Anyone can ask. Whether the answer is right is up to you.
Genie is where people who do not write SQL meet your data: governed dashboards, questions asked in plain English and answered from curated tables, and the shared metric definitions that keep two people asking the same thing from getting two different numbers. Databricks publishes no accuracy figure for it, and that is the right place to start — because what decides accuracy is not the model, it is the curation underneath.
Four things wearing one name.
Databricks has consolidated a lot under the word Genie over the past year, including renaming what was called Databricks One. Here is what each piece actually is, because the names are used interchangeably in places and it matters which one you are buying a rollout of.
AI/BI Dashboards
Dashboards and visualisations built on analytical datasets, with AI assistance through the build. The familiar artefact: a page of charts someone owns, certifies and shares, embeddable in another application.
Genie Spaces, now Genie Agents
Domain-specific environments where the data team configures the trusted tables, the metrics and the business rules that answers are drawn from. This is the artefact that decides quality, and the one that needs an owner. Databricks reports customers created over 1.5 million of them in 2026.
Genie One
The simplified interface for business users, previously called Databricks One, where dashboards, questions and Databricks Apps sit together. Available at workspace level for members with the consumer entitlement, and at account level as one view across every workspace — Databricks says unified login scales past 100,000 users.
Genie Code
The technical counterpart: AI-assisted SQL authoring and fixes, pipeline drafting, and the agentic code converter used in warehouse migrations. Same brand, entirely different audience.
The widened Genie can answer beyond the boundary of a single space, and reach connected sources such as SharePoint and Google Drive through built-in connectors and MCP, with access managed through the gateway. Which means the governance question — who may ask what, of which sources — is now a BI question too. See Unity Catalog.
Six things between the question and the number.
Understanding this sequence is what separates a useful rollout from a disappointing one, because five of the six steps are yours to get right and only one belongs to the model.
- 1
The question
Asked in plain language, with follow-ups, by someone who will not read a data dictionary.
- 2
Permission
Resolved against the asker’s own grants. Nobody can ask their way to data they could not already query.
- 3
Routing
Metadata decides which certified space, dashboard or app should answer, with higher-trust sources preferred.
- 4
Definition
The metric is computed from the shared definition in the catalogue, not inferred from a column name.
- 5
Query
Run against governed tables on warehouse compute, with row filters and column masks applied on the way.
- 6
The answer, with its sources
Returned with the assets it came from, so a figure can be checked rather than believed.
It will sometimes be wrong, and the fix is not a better model
Databricks publishes no accuracy number for Genie, and anyone quoting you one is making it up. What it does publish is the machinery for making answers trustworthy: certified spaces, governed dashboards, shared definitions, and routing that prefers vetted sources and reuses their logic.
A space pointed at raw tables with no descriptions and no agreed metrics will be confidently wrong, and will poison trust in the whole platform in about three weeks. A space over ten well-described tables with nine agreed metrics is genuinely useful. The difference is curation work, and it is the project.
What has to be true before you let people ask.
We will not switch this on for a client until these are done, because the alternative is a fortnight of enthusiasm followed by a year of “we tried that, it does not work”.
A named owner for the space
One person accountable for what it answers. Curation that is everybody’s job is nobody’s.
Ten tables, described in business language
Not forty. Column comments that say what a field means to a human, not what type it is. This is also what the SAP connector brings across for free, if your data comes from there.
The metrics defined and agreed
In the catalogue, with the rule written down and signed off by the people who disagree about it.
A list of questions it must answer correctly
Twenty real questions from real people, with the right answer established by hand. This is your regression test, and you will use it every time the space changes.
A rule for what it must refuse
Questions outside the domain should be declined rather than guessed at. An honest “I do not cover that” is worth more than a plausible fabrication.
Somewhere to report a wrong answer
One click, going to the owner. Without it you will never find out, you will only notice usage quietly stopping.
A good answer in a tool nobody opens is nothing.
The single best predictor of whether this gets used is not answer quality. It is whether the answer appears where the person already is. Databricks supports that in several ways, and choosing the right one is a change-management decision rather than a technical one.
Identity is handled for you at scale: automated sync through Entra and Okta, domains grouped by business concept, custom URLs, and consumer entitlements for people who should see answers and nothing else. Which removes the usual reason a rollout stalls at fifty users.
Pick one channel and do it properly before adding a second. A rollout that lands in Teams for the whole finance team beats one that is technically available in four places and habitually used in none.
No seat licence. One thing to check.
AI/BI dashboards and Genie spaces carry no per-seat or per-licence BI fee — they bill at standard warehouse rates for the compute a question actually uses. For an organisation coming off per-user BI licensing, that alone can change the economics.
The Genie One pricing notices disagree
Databricks has published both a move to usage-based pricing for Genie One products with a free monthly per-user allowance, and a statement that Genie One and Genie Agents usage is free through a stated date. The two notices give different dates, and we have not been able to reconcile them.
If you are planning a rollout to hundreds of people, that is a sentence to resolve with Databricks in writing before you plan the budget. We will read the current pricing page with you rather than quote a figure that may already be stale.
The cost lever that actually matters is what the questions hit. A space pointed at materialized views and metric definitions costs a fraction of one pointed at three years of raw events, for identical answers. See pricing and cost.
Your data team does not get smaller. It gets different.
The pitch for self-service BI has always implied that the analytics team shrinks. It does not, and pretending otherwise is how these programmes lose their sponsors in month four.
The request queue
- Building the hundredth variant of a report.
- Answering “can you just pull me…” four times a day.
- Reconciling two numbers that were never going to match.
- Maintaining dashboards nobody has opened since 2024.
Curation, as a real job
- Describing tables in language a business reader understands.
- Defining and owning metrics, and arbitrating when they are challenged.
- Certifying what is trustworthy and demoting what is not.
- Running the regression questions when a space changes.
- Watching which questions get asked, and what that reveals about the business.
That last one is the underrated prize. The log of what people ask is the clearest statement of what your organisation does not currently know — and it is far more useful than any survey about reporting requirements.
What has to be right underneath.
Genie is a window. These are what it looks at.
Unity Catalog
Tables, files, models, agents and dashboards in one permission model, with column-level lineage, attribute-based policies and a gateway that governs what AI is allowed to touch.
Read it →Data warehousingLakehouse
SQL analytics on Delta Lake and Apache Iceberg tables in your own cloud storage, with serverless compute, self-tuning layouts and a real-time tier for dashboards and apps.
Read it →Application developmentDatabricks Apps
Streamlit, Dash and Gradio apps on serverless compute, with sign-in, permissions and Unity Catalog around them instead of a server you have to ask IT for.
Read it →Where this leads next.
© 2026 Mawkish Technologies. This guide, its diagrams and illustrations are protected by copyright. You’re welcome to learn from it and to quote it with credit and a link. Reproducing it, by hand or through AI, and publishing it as your own is not allowed. Read our copyright notice
What people ask about asking questions.
How is this different from asking a chatbot about a spreadsheet?
Genie answers from governed tables under Unity Catalog, using the metric definitions, dashboards and curated spaces your data team has certified, and it honours the asker's permissions. Metadata routes a question so higher-trust sources win. The next-generation release can also reach beyond a single space and into connected sources such as Google Drive and SharePoint, with access controlled through the gateway.
So the difference is accountability: you can see which certified asset an answer came from, and the person asking cannot see anything they could not already query.
Will it get the answer wrong?
Sometimes, and anyone who tells you otherwise is selling. Databricks publishes no accuracy figure for Genie. What it publishes is the machinery for making answers trustworthy: curated spaces, certified dashboards, shared metric definitions, and routing that prefers higher-trust sources.
That machinery is the whole project. A Genie space pointed at raw tables with no definitions will be confidently wrong; one pointed at ten well-described tables with agreed metrics is genuinely useful. Which of those you get is a modelling decision, not a licensing one.
What does it cost per user?
AI/BI dashboards and Genie spaces carry no per-seat or per-licence BI fee — they bill at standard warehouse rates for the compute a question actually uses. Genie One, the business-user interface, is the part to check: Databricks has published both a move to usage-based pricing with a free monthly per-user allowance and a date through which usage stays free, and the two notices give different dates.
Read the current pricing page before you plan a rollout to hundreds of people. We will read it with you.
Can we put it where people already work?
Yes. Dashboards can be embedded in business applications, and the Genie API is used to reach it from Microsoft Teams, Slack and Glean. There is a web interface at account level, and Databricks lists native mobile apps as coming, available on request through your account team.
In our experience adoption follows the channel, not the quality of the answer. A good answer in a tool nobody opens is a dashboard nobody opens.
Does the data team lose control?
It gains a different job. Nobody hand-builds the hundredth variation of a report, and instead the team curates: which tables are certified, how each metric is defined, which spaces exist and who may use them.
That is more leverage and less drudgery, but it is not less work, and it needs someone's name against it. Platforms like this fail quietly when curation is everybody's job.
Give us twenty questions and the right answers.
Write down twenty questions your people actually ask and what the correct answer is. We will build one curated space that answers them, measure it against your list, and show you the honest hit rate before anyone else sees it. If it is not good enough, you will know in a fortnight rather than a year.
- One domain, one named owner
- Ten tables described in business language
- Metric definitions agreed and certified
- A measured hit rate against your own questions
- A route for reporting wrong answers
Databricks, Lakehouse, Lakeflow, Lakebase, Unity Catalog, Genie, Agent Bricks, OpenSharing, Delta Sharing and Lakewatch are trademarks of Databricks, Inc. Apache, Apache Spark, Apache Iceberg and Spark are trademarks of the Apache Software Foundation. Delta Lake and MLflow are projects of the Linux Foundation. SAP, SAP Business Data Cloud and SAP Datasphere are trademarks of SAP SE. PostgreSQL is a trademark of the PostgreSQL Community Association. All other names belong to their owners. Mawkish Technologies is not affiliated with, endorsed by or a reseller for Databricks, Inc. Everything described here is taken from material Databricks publishes; product names, availability and prices change, so check the current documentation before you commit to anything. Screens shown are our own illustrations with sample data, not screenshots.
At Mawkish, the Junden Method is how we keep Databricks Genie work checked, owned and answered for, whoever does the build.