What you will be able to do
- Explain what an AI/BI Genie space is for and who uses it
- Contrast Genie spaces with AI/BI dashboards inside Databricks AI/BI
- Describe the inputs Genie combines to turn a natural-language question into SQL
- Tell apart the questions a Genie space can answer from the ones it cannot
Key concept
Genie space (now called Genie Agent) — A curated, natural-language chat interface over Unity Catalog data. Business users type questions in plain language, and Genie answers with a generated read-only SQL query, a results table and, where useful, a visualization, all grounded in the datasets and guidance that a data analyst has curated.
1.What a Genie space is for
The exam guide calls them Genie spaces. The current Databricks documentation calls them Genie Agents and says plainly that Genie Agents were formerly known as Genie Spaces. Both names refer to the same thing, and this lesson uses them interchangeably.
A Genie space exists to let business teams get answers from their data without writing SQL. The documentation describes it as a feature that "allows business teams to interact with their data using natural language." The work is split between two roles. Domain experts, usually data analysts, set up the space with datasets, sample queries and text guidelines. Business users then ask questions and get back answers and visualizations.
The end-user documentation names three key capabilities. Users can ask natural-language questions that query enterprise data directly. They get relevant responses derived from organizational data. And they get self-service for questions that fall outside existing dashboards and reports, without needing advanced tools or expertise. That third capability is the real reason a Genie space exists: it covers the long tail of ad-hoc questions that nobody built a chart for.
As an example, the documentation describes a sales manager at a bakery who wants to see the top-selling product over time. They type the question in plain language and Genie produces the visualization automatically.
Checkpoint 1 of 5· Check yourself
Which statement best describes the purpose of an AI/BI Genie space?
Genie spaces are self-service, natural-language Q&A over curated data. Fixed charts describe dashboards, and code assistance is a separate product (Genie Code).
“Genie Agents is a Databricks feature that allows business teams to interact with their data using natural language.”Source: docs.databricks.com
2.Where Genie spaces sit in Databricks AI/BI
Databricks AI/BI has two complementary experiences. AI/BI dashboards handle predefined analytics: a fixed set of business questions answered with visualizations, cross-filtering and scheduled PDF snapshots. Genie spaces are conversational workspaces that can answer a much broader range of questions and improve over time through human feedback and configuration by the data team.
This split shows up on the exam. If a question describes a known, recurring set of KPIs, the answer is a dashboard. If it describes users asking unpredictable follow-up questions in their own words, the answer is a Genie space.
| Aspect | AI/BI dashboards | Genie spaces (Genie Agents) |
|---|---|---|
| Questions answered | A fixed set of business questions | A much broader set of business questions |
| Interaction | Predefined visualizations, cross-filtering | Natural-language chat; adaptable visualizations; asks for clarification when needed |
| How it improves | Author edits the dashboard | Human feedback and data team configuration |
The two experiences also link up. Published dashboards include an Ask Genie button. When you publish a dashboard, Databricks automatically generates a companion Genie space from the dashboard's datasets and visualizations, so viewers can ask follow-up questions without leaving the dashboard. Data teams keep Genie reliable with tools such as instructions, trusted assets, confidence voting and quality monitoring.
Checkpoint 2 of 5· Exam question
A retail analytics team has built an AI/BI dashboard that shows weekly sales by region, but store managers keep asking follow-up questions that the fixed dashboard filters cannot answer, such as comparing arbitrary product bundles across custom date ranges. Which capability should the team introduce to let managers explore the underlying data conversationally instead of waiting for a new dashboard widget?
Correct answer: B — An AI/BI Genie space configured over the curated sales tables, so managers can type natural-language questions and get on-demand SQL-backed answers beyond the dashboard's fixed filters.
- A. Adding another dashboard tab with preconfigured charts still only covers the specific bundle and date-range combinations someone anticipated when building it, not the arbitrary new questions a conversational interface can handle.
- B. This is correct because a Genie space lets business users ask ad hoc natural-language questions over curated Unity Catalog data and receive generated SQL answers, which is exactly the open-ended exploration a fixed dashboard cannot provide.
- C. A faster refresh schedule changes how current the data is, not how flexibly users can query it, so it does not address the need for ad hoc, unanticipated questions.
- D. An external share link only changes who can view the same fixed charts; it does not add any new exploratory or conversational query capability.
Publishing an AI/BI dashboard automatically generates a companion Genie space from the dashboard's datasets and visualizations. Viewers open it with the Ask Genie button.
Sources3
3.How Genie turns a question into an answer
Genie is not one large language model answering from memory. It uses a compound AI system, which means several interacting components work together to interpret the business question and produce an answer. When a user submits a question, Genie parses it, identifies the relevant data sources and filters the available context down to what is most relevant: example SQL queries, table and column metadata, and chat history.
The main output is SQL. Genie picks relevant names and descriptions from annotated tables and columns and turns the question into an equivalent SQL query. It then returns the generated query along with the results table.
| Component | What it contributes |
|---|---|
| Unity Catalog table metadata | Table names, descriptions, and defined primary key (PK) and foreign key (FK) relationships |
| Column names and descriptions | Genie filters for the relevant columns |
| Knowledge store context | Agent-specific metadata added by authors; does not alter Unity Catalog metadata |
| Example SQL queries | Relevant examples selected from the space's SQL Queries |
| SQL functions | All SQL functions added to the space |
| Instructions | Plain-text notes provided as General instructions |
| Prompt and response history | The current chat; the oldest parts are dropped if token limits require it |
Two facts about execution are worth remembering. In most cases the generated query runs on the space's SQL warehouse, which handles concurrency and scale, and Genie retries automatically when needed. The generated queries are also always read-only, so a Genie space never modifies your data. When Genie cannot produce an answer, it can ask follow-up questions to clarify before it responds.
Some table details, such as the owner and table size, are not included by default. To make them available, use views from the Unity Catalog information schema.
Checkpoint 3 of 5· Check yourself
A business user asks a Genie space to "remove last month's duplicate orders." What does the documentation tell you about the SQL Genie generates?
Genie runs its generated SQL on the space's SQL warehouse, and those queries are always read-only, whatever the user asks for.
“Generated queries are always read-only.”Source: docs.databricks.com
Sources1
4.What Genie can and cannot answer
Genie's text-to-SQL model uses chain-of-thought reasoning. It first identifies the relevant columns and instructions, then plans the SQL, then combines the parts into a single query. This makes it good at precise filters and nuanced data questions. It does not interpret results or offer advice, so the best questions are specific and data-driven.
| Genie can answer | Genie cannot answer |
|---|---|
| What were the sales this quarter? | Why were sales low this quarter? |
| Which customers generated the most revenue? | How can I get more revenue from Customer M? |
| What were the top-performing marketing campaigns in 2024? | How can I improve my marketing strategy? |
For exploratory questions that look into causes and patterns, the documentation points to Agent mode. Agent mode builds a research plan, runs multiple SQL queries, refines its approach and returns a report with citations and visualizations.
There are three more boundaries to know: - Structured data only in Chat mode. Standard chat cannot answer questions about PDFs, Word documents or other files. Querying files in Unity Catalog volumes requires Agent mode. - Context is per thread. Earlier questions in a thread inform later ones, so "Only for July 2024" can refine a previous answer. That context does not carry over to other chats. - Non-deterministic output. Like other LLMs, Genie can give different outputs for the same prompt.
Checkpoint 4 of 5· Exam question
A data analyst is setting up a new Genie space for the finance team and needs to define the compute that will actually execute the SQL Genie generates when it answers a question. Which component must the analyst configure as part of the Genie space setup to satisfy this requirement?
Correct answer: D — A SQL warehouse assigned to the Genie space, since Genie compiles natural-language questions into SQL that must run on a warehouse before an answer is returned.
- A. Genie answers questions by generating and running SQL against live data on demand, not by relying on a scheduled batch job that precomputes answers ahead of time.
- B. Genie spaces are built on Databricks SQL and execute generated SQL statements, not notebooks on an interactive all-purpose cluster, so a cluster policy is not the compute that answers questions.
- C. There is no separate model serving endpoint that generated SQL is routed through for scoring; the SQL warehouse assigned to the space is what executes the query and returns results.
- D. This is correct because a Genie space must be associated with a SQL warehouse, and that warehouse is the compute that runs the SQL Genie generates in response to each natural-language question.
Checkpoint 5 of 5· Check yourself
In a new chat, a user asks a follow-up that depends on a definition they gave Genie in a different chat yesterday. Genie seems to have forgotten it. Why?
Thread context applies within one conversation. To make a definition permanent, authors add it to the space's configuration rather than relying on chat history.
“This context does not carry over to other chats.”Source: docs.databricks.com
Exam traps
Each one states something that sounds right. Open it to see what is actually true.
1.A standard Genie chat can explain why a metric changed or recommend business actions.Why is that wrong?
Standard Genie translates questions into SQL. It does not interpret results or offer recommendations. Agent mode is meant for exploratory 'why' questions.
Covered in What Genie can and cannot answer
2.Standard Genie chat can answer questions about PDFs and Word documents stored in the workspace.Why is that wrong?
In Chat mode, Genie works only with structured data. Unstructured files in Unity Catalog volumes need Agent mode.
Covered in What Genie can and cannot answer
Sources
Every claim above is drawn from one of these pages, quoted as it was written on the date shown.
- 1.
“Genie Agents were formerly known as Genie Spaces.”
↩︎ What a Genie space is for“Domain experts configure Genie Agents with datasets, sample queries, and text guidelines to help Genie translate business questions into analytical queries.”
↩︎ What a Genie space is for“Genie Agents use a compound AI system to interpret business questions and generate answers.”
↩︎ How Genie turns a question into an answer“In most cases, Genie generates a SQL query that runs on the agent's SQL warehouse.”
↩︎ How Genie turns a question into an answer“If Genie can't generate an answer, it can ask follow-up questions to clarify before providing a response.”
↩︎ How Genie turns a question into an answer“In Chat mode, Genie works with structured data only.”
↩︎ What Genie can and cannot answer“In Chat mode, Genie works with structured data only.”
↩︎ Exam trap 2“Genie Agents is a Databricks feature that allows business teams to interact with their data using natural language.”
↩︎ Checkpoint“Generated queries are always read-only.”
↩︎ Checkpoint - 2.
“Supports self-service to address questions beyond the scope of dashboards and reports, without requiring advanced tools or expertise.”
↩︎ What a Genie space is for“Context from earlier conversations in a thread is used to inform later conversations in the same thread.”
↩︎ What Genie can and cannot answer“Genie, like other large language models (LLMs), can exhibit non-deterministic behaviors.”
↩︎ What Genie can and cannot answer“However, it does not interpret query results or offer recommendations.”
↩︎ Exam trap 1“However, it does not interpret query results or offer recommendations.”
↩︎ Prediction“This context does not carry over to other chats.”
↩︎ Checkpoint - 3.https://docs.databricks.com/aws/en/ai-bi/conceptsOfficial docs
“Genie Agents can answer a much broader set of business questions than dashboards, and improve over time through human feedback and data team configuration.”
↩︎ Where Genie spaces sit in Databricks AI/BI“When you publish a dashboard, Databricks automatically generates a companion Genie Agent based on your dashboard's datasets and visualizations.”
↩︎ Where Genie spaces sit in Databricks AI/BI“Data teams can configure Genie Agents using tools like instructions, trusted assets, confidence voting, and quality monitoring to ensure reliable, well-governed responses.”
↩︎ Where Genie spaces sit in Databricks AI/BI
Also cited
- https://docs.databricks.com/aws/en/genie-agentsOfficial docs
“A Genie Agent is a domain-specific natural-language chat interface in Databricks where users ask questions of their data”
↩︎ Key concept