An AI/BI Genie space is a curated, domain-specific chat interface where business users ask questions of Unity Catalog data in plain language and get back read-only SQL, a results table and a chart. The Databricks Data Analyst Associate exam gives it four objectives: what a space is made of, how to create one, how to share it, and how to keep it accurate. Since July 2026 the docs call the same thing a Genie Agent.
Last updated September 2026.
The exam guide is from October 2025 and uses the names of that time. During 2026 Databricks renamed three things around Genie, and the questions still use the guide's terms:
| Exam guide says | 2026 docs say | Since |
|---|---|---|
| Databricks Assistant | Genie Code (the coding assistant in notebooks, the SQL editor and the dashboard editor) | 11 March 2026 |
| Databricks One (business-user interface) | Genie One | June 2026 |
| AI/BI Genie space | Genie Agent | July 2026 |
Dates are from the Databricks platform and AI/BI release notes, which state that the capabilities were unchanged by the renames.
A space is the unit of curation. Its parts, in the docs' own terms:
Genie generates one read-only SQL statement per question, runs it on the space's warehouse, and returns the result with an auto-generated visualization. It does not interpret results or recommend actions, and in chat mode it works with structured data only.
Compute is the author's; data is the user's. The generated SQL runs on the warehouse using credentials the author embedded when choosing it, so end users need no warehouse permission. Unity Catalog evaluates data access as the asking user, never the author, so row filters and column masks apply per user, and a question about data the user cannot read returns an empty response.
Almost every sharing scenario on the exam resolves through that rule. A user who gets an empty answer while colleagues in the same group get results is missing SELECT on a table. A space whose author leaves the company stops running until another CAN EDIT user reselects the warehouse, because the embedded compute credentials were the author's.
Creation is short: New, choose datasets, choose a pro or serverless warehouse (serverless recommended), configure. Curation is the work, and Databricks gives it a precedence: SQL expressions for common business terms, example SQL queries for complex questions, and text instructions only as a last resort, because text is global context that dilutes as it grows. Sample questions appear on the chat landing page to show users what the space can answer. A rule that must be applied exactly (finance's definition of net revenue) becomes a trusted asset, not an instruction. Limits worth knowing: 100 instructions and 200 knowledge store snippets per space.
The loop the exam wants: start small, watch, add context, measure. The Monitor tab (CAN MANAGE) lists questions, ratings and review requests. User feedback (Yes, Fix it, Request review) is visible to managers but Genie never learns from it automatically; an editor turns it into synonyms, joins, examples or trusted assets. Benchmarks (up to 500 per space) run each question as a new conversation and score it Good, Bad or Manual review needed against the reference SQL, giving one accuracy number for the set; CAN EDIT can run them. And because the space's default descriptions are the Unity Catalog comments and declared primary and foreign keys become join relationships automatically, fixing metadata upstream helps every space, dashboard and analyst at once.
A Genie space is a curated natural-language chat interface over a small set of Unity Catalog datasets: users ask questions in plain language and Genie returns read-only SQL, a results table and a visualization. Since July 2026 the Databricks docs call it a Genie Agent.
Viewers need CAN VIEW or CAN RUN on the space (the two are equivalent for spaces) plus SELECT on every table the space uses, because data access is evaluated as the asking user. Curators need CAN EDIT, and monitoring, embedding and permission changes need CAN MANAGE.
Trusted assets are parameterized example SQL queries and Unity Catalog SQL functions whose logic an author has verified. When a question matches one, Genie returns a verified answer using that exact logic instead of generating new SQL.
No. Thumbs up, Fix it and Request review are visible to CAN MANAGE users as signals, but Genie does not change its behavior automatically. An editor reviews the flagged exchanges, adds context or trusted assets, and re-runs benchmarks to measure the change.
Yes. Databricks renamed Genie spaces to Genie Agents in July 2026 and states the capabilities did not change. The Data Analyst Associate exam guide, dated October 2025, still uses the term Genie space.
Chapters 35 to 38 of the Data Analyst Associate course cover what a space is made of, creating one, sharing it and optimizing it, with a quiz in each.
Open chapter 35 →