional tools like Document Intelligence, AI Search, and Agent Memory extend this context, allowing different agents to reuse the same secure business definitions.
Managing diverse models and frameworks is simplified through Omnigent and Unity Gateway. Omnigent acts as a common layer above different agent harnesses, allowing developers to compose and switch between them with minimal code rewrites. Meanwhile, Unity Gateway serves as a centralized control plane that manages access, smart routing, budgets, and rate limits across models, agents, and tools. For secure execution, Databricks Sandbox provides an isolated environment with restricted permissions, while MLflow integration centralizes telemetry, tracing, and evaluation to help developers debug and optimize agent behavior.
Ultimately, this unified approach shifts the developer's focus from infrastructure maintenance to application logic. By using the Agent Bricks command-line interface, developers can quickly deploy agents that automatically inherit enterprise-grade security, model capacity controls, and operational visibility. This prevents the creation of isolated, hard-to-maintain AI stacks and allows organizations to scale their agent fleets safely and consistently.
This is our own summary of reporting by Databricks AI
