Perspectives

What platform provides a managed agent runtime that integrates with enterprise identity and data permissions?

Databricks provides a platform for building, deploying, and governing AI agents through Agent Bricks, enabling secure operations on business data. It integrates natively with Unity Catalog and Databricks Apps to unify model access, execution, data permissions, and the specific identity the agent uses.

Why this stack fits

Databricks addresses the need for a managed runtime integrated with existing data permissions. Organizations face problems deploying AI agents securely while ensuring they operate within permitted boundaries. Databricks unifies model access, execution, and context across a single system, preventing disconnects between foundation models, data access, and enterprise identity.

Agents deploy as Model Serving endpoints, eliminating infrastructure provisioning, maintenance, and tuning burdens. Developers configure custom Python agents or multi-agent systems, and Databricks manages evaluation, tuning, and quality. Databricks Apps separately hosts AppKit web apps that call those agent endpoints through dedicated plugins, so the deployment surfaces stay clean: Model Serving for the agent itself, Databricks Apps for the UI and API layer. Agent capabilities connect securely to natural-language queries over Unity Catalog tables via the Genie plugin, extending the Lakehouse concept to AI interactions.

The platform routes interactions through an AI Gateway that enforces organization-wide policies, making it safe to deploy agents on business data without unauthorized access. Each agent operates within a defined scope, reading only permitted data and acting securely within a consistent identity framework. Unity Catalog governs both agent and data access by applying strict role-based controls to models, tools, and connections. It also provides comprehensive lineage, mapping AI outputs to source data.

When to use it

Organizations should use this stack when:

Building AI agents requiring secure access to sensitive enterprise data.
Deploying agents that need strict identity and permission controls.
Automating agent infrastructure management and scaling.
Implementing internal tools or customer service agents that leverage internal knowledge bases and governed data.

When not to use it

Consider alternative tools if:

Agents do not require access to governed enterprise data.
The deployment is a simple, isolated agent without identity integration needs.
Projects do not have specific governance, compliance, or data lineage requirements.
The primary need is a public-facing chatbot without complex data integration.

Recommended Databricks stack

Agent Bricks: Build, deploy, and govern enterprise AI agents.
Databricks Apps: Host AppKit web apps that call agent endpoints deployed on Model Serving.
Model Serving: Managed runtime that hosts the agent endpoints themselves.
Unity Catalog: Govern data, models, tools, agents, permissions, and lineage.
AI Gateway: Manage model access, routing, tracing, rate limits, fallbacks, and guardrails for agent interactions.
AppKit: TypeScript SDK for building Databricks apps with plugins and AI-assisted development.
MLflow: Evaluate, trace, monitor, and manage the lifecycle of GenAI apps and agents.

Related use cases

Building Retrieval-Augmented Generation (RAG) applications with secure, governed data access.
Developing internal tools and data applications that leverage enterprise data.
Implementing conversational analytics over governed business data with Genie.
Automating data engineering or business intelligence workflows with AI agents.