Unity Catalog in 2026: Business Glossary, Domains, and the Semantic Layer Your AI Agents Actually Need

TL;DR

• The shift: Unity Catalog is no longer just where data is governed. It is now where business meaning is defined and governed, for people and agents alike.
• Business Glossary (Preview coming soon): authoritative concepts, terms, and taxonomies. Genie Code can draft glossary pages and flag definitions that drift from how data is actually used.
• Domains (Public Preview): organise data and AI assets into business-aligned categories so an agent gets scoped, relevant context instead of the entire catalog.
• Governance Hub (Private Preview): a single command center for stewards and admins to monitor governance posture, spot risks, and prioritise remediation across data, AI, cost, and performance.
• The catch: none of this writes itself. Glossary, domains, and stewardship are foundation work. The platform gives you the control plane; someone still has to define the meaning.

Why this lands now: the agentic era needs shared meaning

Databricks frames its 2026 strategy around three words: control, context, and choice. Unity AI Gateway, which we covered separately, is the control plane for the AI runtime. This set of announcements is the context plane.

The framing in the announcement is blunt: "the agentic era is here. Hundreds of thousands of agents are now acting on enterprise data." An agent querying your tables does not know that "active user" means one thing to finance and another to product, or that "revenue" excludes refunds in one report and includes them in another. Point it at raw tables and it will produce confident, plausible, wrong answers, because the meaning lives in people's heads and in scattered SQL, not in the platform.

Unity Catalog's answer is to make that meaning a first-class, governed object. With more than 14,000 organisations now governing data and AI on Unity Catalog, it becomes, in Databricks' words, the "single, shared source of meaning" for both people and agents.

Business Glossary: authoritative meaning, not tribal knowledge

The Glossary (Preview coming soon) lets you define the authoritative concepts, terms, and taxonomies that describe your business, or import the ones you already have. Glossary pages connect to the underlying data and to each other, so a term like "churn" links to the tables, metrics, and related concepts that define it.

Two things make it more than a wiki. First, Genie Code can draft new glossary pages, suggest refinements, and, importantly, flag definitions that have drifted from how the data is actually used, so your glossary does not quietly rot. Second, the whole team curates it together through suggestions, comments, and domain-level ownership, so meaning stays owned rather than abandoned.

For an AI agent, this is the difference between guessing what a term means and reading the governed definition your business agreed on.

Domains: scoped context instead of the whole catalog

Domains (Public Preview) organise your data and AI assets into business-aligned categories, finance, supply chain, marketing, and so on. The point is scope. Instead of giving an agent the entire catalog and hoping it finds the right table, you give it the relevant domain, with the relevant context.

Humans browse domains and agents query them through an internal marketplace, with certification and stewardship signals showing how reliable each asset is. AI-driven domain suggestions are coming, so you do not have to organise everything by hand from scratch. The practical effect: agents retrieve from a curated, trustworthy slice of your estate rather than rummaging through everything.

Governance Hub: a command center for stewards

Governance Hub (Private Preview) is a centralized command center for data stewards and admins. From one place you can monitor your governance posture, identify risks, prioritise remediation, and scale governance operations across data, AI, cost, and performance.

This is the piece that turns governance from a project into an operating function. Most enterprises do a governance push, then watch it decay because nobody owns the day-to-day. A command center with posture monitoring and prioritised remediation is what keeps it alive once the consultants leave.

The supporting cast: Metrics, ABAC, and one address for everything

Three more pieces complete the picture, and they matter for context.

• Governed Metrics. Define business KPIs (revenue, churn, active users, margin) as governed, reusable objects, then query them consistently from SQL, BI tools, APIs, and agents. They are generally available and open source (Open Semantic Interchange ready), can be materialised for speed (Public Preview), and can be bootstrapped from Power BI and Tableau (Beta). Metrics, Glossary, and Domains together feed the Genie Ontology, a continuously learned enterprise context layer. This is how an agent comes to understand your business rather than just your schema.
• Attribute-Based Access Control. Row filtering and column masking are now GA; ABAC grant policies for models, tag propagation, and identity and context attributes are in Beta or Preview. This is how access scales without hand-granting every table. (We have written before on Unity Catalog governance for business leaders.)
• One address for every asset. A new four-level namespace (metastore.catalog.schema.table) gives a single address for every asset across accounts, regions, and clouds, with unified discovery, one set of policies, and one audit trail. Cross-region governance is coming in preview. For multinational enterprises, that is the difference between governing one estate and governing several.

The catch, and where it gets real for you

Read the list again and notice what it is not. Databricks did not announce a button that organises your business for you. It announced the place to put the organisation once you have done it.

A glossary only helps if someone defines the terms and keeps them honest. Domains only scope context if someone groups the assets sensibly and certifies the reliable ones. The Genie Ontology only gets smart if the glossary, domains, and metrics underneath it are real. Governance Hub only protects you if there is a stewardship function actually using it. The platform moved the ceiling; the floor is still yours to build.

This is the same pattern across the whole Summit 2026 announcement set: every new agentic capability quietly assumes a governed, well-described foundation. CustomerLake assumes a real Customer 360. Genie assumes a real ontology. Unity AI Gateway assumes a governed catalog. And the ontology those agents read from assumes a real glossary and real domains. Meaning is the new prerequisite.

The European angle

For regulated European enterprises, a governed semantic layer is more than convenience. When an agent answers a question or takes an action, an auditor will eventually ask what definition it used, which data it touched, and who certified that data. A business glossary, domains with stewardship signals, governed metrics, and Governance Hub posture monitoring are precisely the artefacts that answer those questions. Combined with cross-region governance and the residency controls in the wider platform, this is what lets a risk team say yes to agents rather than no.

How to get "agent-context-ready"

The work is foundation work, and it pays off whether or not you adopt every feature on day one.

• Define the business glossary that matters. Start with the twenty terms your reports argue about (revenue, churn, active customer, margin). Govern them, own them, and link them to the data.
• Organise your assets into domains. Group by business area, certify the trustworthy assets, and assign stewards. This is what gives agents a clean slice to work from.
• Govern your metrics centrally. One definition of each KPI, reusable across SQL, BI, and agents, so the agent and the dashboard never disagree.
• Stand up a stewardship function. Governance Hub is a tool; it needs owners. Decide who watches posture and prioritises remediation before you scale agents.
• Map it to residency and audit. For regulated workloads, make sure the meaning, the data, and the access decisions are all traceable.

Do this and your agents inherit a business they actually understand. Skip it and they inherit your mess, faster.

The Cosmos Thrace perspective

This is squarely the work we do. We are a Databricks Silver Partner, and the unglamorous foundation, classifying data, defining meaning, organising domains, standing up stewardship, is exactly where we spend our time so that the clever things on top are trustworthy. We have delivered dozens of data platform implementations across Europe, many on Databricks, with more than $50M saved for clients in 2025, a 100% client retention rate, and 106 million data points moved daily.

Our honest read: Glossary, Domains, and Governance Hub are some of the most quietly important things Databricks shipped this year, because they are what make every agent above them trustworthy. The enterprises that win in the agentic era will be the ones who treated business meaning as infrastructure, not documentation. Define the meaning, govern it, and the agents get smarter. Skip it, and no amount of model quality will save you.

Sources

Databricks blog: What's new with Unity Catalog at Data + AI Summit 2026

Databricks product page: Unity Catalog Semantics

Databricks blog: General Availability and Open Sourcing of Unity Catalog Business Semantics