How-to · Data · Databricks Genie · Oct 19, 2026

Set up a Databricks Genie Agent on Unity Catalog tables, and pass a benchmark before sharing it

Databricks Genie Agents answer plain-language questions on Unity Catalog tables and show the SQL they ran. A wrong join still looks confident. This page shows how to set up one Agent with example SQL and pass a benchmark before anyone else uses it.

How-to · Oct 19, 2026 · Databricks Genie · Databricks Genie vs Snowflake Cortex

Who this is for

Data and ops teams whose trusted tables already sit in Unity Catalog. The job here is a Genie Agent benchmark before share. For classifying ticket text with Snowflake Cortex and tracking credits — a different job — see Snowflake Cortex classify ticket text.

How to do it

This is a how-to, not a ranking of lakehouse assistants. Per the Databricks Genie review, a Genie space is now called a Genie Agent; you add Unity Catalog tables, example SQL, read generated queries, run benchmarks, and may use Genie One. The ground rule: no share until benchmark questions pass with a data owner sign-off.

Simple mark for the Databricks how-to
Simple mark for this how-to. The diagram below is the three-step path, not a product screenshot.
Diagram of three steps: Pick tables (Unity Catalog); Example SQL (+ benchmarks); Owner signs (then share)
Simple workflow for this how-to: pick three to five Unity Catalog tables, then add example SQL and write benchmark questions, then read generated SQL on failures; share only after pass. A plain diagram, not a product screenshot.

1. Pick three to five Unity Catalog tables

What it is: Choose a small, governed set of tables with clear owners — not the whole catalog.

  • Name tables and owners.
  • Prefer curated marts over raw landing zones.
  • Grant the Genie service principal only what it needs.
  • Write the business questions this Agent should answer.
  • Keep Genie One extras out of the first pilot if you do not need them.
Concrete example (revenue ops mart)

The data owner picks orders, customers, and product tables in the finance mart. Raw event logs stay out.

Try this

List three to five table names and the owner email before you create the Agent.

2. Add example SQL and write benchmark questions

What it is: Add example SQL for the joins you trust, then write benchmark questions with expected answers before anyone chats casually.

  • Add example SQL that shows the correct grain and filters.
  • Write at least ten benchmark questions with expected numbers or rows.
  • Include two questions that should fail closed.
  • Freeze the benchmark set.
  • Assign a grader who can read SQL.
Concrete example (monthly bookings questions)

Examples show how bookings exclude internal test accounts. Benchmarks include “bookings last month” and a trick question about a retired product code.

Try this

Save example SQL and ten benchmark questions before open chat.

3. Read generated SQL on failures; share only after pass

What it is: Run the benchmarks, open the generated SQL on every miss, fix instructions or examples, and share only after the data owner signs the pass.

  • Open SQL on every wrong answer.
  • Fix example SQL or instructions; do not shrug.
  • Re-run the full benchmark after changes.
  • Data owner signs the pass record.
  • Share the Agent only then.
Concrete example (data owner sign-off)

Two benchmarks fail on test-account leakage. Examples are fixed, benchmarks re-run clean, and the owner signs before Finance gets access.

Try this

Do not share the Agent until the owner signs a passed benchmark sheet.

Pricing context (October 2026)

This how-to isn’t a price sheet. Per the Databricks Genie review and Databricks’ Genie pricing page (checked Oct 19, 2026), Genie One and Genie Agents LLM usage is free until January 31, 2027; afterward each user gets 150 free DBUs per month (Databricks notes 150 DBUs = $10.50 in US East). SQL warehouse compute is billed separately. Dated October 2026.

Checked Oct 19, 2026 on the vendor’s own pricing page: Databricks’ Genie pricing page lists Genie One and Genie Agents LLM usage free until January 31, 2027, then 150 free DBUs per user per month (Databricks notes 150 DBUs = $10.50 in US East). SQL warehouse compute is billed separately.

Uncertainty: The only prices this site sells here are the optional $9 AI toolkit spreadsheet and $19 AI toolkit audit.

When this helps (honest take)

Useful when your trusted tables are in Unity Catalog and a SQL-literate owner will grade benchmarks.

Not useful when you wanted chat without reading SQL, or tables are ungoverned.

Related options (educational)

Where to look next for related jobs (educational, not a ranking)
NeedWhere to lookWhy
Databricks GenieDatabricks GeniePrimary path
Genie vs CortexDatabricks Genie vs Snowflake CortexCompare patterns
Snowflake Cortex classifySnowflake Cortex classify ticket textA different job

FAQ

Is Genie space the same as Genie Agent?

Per our review, Databricks now calls Genie spaces Genie Agents.

Is Genie free right now?

Databricks’ Genie pricing page listed LLM usage free until January 31, 2027 when we checked Oct 19, 2026. Warehouse compute is separate. Confirm on Databricks’ page.

Optional next step

If you want a one-page toolkit map for the tools you already pay for, use the optional $9 AI toolkit spreadsheet. If you want a short review of seat waste and unused tools, use the optional $19 AI toolkit audit. Neither is required to follow this how-to.

Affiliate disclosure: Some outbound product links may be affiliate links. We may earn a commission if you subscribe. Prices and features change — always verify on the vendor site. Full policy: disclosure.

Next steps