Genie Ontology explained: How Databricks Turns AI Context into Shared Infrastructure

What Genie Ontology means for every AI experience you build on Databricks?

Learn how Databricks Genie Ontology uses OntoRank, Unity Catalog and organizational knowledge to create a reusable context layer for Genie Agents, Genie One and enterprise AI applications.

If you’ve built an enterprise AI assistant, you know the ritual.

You start writing instructions. “Revenue means recognized revenue, net of refunds, from the billing system. Customer means an account with at least one paid invoice. Active user means a session of at least 30 seconds on a core feature in the last 30 days. Use the sessions table. Do not use the users table.” You explain your metrics, your table relationships, your edge cases. You hit save. You test it. It works. 

Then another team builds their own AI assistant. They explain the same business, slightly differently. Their “active user” uses a 7-day window. Their “revenue” includes refunds. Their “customer” counts trials. Now you have two AI assistants giving two different answers to the same question. Both look confident. Nobody gets an error. Unless someone happens to compare them, nobody notices.

This happens with every AI tool your organization builds. Genie Spaces, Copilots, RAG chatbots, internal agents. Every single one needs to be taught what your business means. And every single one gets taught differently.

Why Enterprise AI assistants keep giving different answers?

The model isn’t the hard part. Getting an LLM (large language model) to answer a business question correctly once is manageable. The hard part is making it answer correctly six months later, when the business has changed, the team has turned over, and the instructions nobody remembers writing are quietly out of date.

Context, in this sense, is the organizational knowledge that sits between raw data and a correct answer: what a term means in your company, which table is authoritative, how a metric is calculated, which version of “revenue” finance uses versus sales. It’s not in the schema. It’s not in the model. It lives in people’s heads, in Slack threads, in dashboard descriptions, and in the instructions you write into AI tools by hand.

Every team is solving this the same way:

• Writing longer prompts with more business rules
• Building RAG pipelines (Retrieval Augmented Generation — a technique that fetches relevant documents before the AI answers)
• Curating examples and sample questions
• Manually describing datasets, columns, and KPIs

It works. Until it doesn’t. The person who wrote the instructions leaves. The business changes and nobody updates the prompt. A new team starts from scratch because they don’t know the old context exists. Documentation rots quietly, and the AI keeps answering confidently from whatever it has.

The core problem: AI doesn’t actually understand your business. It only understands the context you’ve supplied. If that context is incomplete, siloed, or stale, your AI’s understanding is too. And it won’t tell you.

Maintaining context has quietly become its own engineering discipline one most teams didn’t sign up for and can’t sustain long-term.

Why context becomes siloed across Databricks Genie Spaces?

If you’ve used Genie Spaces, a Databricks feature for setting up AI assistants over your data, where you manually describe tables, metrics, and business rules, you’ll recognize this immediately.

Genie Spaces were a genuine step forward. Before them, natural language to SQL tools were essentially context-blind. They got your schema and were expected to figure out the rest. 

Genie Spaces changed that by giving teams a place to encode business context explicitly. Teams that invested in good instructions got useful answers. But here’s what nobody tells you up front: that investment doesn’t compound. It decays.

The same structural problems that affect every AI tool hit Genie Spaces hard:

Across your organization, context has no home. It gets copied into prompt templates, Genie Spaces, RAG pipelines, fine-tuning datasets, each carrying its own version of what “active user” means, none of them talking to each other. Databricks published that customers created over one million Genie Spaces in 2026 alone. That’s a lot of diverging definitions.

From semantic layers to a shared AI context layer

Yes, with the semantic layer. And the parallel is worth sitting with.

Years ago, the same thing happened with SQL logic. The same revenue calculation duplicated across fifty dashboards, each slightly different, none of them authoritative. The fix wasn’t better copy-paste discipline. It was dbt, Looker, MetricFlow: define the logic once, let everything draw from it. SQL became infrastructure instead of per-dashboard maintenance.

The difference this time: the semantic layer was a deliberate engineering project. Someone had to sit down and define revenue in dbt. Genie Ontology is betting it can do most of that automatically, by reading what your organization already produces. That’s the harder and more interesting claim, and the one worth scrutinizing.

What is Databricks Genie Ontology and how does it work?

Genie Ontology is a context layer built into the Databricks platform, announced at Data + AI Summit in June 2026. Instead of you writing context into Genie, Genie reads context from everything your organization already produces.

What it connects to:

• Your Databricks environment: tables, queries, dashboards, pipelines
• Unity Catalog (Databricks’ governance layer for centralizing permissions, metadata, and lineage): certified metric definitions, glossaries, semantic tags
• 50+ external tools: Slack, Jira, Google Drive, SharePoint, Confluence, email, calendar

From all of this, it extracts snippets of knowledge — what a term means, how a metric is defined, which tables relate to which concepts — and organizes them into a continuously updated knowledge graph. Think of it as a map of how your business understands its own data, rebuilt automatically every time something changes.

Every product in the Genie family draws from it:

Genie Spaces is the most familiar surface, but Genie Ontology is the shared foundation underneath everything. One definition of “active user,” everywhere.

Which Databricks products use Genie Ontology?

What is OntoRank and how does it rank business context?

This is the right question, and the answer is more interesting than most coverage gives it credit for.

The same word (“engagement,” “active user,” “churn”) might appear in a dozen places across your organization, defined a dozen different ways. A two-year-old wiki page. A certified metric in Unity Catalog. A Slack message from a sales rep. A dashboard built before the product team changed how they measured retention.

When Genie Ontology crawls all of these, it has to pick one. That’s where OntoRank comes in.

OntoRank is Databricks’ ranking mechanism for knowledge snippets, inspired by Google’s PageRank. PageRank figured out which web pages to trust by analyzing the graph of links pointing to them. OntoRank applies the same principle to business definitions, across fundamentally different asset types: tables, documents, Slack threads, dashboards.

The four OntoRank ranking signals:

The old wiki definition loses. The certified metric maintained by data engineering and referenced across a dozen dashboards wins.

One honest caveat: the signals are public but the algorithm isn’t. Whether OntoRank runs iteratively like true PageRank or is a weighted scoring function, how it handles organizations with no usage history, and what happens when two high-authority sources genuinely disagree, Databricks hasn’t published any of that. Worth asking about before you rely on it in regulated environments.

Genie Ontology example: How a context-aware AI agent answers business questions?

A context-blind agent asked “what’s our churn rate?” gives you an industry benchmark and a formula. Plausible, confident, completely disconnected from your actual data.

A context-aware agent grounded in Genie Ontology knows:

• Your finance team defines churn as MRR lost from cancellations and downgrades divided by opening MRR
• The authoritative source is the billing_events table
• This definition was certified by the CFO, scores 91/100 in OntoRank
• Last month’s number was 2.3%

The same pattern plays out across any question that depends on organizational context:

• “How many active users do we have?” – Which definition? 30-day session? 7-day login? Any account ever created?
• “What was LATAM revenue in Q1?” – Which territories count as LATAM in your company?
• “Did engagement drop last week?” – DAU/MAU ratio, or feature interaction sessions?

These aren’t hard questions for a capable model. They’re questions that require knowing your organization specifically, and that’s exactly what has been missing.

The numbers: Databricks’ internal benchmark on 28 real-world data analysis questions (June 2026) shows 84.5% first-attempt accuracy with Genie Ontology vs. 52.4% for the strongest general-purpose coding agent, with answers returning about twice as fast. It’s a self-reported benchmark on an anonymized competitor set, so treat it as directional rather than definitive. Snowflake reported similar improvements (just over 80% accuracy on complex queries) when they added context to their own agents around the same time.

Context-blind AI vs context-aware AI:

What Genie Ontology improves and what it does not fix?

Context accuracy and computational accuracy are different problems. Worth being clear about which one this solves.

What Genie Ontology improves:

✓ Which definition of “churn” the agent uses

✓ Which table it queries

✓ Whether it knows your org’s terminology

✓ Consistency of answers across teams and tools

What it doesn’t fix:

✗ Whether the pipeline computing churn has a bug

✗ Whether a high-authority definition is wrong for a specific edge case

✗ Whether the SQL the agent generates is logically correct

Better context is a necessary condition for trustworthy enterprise AI. It’s not a sufficient one.

Can Genie Ontology be audited?

This is the question that should give compliance teams pause. Can you export a snapshot of the ontology state? Can you see exactly which sources contributed to a definition, and prove who approved changes? Without that, you risk trading one governance problem for another, a shadow semantic layer that’s automatic, invisible, and hard to defend. Databricks hasn’t published clear documentation on this yet. Ask your account team directly before committing to production use in regulated industries.

How does OntoRank handle cold starts?

OntoRank relies on usage signals: how many people and queries rely on a definition. A new organization or a new dataset has no usage history, which means the ranking degrades to creator authority and recency alone. The system gets more accurate over time, but early results may not reflect what you’d expect.

Can Genie Ontology resolve conflicting business definitions?

Genie Ontology can rank which definition is most used and most trusted. It cannot resolve the underlying disagreement that caused the definitions to diverge in the first place. If finance and product have genuinely different, legitimate reasons to define “churn” differently (and often they do) OntoRank picks a winner but doesn’t fix the business problem. Someone still has to make that call. The system automates discovery and ranking; it doesn’t replace the governance conversation.

How to prepare for Databricks Genie Ontology?

Genie Ontology is in public preview (not available to everyone), and GA hasn’t been announced. You request access through your Databricks account team.

That said, the work you do now directly determines how well it performs when you get access. The ontology is only as good as the context it can find. If your Unity Catalog semantics are a mess, the ontology starts from a weak foundation.

Step 1: Define trusted business metrics in Unity Catalog: Start with the 5–10 metrics your teams argue about most — churn, active users, revenue, engagement. Define them once, officially, in Unity Catalog Glossary and Metrics. These become the highest-authority inputs OntoRank has to work with. If you’ve been putting this off because it felt like documentation nobody would read, the calculation just changed.

Step 2: Audit existing Genie Spaces context and instructions: Your existing instructions get absorbed as manual context and treated as high-authority input. But stale or inaccurate instructions will compete with your certified metrics in OntoRank — and might win if they have higher usage signals from months of queries. Review them before the ontology ingests them. Delete anything outdated.

Step 3: Map where Enterprise AI context lives across your organization: Before the ontology crawls your organization, know what it will find. Which definitions exist only in a dashboard query? Which metrics are defined differently in Slack vs. your BI tool? Which tables have no documentation at all? This audit tells you where the ontology will be weakest — and where to focus first.

Step 4: Request early access: Earlier access means more time to observe OntoRank behavior before you’re relying on it. Contact your Databricks account team.

Is Genie Ontology the future of Enterprise AI context?

Context has always been infrastructure. It just took us a while to build it that way.

For years, every team taught their AI tools the same business logic separately. The result was predictable: inconsistent answers, stale definitions, and a maintenance burden nobody planned for.

Genie Ontology is Databricks’ attempt to fix that at the root, context extracted automatically, ranked by authority, updated continuously, shared across every AI experience your organization builds.

The caveats are real. It’s in preview. OntoRank’s internals aren’t public. The auditability question is unanswered. And no algorithm substitutes for the governance work of getting your organization to agree on definitions in the first place.

But the direction is right. We centralized SQL logic with the semantic layer and stopped copying calculations into every dashboard. The same shift is now available for context. The organizations that make it will build AI systems that actually know their business. The ones that don’t will keep paying the context tax.

Have the governance conversation before the ontology has it for you. Then request early access through your Databricks account team.

Genie Ontology is in public preview on Databricks as of June 2026. This post was written in July 2026 based on publicly available information at launch.

Frequently asked questions

What is Genie Ontology? Genie Ontology is Databricks’ context layer, announced at Data + AI Summit in June 2026 and currently in gated public preview. It automatically extracts business definitions from your Databricks data and 50+ connected tools, ranks them by authority using OntoRank, and shares them across every AI experience you build — Genie Spaces, Genie One, Genie Agents, Genie Code, and external tools via MCP.

What is OntoRank? OntoRank is the ranking mechanism inside Genie Ontology, inspired by Google’s PageRank. When multiple sources define the same term differently, OntoRank ranks them by four signals: creator authority, usage breadth, dataset linkage, and recency. The most trusted definition wins and gets used across all Genie experiences.

Is Genie Ontology available yet? Genie Ontology entered public preview in June 2026. Genie One, Genie Agents, and Genie Code are generally available. GA for Genie Ontology hasn’t been announced — request access through your Databricks account team.

Does Genie Ontology replace Genie Spaces? Not exactly. Genie Spaces are evolving into Genie Agents, which are more capable. Genie Ontology is the context layer underneath both. Think of Genie Spaces as the previous generation and Genie Agents as the next — sharing the same foundation.

What’s the difference between Genie Spaces, Genie One, and Genie Agents? All powered by Genie Ontology, different surfaces. Genie Spaces are scoped assistants tied to a specific dataset. Genie One is the general-purpose AI coworker for business users — answers questions across your whole data estate, produces documents, takes actions in external tools. Genie Agents are reusable autonomous agents you build for specific workflows and share with your team.

What happens to the context I already wrote in my Genie Spaces? It’s preserved and treated as high-authority manual context. It becomes one input among many rather than the only one — but as noted above, stale instructions can still compete with certified metrics. Review them before the ontology absorbs them.

Does Genie Ontology work outside of Databricks? The ontology is built from Databricks data plus 50+ connected external apps. The context it produces can be accessed by external AI tools via MCP (Model Context Protocol — an open standard for connecting AI tools to external data sources), so Claude, Copilot, or custom agents can use the same governed context.

Is it secure? Genie Ontology respects Unity Catalog permissions. Snippets are permission-gated — an agent only sees context the user is entitled to see.

How is Genie Ontology different from Snowflake’s offering? Snowflake launched Horizon Context (via Cortex Sense) weeks before Databricks announced Genie Ontology. The architectural difference: Genie Ontology pre-builds a knowledge graph and ranks definitions by trust before any question is asked. Cortex Sense retrieves and reorders definitions at query time. Both reported similar accuracy improvements — roughly 80–85% on complex queries — suggesting the context problem is real regardless of which platform you’re on.

How is this different from RAG? RAG (Retrieval Augmented Generation) fetches relevant documents at query time — it searches while it answers. Genie Ontology builds the knowledge graph before anyone asks a question. It’s faster, and it’s permission-aware in a way generic RAG pipelines typically aren’t.

Does Genie Ontology work with dbt? Not directly documented at launch. Unity Catalog Metrics — which feeds the ontology — can be aligned with dbt semantic models, but the integration path isn’t officially specified yet. Worth confirming with your account team if dbt is central to your semantic layer

Sources

• Introducing Genie One, Genie Agents, and Genie Ontology — Databricks Blog, June 2026
• Databricks Launches Genie One: All-New Agentic Coworker for Every Team — Press release, June 2026
• Genie One product page — Databricks
• Unifying Data and Governance in the Agentic Era: What’s New with Azure Databricks — Databricks Blog, June 2026
• Chat in Genie One — Azure Databricks documentation, Microsoft Learn

• From RAG to ontology: Databricks bets on context as the key to trusted AI agents — CIO, June 2026
• Not PageRank, but OntoRank: Databricks Genie Ontology brings context and authority to AI — ITdaily, June 2026
• Databricks lets a Genie One AI agent coworker out of its magic lamp — Blocks & Files, June 2026
• Databricks Data + AI Summit 2026: All 20+ major launches — Flexera, June 2026
• Key takeaways from Databricks Data + AI Summit 2026 — Atlan, June 2026

• Genie Ontology: Databricks’ New Context Layer for AI Agents — Sreedeep Sinha, Medium, June 2026
• Genie Ontology VS Genie One VS Genie Agents — Harshit Dawar, Medium, June 2026
• The Ontology Renaissance: Why a Decades-Old Discipline Now Powers AI — DataMy, June 2026

• What Is Genie Ontology? Databricks’ Continuously Learned Context Layer Explained — Typedef.ai, June 2026
• What is Genie Ontology? Databricks’ Context Layer, Explained — Atlan
• What Is Databricks Genie Ontology? — Dawiso
• Genie Ontology and the Atlan Context Layer: Better Together — Atlan
• What is Databricks Genie One? — Atlan

About Datapao

Datapao is a Databricks-focused data and AI consulting and training company operating in the UK and across Europe, with delivery hubs in the UK and Central Europe. The company helps organisations design, build, and operate modern data platforms while enabling internal teams through hands-on Databricks training. Our approach combines deep engineering expertise with structured knowledge transfer to ensure sustainable, long-term adoption of data and AI.

Get in touch to discuss where to start.