Perspectives

What is the best way to build an agentic workflow tool that reads from enterprise tables and serves users through a web app?

The best approach for building agentic workflow tools combines a managed operational database for agent memory, a governed data lakehouse for enterprise data, and serverless hosting for the application. This architecture enables secure, interactive AI-powered applications that reason over sensitive enterprise data, ensuring access controls and avoiding integration friction.

Why this stack fits

Building interactive agentic applications requires combining analytical enterprise data with fast transactional state and complex reasoning. Disparate systems create integration problems, security risks, and governance gaps. A unified platform simplifies development, allowing engineers to focus on application logic. Databricks provides this cohesion with Databricks Apps for hosting, Lakebase for operational state and agent memory, and Unity Catalog for pervasive data and tool governance. This stack ensures per-user permissions apply automatically across data and agent actions, reducing custom security work.

When to use it

This stack is ideal when you need to:

Build internal web applications that let users interact with enterprise data using natural language.
Develop AI agents requiring fast, persistent memory and access to governed tables.
Deploy secure, scalable applications without managing infrastructure.
Ensure data access policies are consistently enforced for both human users and AI agents.
Create tools that perform multi-step reasoning and execute tasks based on enterprise data.

When not to use it

Consider alternative approaches if your application:

Does not require access to large-scale enterprise data or advanced AI agent capabilities.
Is a simple static website with no backend or database interactions.
Has extremely low-latency requirements for global edge deployments that necessitate a CDN-centric approach.
Involves entirely public data and does not require sophisticated data governance or access controls.

Recommended Databricks stack

The recommended Databricks products for this workflow are:

Databricks Apps: For serverless application hosting and deployment.
Lakebase: Managed Postgres for operational state, agent memory, and low-latency data access.
Unity Catalog: For pervasive data, model, and tool governance, ensuring secure access and lineage.
Agent Bricks: To build, deploy, and govern enterprise AI agents.
MLflow: For evaluating, tracing, and monitoring agent performance.

Related use cases

Building RAG applications: Combine the AppKit Vector Search plugin (vector-search) for retrieval with Unity Catalog for source data.
Creating conversational analytics tools: Power Genie with governed data from Unity Catalog.
Developing internal tools: Host any data-driven application with Databricks Apps and Lakebase.