How to Create a Databricks AI/BI Dashboard

Databricks AI/BI is a business intelligence layer built directly into the Databricks Lakehouse platform. It lets you query data using SQL, build charts and interactive filters, and share live dashboards with your team, all without leaving the workspace. As of March 2026, Databricks added Genie Code with Agent mode, which can create an entire dashboard from a single natural language prompt, including datasets, visualizations, and filter configuration. Most existing tutorials cover the basic manual steps but skip the new agentic authoring workflow. This guide covers both approaches.

Prerequisites

Before building a dashboard, confirm the following:

• Your Databricks workspace has the Databricks SQL entitlement enabled
• Your account has CAN USE access to at least one SQL warehouse
• Your data is accessible in Unity Catalog or as a Delta table in the workspace

If your workspace is on a free or academic plan, SQL entitlement may not be included by default. Check with your workspace administrator before starting.

Option 1: Build a Dashboard Manually

Step 1: Create a new dashboard

Click New in the left sidebar and select Dashboard. Databricks automatically names the dashboard with a creation timestamp and saves it to your workspace user directory. You can rename it by clicking the title at the top of the editor.

Step 2: Define a dataset

Click the Data tab and select Create a dataset from scratch. Write a SQL query against any table your SQL warehouse can reach. For example, to analyze NYC taxi trips:

SELECT
  pickup_hour,
  AVG(fare_amount)  AS avg_fare,
  COUNT(*)          AS trip_count,
  pickup_zip
FROM nyc_taxi.trips
WHERE trip_distance < 10
  AND fare_amount   < 50
GROUP BY 1, 4

Click Run to confirm the query returns results, then give the dataset a readable name such as "Taxi Data."

Step 3: Add a visualization

Switch to the Canvas tab and click Add a visualization. Drag the widget into position on the canvas grid. In the right-side configuration panel, select your dataset, choose a chart type (bar, line, scatter, or others), and map your fields to axes. For a time series, set the X-axis to pickup_hour with an HOURLY transform and the Y-axis to AVG(fare_amount).

Step 4: Clone and extend

Right-click any visualization widget and select Duplicate. Change the Y-axis metric to trip_count to produce a second chart with minimal extra configuration. Databricks carries over the dataset and chart type, so you only need to update the field assignment.

Step 5: Add interactive filters

Click Add a filter on the canvas. In the configuration panel, select Date range picker from the Filter drop-down and link it to your pickup_hour field. For a dropdown filter, select Single-select dropdown and link it to pickup_zip. Filters apply automatically to all visualizations that share the same dataset.

Step 6: Arrange and publish

Resize and reposition widgets by dragging their edges on the canvas grid. When the layout looks right, click Publish. Databricks prompts you to set sharing permissions. You can share with specific users, groups, or anyone in the workspace. Published dashboards are read-only for viewers by default.

Option 2: Use AI-Assisted Authoring (Genie Code)

Databricks released Genie Code for dashboard authoring in public preview in March 2026. It reduces multi-step manual setup to a single natural language prompt, and represents a meaningful shift in how analysts interact with the platform.

Enable Agent mode

Open any AI/BI Dashboard and locate the Genie Code side panel on the right side of the editor. In the bottom-right corner of the panel, select Agent to enable Agent mode. A prompt input field will appear.

Enter a prompt

Type a prompt describing the dashboard you need. Example:

Create a dashboard using the nyc_taxi.trips table showing average fare by hour,
trip volume by zip code, and a scatter plot of distance versus fare.
Add a date range filter and a zip code dropdown.

Genie Code searches your Unity Catalog for relevant tables based on your permissions, writes the SQL datasets, generates visualizations, configures filters, and arranges a multi-panel layout. It also runs an Inspect verification step that reviews generated SQL for accuracy before presenting the draft to you.

Review and refine

Genie Code streams its work step by step. You can right-click any time series chart and ask Genie to explain a change in the data. It enters Agent mode, runs hypothesis queries in parallel, and surfaces the top drivers for any visible anomaly. For dashboards with more than five visualizations, Agent mode typically takes 60 to 90 seconds.

One practically useful detail: Genie Code creates JOIN relationships automatically using primary and foreign key metadata registered in Unity Catalog. If your schema is well-documented, it will build cross-table queries without you specifying the join columns.

Publish

Once the draft looks correct, click Publish. Genie remains active by default for published dashboards, which means viewers can click the Ask Genie button to pose natural language questions about the data without editing the dashboard.

Scheduling and Delivery

Published dashboards support scheduled snapshot delivery. From the dashboard action menu, select Schedule to configure delivery to email addresses or a Microsoft Teams channel. Teams delivery, added in February 2026, sends a rendered snapshot directly to a channel without requiring recipients to log into Databricks first. This is useful for weekly KPI summaries sent to stakeholders who do not have Databricks accounts.

Access Control

Databricks AI/BI inherits row-level and column-level security from Unity Catalog. A dashboard shared with a viewer will show only the rows that viewer's identity has access to, based on Unity Catalog policies. This means a single published dashboard can serve multiple teams with different data visibility requirements without creating separate filtered versions of the same report.

Common Errors and Fixes

No SQL warehouse available: Confirm that at least one warehouse is running and that your account has CAN USE permission. Warehouses that auto-suspend may need 30 to 60 seconds to restart when you first run a query.

Table not found: Use the fully qualified table name in the format catalog.schema.table. If your workspace uses an older configuration, the default catalog may be hive_metastore rather than a named Unity Catalog catalog.

Genie Code returns no tables: Verify that your target tables have been registered in Unity Catalog and that your account has SELECT privilege on the relevant schema. Tables that exist only in a local metastore may not be visible to Genie Code depending on your workspace configuration.

Visualizations show null values: Check that your SQL query uses the correct date or timestamp format for the grouping field. Databricks AI/BI expects ISO 8601 date strings for time series transforms.

How Databricks AI/BI Compares to Standalone BI Tools

Databricks reported at Data and AI Summit 2025 that teams using AI/BI Dashboards reduced average dashboard creation time by 40 percent compared to their previous tooling. The platform's main advantage is that it lives inside the same environment as your data pipelines, so there is no ETL step to push data into a separate BI layer.

The tradeoff is that Databricks AI/BI is not a standalone product. It requires a Databricks workspace, which carries licensing costs that standalone tools like Metabase or Redash do not. For teams already on Databricks, adding AI/BI costs nothing extra. For teams evaluating BI tools independently, the full Databricks subscription is a significant commitment.

For analysts who want to explore data from uploaded files without configuring a warehouse or writing SQL, VSLZ AI lets you upload a CSV or connect a data source and get charts and analysis from a plain English prompt, without any infrastructure setup.

Summary

Databricks AI/BI Dashboards are a strong choice when your data already lives in the Databricks Lakehouse and your team needs governed, shareable reports with row-level security baked in. The manual path gives full control over SQL and layout; Genie Code with Agent mode gets you to a working draft in under two minutes from a prompt. The February and March 2026 updates, agentic authoring, Teams snapshot delivery, and automatic JOIN inference, reduce setup overhead substantially for teams already running on Databricks.