Overview
This section is worth 12% and has four objectives. The October 2026 guide calls the feature AI/BI Genie Agents and keeps Genie spaces in brackets, so expect both names. A Genie Agent lets business users ask questions in natural language and get answers generated as SQL over a curated set of Unity Catalog data.
The objectives follow the lifecycle. First, describe its purpose, features and components. Second, create one by choosing a SQL warehouse, curating tables, views and metric views, writing domain-specific instructions, adding sample questions, and vetting example queries and functions as trusted assets. Third, assign permissions and distribute it through embedded links, external app integrations and channel apps such as Microsoft Teams or Slack. Fourth, optimise it by monitoring questions, accuracy and feedback, updating instructions and trusted assets, running benchmarks and refreshing Unity Catalog metadata.
How it's tested
Questions follow the build, share and improve lifecycle. You may be asked which component provides certified answers, how to teach Genie a business definition, what users need to ask questions, or how to measure whether accuracy improved after a change. Look for the smallest curation change that fixes the described behaviour.
-
Improve the answer
Genie misreads a term or returns wrong figures, and you choose between instructions, trusted assets, metadata comments or narrowing the data.
-
Grant access
Users cannot open a space or get errors when asking, and you identify the missing space permission, warehouse permission or data grant.
-
Measure quality
A team wants to know whether changes improved accuracy, pointing to benchmarks and the monitoring of questions and feedback.
Users ask a Genie space for active customers and get inconsistent counts because the definition needs three conditions. Adding more tables would widen the search space. A general instruction helps but can still be interpreted loosely. A parameterised example query or a Unity Catalog function registered as a trusted asset encodes the exact logic, and answers that use it are marked trusted, so that is the strongest fix.
Objectives 4
Official objectives from the Databricks outline, in outline order.
7.1 Describe the purpose, key features, and components of AI/BI Genie Agents (Genie spaces).What a Genie Agent is and what it is made of.
- What it means
- A Genie Agent is a natural language interface that lets business users ask data questions and receive answers as tables, charts and summaries, generated by translating the question into SQL. Its components are the curated data (tables, views, metric views), a SQL warehouse, instructions, sample questions, trusted assets, benchmarks and monitoring.
- What to understand
- Genie uses the Data Intelligence Engine together with Unity Catalog metadata, the instructions and example SQL that authors provide, to write queries; users can view the generated SQL. Answers run on the chosen SQL warehouse with the asking user's permissions. Genie is designed for a focused business domain, not a whole lakehouse, and it can ask clarifying questions when a request is ambiguous. It can also be enabled from AI/BI Dashboards.
- How it shows up
- Questions describe a self-service analytics goal and ask what Genie provides or which component serves a stated purpose.
- Common confusion
- Assuming Genie answers from its own training data rather than generating SQL over curated tables.
7.2 Create Genie Agents (Genie spaces) by defining reasonable sample questions and domain-specific instructions, choosing SQL warehouses, curating Unity Catalog datasets (tables, views...), and vetting queries as Trusted Assets.Building a focused, well-instructed Genie space.
- What it means
- Creating a space means choosing a pro or serverless SQL warehouse, selecting a small set of relevant Unity Catalog tables, views or metric views, writing instructions that explain business terms and rules, adding sample questions that show users what to ask, and adding example SQL queries or SQL functions vetted as trusted assets.
- What to understand
- Keep the data focused, ideally a handful of well-documented tables, since too many tables reduce accuracy. Good column comments and descriptions matter because Genie reads them. Instructions can be plain text, example SQL queries and join definitions. Trusted assets are parameterised example queries and Unity Catalog functions; when Genie answers with one, the response is labelled trusted, signalling verified logic. Prepare data with clear names and pre-joined views where needed.
- How it shows up
- A scenario asks for the best way to teach Genie a business rule or to give users verified answers.
- Common confusion
- Adding every table in a schema to improve coverage, which usually lowers accuracy.
7.3 Assign permissions via the UI and distribute Genie Agents (Genie spaces) using embedded links, external app integrations, and channel apps.Sharing Genie and the access users need.
- What it means
- Genie space permissions are granted through the Share dialog, with levels such as Can View, Can Run, Can Edit and Can Manage. Distribution options include sharing a link, embedding in other pages, calling Genie through its API from external apps, and connecting it to channel apps such as Microsoft Teams or Slack.
- What to understand
- Access to the space is not enough: users also need CAN USE on the SQL warehouse and SELECT on the underlying data, because queries run as them, so Unity Catalog row filters and column masks still apply. Authors need the right privileges on all included tables. The Conversation API lets developers integrate Genie into custom apps and workflows. Channel apps bring questions into the tools people already use.
- How it shows up
- A user can open a space but every question fails; the missing piece is data or warehouse permission.
- Common confusion
- Thinking a Genie space bypasses Unity Catalog permissions on its tables.
7.4 Optimize AI/BI Genie Agents (Genie spaces) by tracking user questions, response accuracy, and feedback; updating instructions and trusted assets based on stakeholder input; validating accuracy with benchmarks; refreshing Unity Catalog metadata.Monitoring, benchmarking and refining Genie answers.
- What it means
- Optimising is a feedback loop. The monitoring view shows the questions users asked, the responses, and thumbs-up or thumbs-down feedback, including requests for review. Authors update instructions, add trusted assets, improve metadata and then run benchmarks, a set of test questions with expected SQL or answers, to measure accuracy.
- What to understand
- Benchmarks give an objective score before and after a change, so you know whether an edit helped. Low-rated questions show where terms are misunderstood or data is missing. Refreshing Unity Catalog metadata, such as column comments, descriptions and new columns, keeps Genie aligned with the current schema. Involve subject matter experts to confirm the correct answers, then encode them as trusted assets or instructions.
- How it shows up
- Expect a scenario asking how to prove that instruction changes improved accuracy; the answer is benchmarks.
- Common confusion
- Judging accuracy from a few manual questions instead of a repeatable benchmark set.
High-yield concepts
The terms, rules and formulas this domain turns into questions most often.
-
Genie Agent (Genie space)
Natural language interface over curated Unity Catalog data.
- Remember
- Both names appear in the 2026 guide.
- Tested as
- Naming and purpose questions.
- Trap
- Treating it as a general chatbot.
-
Instructions
Text, example SQL and join guidance for Genie.
- Remember
- Teach business terms and rules.
- Tested as
- Accuracy improvement.
- Trap
- Writing vague or conflicting rules.
-
Sample questions
Starter questions shown to users.
- Remember
- Show scope and phrasing.
- Tested as
- Space creation.
- Trap
- Confusing them with benchmarks.
-
Trusted asset
Vetted example query or SQL function.
- Remember
- Answers using it are marked trusted.
- Tested as
- Verified logic scenarios.
- Trap
- Thinking any instruction makes answers trusted.
-
Benchmark
Test questions with expected answers.
- Remember
- Measures accuracy objectively.
- Tested as
- Validation questions.
- Trap
- Relying on ad hoc spot checks.
-
Monitoring and feedback
Record of user questions and ratings.
- Remember
- Shows where Genie struggles.
- Tested as
- Optimisation loop.
- Trap
- Ignoring review requests.
-
SQL warehouse for Genie
Compute that runs Genie's generated SQL.
- Remember
- Pro or serverless.
- Tested as
- Creation and access.
- Trap
- Assuming classic warehouses work.
-
Unity Catalog metadata
Table and column names, comments and descriptions.
- Remember
- Genie reads it to write SQL.
- Tested as
- Quality improvement.
- Trap
- Leaving columns undocumented.
-
Conversation API
Programmatic access to Genie.
- Remember
- Integrates Genie into external apps.
- Tested as
- Distribution questions.
- Trap
- Confusing it with dashboard embedding.
-
Channel apps
Genie in tools like Teams or Slack.
- Remember
- New distribution option in the guide.
- Tested as
- Distribution questions.
- Trap
- Expecting users to visit the workspace.
-
Focused data scope
Small set of relevant tables.
- Remember
- Fewer tables, better accuracy.
- Tested as
- Design scenarios.
- Trap
- Adding a whole schema.
Side by side
Concepts candidates mix up, compared on the point that decides the answer.
Ways to steer Genie
| Concept | Meaning | Key difference | Common trap |
|---|---|---|---|
| General instructions | Plain-text guidance on terms and rules. | Flexible but interpreted by the model. | Expected to guarantee exact logic. |
| Example SQL query | Query showing how to answer a question type. | Teaches patterns and joins. | Not vetted unless a trusted asset. |
| Trusted asset | Vetted parameterised query or function. | Answer flagged as trusted. | Confused with sample questions. |
| Column comments | Unity Catalog metadata on fields. | Improves column selection everywhere. | Overlooked in favour of instructions. |
Evaluation tools
| Concept | Meaning | Key difference | Common trap |
|---|---|---|---|
| Benchmarks | Set of questions with expected answers. | Repeatable accuracy score. | Skipped after changes. |
| Monitoring | Log of real user questions and responses. | Shows actual usage. | Treated as a score. |
| User feedback | Thumbs up or down and review requests. | Points to specific failures. | Ignored when updating instructions. |
Exam traps
Each trap is an answer that looks right. Learn the clue that should change your mind.
-
More data for better answers
- What candidates pick
- Add all tables from the schema.
- Why it looks right
- Broader coverage sounds helpful.
- Why it's wrong
- Extra tables give Genie more ambiguous choices; accuracy improves with a small, well-documented set of relevant tables.
Clue to watch for Genie picks the wrong table or column.
-
Space access is enough
- What candidates pick
- Grant Can Run on the space only.
- Why it looks right
- The space is what users open.
- Why it's wrong
- Queries run as the user, so they also need CAN USE on the warehouse and SELECT on the curated tables.
Clue to watch for Users open the space but questions fail.
-
Instructions for exact logic
- What candidates pick
- Write a text instruction defining churn.
- Why it looks right
- Instructions are the obvious lever.
- Why it's wrong
- Text can still be interpreted loosely; a vetted query or function as a trusted asset encodes the exact definition and marks answers trusted.
Clue to watch for A precise, multi-condition business definition.
-
Bigger warehouse for accuracy
- What candidates pick
- Increase the SQL warehouse size.
- Why it looks right
- Compute fixes many problems.
- Why it's wrong
- Warehouse size affects speed, not whether Genie writes the correct SQL; accuracy depends on curation, metadata and instructions.
Clue to watch for Answers are wrong, not slow.
-
Spot checks as validation
- What candidates pick
- Ask a few questions after each change.
- Why it looks right
- It is quick.
- Why it's wrong
- Spot checks are inconsistent; benchmarks rerun the same questions with expected answers and give a comparable accuracy score.
Clue to watch for Prove the change improved accuracy.
-
Genie bypasses governance
- What candidates pick
- Genie shows all rows to every user.
- Why it looks right
- It runs centrally.
- Why it's wrong
- Answers run with each user's permissions, so row filters and column masks in Unity Catalog still apply.
Clue to watch for Users in different groups see different results.
Answer framework
Use these steps for Genie Agent design, access and quality questions.
- Locate the lifecycle stage: describing Genie, building a space, sharing it, or improving answer quality.
- When building, keep a focused set of documented tables, choose a pro or serverless warehouse, and add instructions and sample questions. Too many tables or vague column names make Genie guess more often.
- For precise business logic or verified answers, add a vetted example query or SQL function as a trusted asset.
- For access problems, check space permission, warehouse CAN USE and Unity Catalog SELECT, remembering queries run as the user. A user without SELECT on a table gets an error even inside a shared space.
- For quality, review monitoring and feedback, update instructions or metadata, then rerun benchmarks to confirm improvement.
Practice question
Want more like this? The Databricks DA Assoc practice test has questions for every domain.
Checklist
0 of 11
Tick each one you can do without notes. Your ticks stay in this browser.
Study strategy
Learn this section by building a Genie space, because the exam scenarios mirror the authoring workflow. Use two to four related tables, for example orders, customers and products, and first add clear table and column comments in Catalog Explorer.
Create the space on a serverless or pro warehouse, add the tables, write three instructions that define business terms such as active customer or net revenue, and add five sample questions. Ask ten questions of your own and read the generated SQL for each. Where an answer is wrong, decide whether the cause is metadata, instructions or missing logic, and fix it. Turn at least one definition into a parameterised example query and mark it as a trusted asset, then ask a matching question and look for the trusted label.
Next, create a benchmark set of eight questions with expected answers and run it before and after a change. Open the monitoring view and give feedback on some answers. Finally, read the documentation pages on Genie permissions, the Conversation API and channel integrations so you can answer distribution questions. Because the guide now says Genie Agents, check that your practice material uses both names.
FAQ
How much of the Databricks Data Analyst exam covers Developing, Sharing, and Maintaining AI/BI Genie Agents?
This section is 12% of the exam, roughly five of the 45 scored questions, per the official certification page.
Are Genie Agents and Genie spaces the same thing?
Yes. The October 2026 guide uses AI/BI Genie Agents with Genie spaces in brackets, so expect either name in questions.
What are trusted assets?
Trusted assets are vetted, often parameterised example SQL queries and Unity Catalog functions. When Genie answers with one, the response is labelled trusted.
Which warehouses can a Genie space use?
Genie requires a pro or serverless SQL warehouse. Serverless is generally recommended for fast start-up.
Do Genie answers respect row filters and column masks?
Yes. Generated queries run with the asking user's permissions, so Unity Catalog controls apply.
How is Genie accuracy measured?
With benchmarks: test questions with expected answers that you rerun after changes to compare accuracy.
Other Data & Analytics certifications
Where the same subject shows up in other exams. Cards marked "Same subject" open the matching domain.
Test your Domain 7 knowledge
Test whether you can build, share and improve Genie Agents, including trusted assets, permissions and benchmarks, with timed practice questions.
Keep going
- Training Training Built on your Weak areas
- Practice test Questions by domain, with explanations
- Mock exam Full length, timed like the real exam
- Readiness test Find your weakest domain in minutes
- Cheat sheet Key terms and formulas on one page
- Ask the Databricks DA Assoc AI tutor Get any concept explained your way