Microsoft IQ and Databricks Genie sell business context as a platform. Own the definitions, not the copy
Microsoft IQ is generally available and Databricks put Genie inside ChatGPT, Claude and Copilot. Business context is now the layer every assistant reads. Where it lives sets your next switching cost.

Ask three AI assistants in the same company what last quarter’s churn was, and you can get three numbers. Not because the models are bad, but because each one rebuilt the definition of churn from whatever tables and documents it could reach. In one week of November, the two largest enterprise data platforms told customers the same thing: stop letting each assistant work out what your business means, and let it read the answer from one governed place.
Two announcements, one direction
Databricks, November 11 and 13. Genie Ontology became generally available: a layer of business definitions, metrics and relationships, learned from dashboards, SQL and notebooks and governed by people in Unity Catalog. The same week, Databricks shipped Genie plugins for ChatGPT, Claude, Microsoft Copilot Cowork and Grok Bot, with an MCP server for any other agent. The assistant hands Genie a business question, or asks only for the missing context; Genie writes and runs the SQL with the user’s permissions and returns a cited answer. A Databricks admin has to turn it on first, and interactive charts and ontology citations only appear in clients that support the MCP Apps extension. A follow-up post with OpenAI described three patterns on top of it: ChatGPT with the Genie plugin, Codex routed through Unity Gateway, and custom agents running GPT models next to the data.
Microsoft, November 17. Microsoft IQ became generally available as what Microsoft calls an enterprise context management platform. It brings together Work IQ for people and work, Fabric IQ for business data and its metrics, Foundry IQ for unstructured content and Web IQ for outside information, with what Satya Nadella described as hundreds of new connectors. Microsoft says the context capabilities are generally available across Microsoft Copilot, GitHub Copilot, Foundry and “any agent harness via MCP.” It comes with no new SKU or subscription. Business plugins, which let domain experts write guidance that agents inherit down an organizational hierarchy, with sensitivity labels and change auditing, are in the Frontier early-access program.
Where the meaning of your business lives
Definitions you own
- Metric formulas
- Entity relationships
- Approved sources
- Expert guidance
Versioned, exportable, reviewed
Context platform
- Microsoft IQ
- Genie Ontology
- Others
Stores, governs and serves them
Assistants you rent
- Copilot
- ChatGPT
- Claude
- Custom agents
Read through plugins and MCP
Why this is the right idea
This is the direction we argued for when SAP data started leaving SAP: what agents need is not more data access but for the meaning to travel with the data. Databricks makes the case with numbers from its own benchmark, built with Turing: Genie with its ontology passed 91% of business tasks against 45% and 58% for two unnamed frontier agents using the same workspace through a toolbelt, with less than half the tokens. It is a vendor-run test on a fictional company, and the gap will be smaller in a young deployment, but the direction is plausible: an agent that can read the certified definition of “commercial planning rate” does not have to guess it.
Microsoft’s version adds something Databricks does not emphasize: guidance, not just data. A business plugin that tells every agent in the finance department which sources are authoritative and which procedure to follow is closer to a policy manual than to a semantic layer.
Why it is also the next lock-in
Both vendors frame their layer as the neutral place every assistant can read. Both are right that it can be read from anywhere. But a layer that every assistant reads is a layer that is expensive to move. Over a year, an ontology accumulates thousands of certified definitions, learned relationships and expert corrections. That accumulation is the valuable part, and it lives in a format and a service that belong to one vendor. We made the same point about agents’ memory and skills becoming the new switching cost; business context is the same story at company scale.
The answer is not to avoid these platforms. It is to keep ownership of the part that is yours.
How to adopt a context platform without being owned by it
Keep the source of truth for definitions outside the runtime. Metric formulas and entity definitions should exist as reviewed, versioned text or code your team controls, from which the platform is populated, not only as objects inside it.
Know which layer is authoritative for which question. If Fabric IQ and Genie Ontology both define revenue, one of them is wrong eventually. Assign each domain a single home.
Test portability once, early. Export the definitions of one domain and load them into a second tool. If that takes a month, you have learned your switching cost while it is still small.
Treat expert guidance as a controlled document. Microsoft’s business plugins and similar features turn procedures into instructions agents follow. In regulated work, that guidance needs an owner, a review date and a change history, like any SOP.
Measure consistency, not just accuracy. Ask the same business question through two assistants every week and compare. When context is shared, the answers should match; when they do not, you have found a definition living in the wrong place.