A CFO asks for last quarter's revenue. Finance says 4.2 million. The growth team, looking at their own dashboard, says 4.8. Both numbers are "right." One counts bookings, the other counts recognized revenue, and nobody wrote that distinction down anywhere a machine could read it. Now point a natural-language tool at the same warehouse and ask it the same question. It does not resolve the disagreement. It picks a side, sounds confident, and you have no idea which analyst it just agreed with. Databricks built the Genie Ontology to deal with exactly this problem, and it is worth understanding what it actually does, and what it asks of you, before deciding whether it solves the parts that matter.
What the Ontology is
The Ontology is a map of your data and business that Genie One generates and maintains automatically. Databricks frames it as a continuously improving, automatically inferred knowledge graph of business terms, entities, and KPIs that grounds Genie's answers. It launched alongside Genie One and Genie Agents at the June 2026 Data and AI Summit and is currently in preview.
The problem it targets is the one every enterprise AI tool runs into. An agent can read your table schemas, but it does not know what the data means to your business. That meaning is scattered across dashboards, queries, pipelines, wikis, tickets, documents, and chat threads. When the agent cannot find it, it fills the gap with inference, and inference is how you get answers that are generic, or confident and wrong. The Ontology is Databricks' bet that this context can be extracted automatically instead of hand-written into prompts one definition at a time. It is the same problem conversational analytics tools are all converging on: the model is rarely the bottleneck, the missing business context is.
How it works
The mechanism comes down to a few moving parts.
Snippet extraction. Genie pulls small pieces of knowledge, called snippets, from your tables, queries, dashboards, docs, pipelines, and connected apps. It does not stop at Databricks assets. It reads from more than 50 connected applications including Slack, Jira, Google Drive, SharePoint, and Salesforce. The snippets come in three kinds:
Metric definitions capture how a term is calculated, like an active user being a distinct user deduplicated across platforms.
Authoritative sources point to the trusted table for a fact, like revenue coming from a specific certified table.
Business rules encode conditional logic, like a lead only counting as qualified once a demo is booked.
Authority scoring. This is the part Databricks treats as the core idea. Every snippet gets an authority score, and at the launch keynote the team named the ranking method OntoRank, modeled on PageRank. Instead of ordering web pages by inbound links, it orders definitions by how trustworthy they look across your company. The score weighs five things:
Where the definition came from.
The authority of the person or source that created it.
How often people rely on it.
How closely it ties to certified and widely used data assets.
How fresh it is.
Conflict resolution. The scoring exists to solve the revenue problem from the top of this post. When finance and sales each define a term their own way, the Ontology does not guess. It surfaces the definition with the strongest trust signals and answers from that one.
Permission gating. Snippets inherit Unity Catalog permissions. When you ask a question, Genie ranks the most relevant snippets, resolves conflicts, and answers using only the sources you are allowed to see. Nobody gets an answer built on data they were never cleared for.
Why it helps performance. Because Genie retrieves answers from governed definitions instead of reasoning over scattered fragments, Databricks says it returns more accurate answers, faster, at lower token cost. Their internal benchmark claims Genie answered 84.5% of questions correctly on the first attempt against 52.4% for the strongest general-purpose coding agent, at roughly twice the speed. Worth holding lightly: that is a Databricks-run 28-question suite, not an independent benchmark. For an independent reference point on data-reasoning accuracy, Genloop publishes its results and holds the top spot on the public Spider 2.0 benchmark.
How you interact with it
From a user's seat, the Ontology is mostly invisible. It works under the hood, and the one direct touchpoint is the citation icons in a response. Click them and you see which snippets fed the answer.
It also takes manual input. You can author definitions yourself through Unity Catalog metric views, also called Business Semantics, which become one trusted, curated input among the many the graph reads automatically. This is the governed place your definitions live, and Genie treats it as a high-authority source.
The curation underneath the automation
Here is the part that gets glossed over. The auto-extraction sits on a curated foundation, and the most trustworthy answers still lean on definitions a person builds by hand.
Databricks is explicit that a Genie Space is configured by a SQL-proficient analyst. The manual surface area is real:
KPIs are entered as Measures, where someone writes the name, the SQL calculation, and the synonyms.
Filters and fields are each defined as SQL expressions with their own names and synonyms.
Example queries are authored as gold-standard templates for the questions you expect.
Join cardinality is declared by hand so Genie does not double-count a metric.
Verified answers are marked manually to signal which responses to trust.
The docs call curating a space an iterative process you keep returning to. Every new metric, renamed field, or shifted definition is another edit, and that edit sits with your scarcest people. The Ontology softens the cold start by auto-extracting, but it does not remove the curation. It rides on top of it. For a data leader weighing rollout, that ongoing curation load is the cost that does not show up in a demo.
Where the Genloop design philosophy splits
The Ontology is a strong answer to the conflicting-definitions problem. The interesting comparison is with a system built on a different premise about what to model, how to learn it, and where it lives. Genloop is structured around a few pieces, each lining up against a part of how Genie works.
What gets modeled: the Living Context Graph
Genie's snippets (metric definitions, authoritative sources, business rules) are all properties of the data. They describe what is in the warehouse and which version to trust.
Genloop's Living Context Graph goes deeper, across four layers, two of which never touch the warehouse:
Data architecture. Tables, columns, relationships: the layer Genie also covers.
Processes. KPIs, playbooks, and reference docs, applied the same way every time.
Decisions. The corrections your team made, stored and reused on every future answer.
People. Whose validated knowledge should shape how the system answers for everyone else.
The last two are the difference. Genie maps the data; the Living Context Graph maps the data plus how your organization works and decides, so a correction and its reasoning become part of the model instead of dying in a Slack thread.
How it stays current and how you steer it: the Self-Learning Loop and Context Hub
Genie's reliable definitions are curated, and a curated snapshot starts decaying the moment the business changes. Someone has to refresh it by hand.
Genloop's Self-Learning Loop inverts that. It builds the context graph the moment you connect your data, then keeps updating from real usage: every interaction, validated answer, and piece of feedback. The more you use it, the sharper it gets, and the human role shifts from authoring definitions to reviewing the ones the system proposes.
That learning also stays visible. Genie's knowledge lives per space and updates underneath without showing you the change; citations name an answer's sources but never surface a definition it just learned.
Genloop's Context Hub manages all four layers in one place, with a governance flow that shows the old value next to the new one and waits for you to approve, reject, or edit before it sticks. So you get learning you can read, and one context that holds across the platform, Slack, and Claude via MCP: one correction enforced everywhere instead of re-entered in each place.
The two systems diverge at almost every layer:
What gets modeled. Genie covers snippets at the data layer: metrics, sources, rules. Genloop's Living Context Graph reaches further, spanning data, processes, decisions, and people.
How KPIs get defined. Genie leans on a SQL-proficient analyst to enter each as a Measure, plus auto-extracted snippets. Genloop infers them from real usage through its Self-Learning Loop, then confirms them.
How it stays current. Genie auto-extracts and re-scores snippets, with curated spaces refreshed by hand. Genloop learns continuously from every interaction, validated answer, and piece of feedback.
How definitions are ranked. Genie uses an OntoRank authority score across five trust signals. Genloop carries forward whatever a person has validated.
What it shows you. Genie shows citations for the sources behind an answer. Genloop shows the change itself: the old value beside the new one, editable.
Reach of a correction. Genie keeps it per space. Genloop holds it across the platform, Slack, and Claude via MCP.
Choosing between Genie One and Genloop?
If your team lives inside Databricks, you have SQL-fluent analysts with time to curate, and you want a knowledge layer that respects Unity Catalog, the Ontology fits well. OntoRank is a genuinely smart approach to the conflicting-definitions problem, the permission model is sound, and the citation trail is honest about what each answer used.
If your constraint is analyst capacity, or you want the system to model how your team decides and not just what is in the warehouse, the trade-offs point elsewhere. A context graph that reaches into processes, decisions, and people, a loop that learns those from usage, and a hub that shows you the change before it sticks, together keep the effort low and the context current at the same time.
The Ontology answers how a machine can build and rank its own map of your business. The open questions are who pays the curation cost underneath it, whether the trusted definition produces the correct number, and whether you find out when the map starts pointing somewhere wrong.
If you want to see what a context layer that learns from usage and shows its work looks like on your own warehouse, try Genloop for free.
