Every self-serve analytics rollout hits the same wall. The tool works, the data is governed, and then a business user asks for "active users last quarter," and the answer is wrong, because the assistant does not know that your company deduplicates users across platforms. The gap was never the SQL. It was the business context that lives in people's heads, and nowhere can the model read.
The usual fix is to write it all down. Instructions, metric definitions, glossaries, one painstaking entry at a time. That works, and it does not scale. Nobody has the hours to hand-author every definition an assistant might need across a real estate.
Genie Ontology changes the shape of that problem, and as of this month, the part that does the heavy lifting is on by default for every customer.
What is Genie Ontology?
The Genie Ontology is the context layer that provides Genie One with a business-aware map of your organisation. It has two halves, and the distinction is the whole point.
The first half is context, you author. Unity Catalog semantics: metric views, domains, and Pages. You define these, you govern them, you certify them. This is the deliberate, human-modelled layer, and it is exactly what you would expect to maintain by hand.
The second half is context Genie infers. Genie One automatically extracts and maintains a map of snippets from assets you already have: your metric views, dashboards, SQL queries, and Genie Agents. You did not write these snippets. Genie read your existing work and drew it out.
That second half is what went ungated Public Preview and, "on" by default this month. It is the shift worth paying attention to.
What an inferred snippet looks like
The docs give three concrete shapes, and they map onto the exact things people complain Genie gets wrong.
Metric definitions. "An active user is a distinct user, deduplicated across all platforms." That is the definition that was trapped in a dashboard query, now readable as context.
Authoritative sources. "Revenue questions should be answered using the curated Finance Genie Agent." A routing rule, inferred from how your assets are actually used.
Business rules. "A qualified lead only counts once a demo is booked." The kind of logic that lived in one analyst's head and broke every answer that did not know it.
None of these was handwritten for the ontology. Genie extracted them from the work your teams already did.
Why is this a governed inference, not a black box?
Inference makes governance-minded people nervous, and it should, so here is what actually holds it in check.
Every snippet gets an authority score based on where it was generated from, how often it is used, and how fresh it is. Stale or rarely-used context loses weight. Snippets are also gated by Unity Catalog permissions, so a snippet built from data you cannot see is a snippet you never get. When you ask a question, Genie One ranks the most relevant snippets, resolves conflicts between them, and answers using only the sources you are permitted to see.
And you can see the working. Answers carry citation icons that show which snippets were used. This is the line that matters: the inference is scored, permissioned, and cited, so you can audit why an answer came out the way it did. It is not Genie guessing your business in the dark.
There is a speed payoff, too!
Accuracy is the obvious win, and it is not the only one. Because the ontology ranks a small set of relevant snippets before answering, Genie One crawls and queries less data to get to a response. The docs are explicit that this improves both accuracy and latency. For a platform team, that matters at scale: a curated context layer is cheaper to run against than an assistant that re-derives your business logic from raw assets on every question. Better answers and less work per query is a rare combination, and it is the direct result of the ontology doing the narrowing up front.
Authored and inferred are not rivals
It is tempting to read "Genie infers your context" as "you no longer need to write anything". That is the wrong read, and it is worth being precise, because a nearby idea on this forum is about replacing memory with configuration.
The two halves do different jobs. Inferred snippets cover the vast surface of definitions and rules already implicit in your assets, the ones you would never find the time to write out. Authored semantics, your metric views and Pages are for the definitions you want to pin down, govern, and certify as the single source of truth. Inference gives you breadth with no authoring effort. Authoring gives you control where control matters. You want both, and the ontology is designed to hold both.
What good looks like
The teams that get the most from this share a pattern.
They treat inferred context as a starting map, not a finished one. They read what Genie inferred, and where a snippet is close but not quite right, they promote it into an authored metric view or Page so it becomes governed truth. They use the authority score as a signal for what to certify next. And they lean on the citations, checking which snippets drove an important answer rather than trusting the number at face value.
The move here is not "turn it on and walk away". It is "let inference show you what your estate already knows, then govern the parts that deserve governing".
Getting started
There is very little to switch on. The ontology is on by default, and inferred snippets are now available to every customer without a special access request. The smallest useful first step is to open Genie One, ask a question you know your business has a specific definition for, and click the citation icons to see which snippets it used. That one action tells you what Genie already understands about your data, and where the gaps are.
A few questions for you:
Are you comfortable letting Genie infer your definitions, now that the inferred snippets are authority-scored, permission-gated, and cited?
Where do you draw the line between the context you let Genie infer and the context you insist on authoring and certifying yourself?
Have you checked the citation icons on a real answer yet, and did the snippets Genie used match what you expected?
For more on getting Genie grounded in your business 🧞👇🏻
Feel free to connect with me directly or drop a DM: linkedin.com/in/lingeshwarankanniappan
Sources
Chat in Genie One, including the Genie Ontology section: learn.microsoft.com/en-us/azure/databricks/genie-one/chat
Unity Catalog semantics and Pages: learn.microsoft.com/en-us/azure/databricks/uc-semantics/pages
AI/BI and Genie One release notes 2026: docs.databricks.com/aws/en/ai-bi/release-notes/2026