What happened
Databricks says its new Genie plugins work natively in ChatGPT, Claude, Microsoft Copilot Cowork and Grok Bot. Organizations whose agent client does not have a native plugin can instead connect to Genie through the company’s MCP server. The product surface is intended to let teams keep business context in Databricks while exposing it to assistants they already use.
At the center of the announcement is Genie Ontology. Databricks describes it as a governed context layer containing certified definitions, relationships and source context. Its stated purpose is to give an assistant a business map before it begins searching tables, documents or MCP-connected services. In the company’s framing, an agreed definition of an active user, a revenue metric or a pipeline measure can travel with the request instead of being inferred from schemas or scattered documentation.
Databricks describes two ways an assistant can use Genie. First, the assistant can hand a business-related question to Genie. Genie determines the relevant business terms through Genie Ontology, identifies data the user is allowed to see, writes and runs SQL, and returns an answer with citations. If analysis needs a specialty, Databricks says Genie can route it to a domain-specific Genie Agent: a pipeline-attribution question can go to a marketing agent, for example, while an operational-cost question can go to a finance agent.
Second, an assistant can retain ownership of a broader task and call Genie only for the context it lacks. Databricks gives examples such as an agent editing files, drafting a plan or posting to Slack after retrieving governed business facts. The company also says Genie can draw insights from unstructured enterprise data, alongside the table-oriented retrieval and SQL workflow it describes.
The integrations are presented through example workflows rather than deployment metrics. A ChatGPT Dot can monitor a feature launch, ask Genie about adoption against a certified active-user definition, compare the result with targets and surface related support themes. In a Claude example, an analyst investigating a regional revenue miss requests revenue, order volume and average deal size by product and channel, then uses Genie to search pricing memos, sales-call notes and competitor reports. Databricks says the resulting analysis includes citations and totals that match the official forecast.
For Microsoft Copilot Cowork, Databricks describes an account-planning flow that retrieves governed usage and account trends from Azure Databricks, then produces a Word briefing, an Excel overview and a presentation. The example says Cowork waits for approval before sending anything. In the Grok Bot example, an always-on marketing teammate identifies campaigns missing targets, investigates them, drafts budget shifts with attached evidence and prepares re-engagement audiences from product-usage and CRM data.
Clients supporting the MCP Apps extension can receive interactive visualizations and Genie Ontology citations directly in the conversation, according to Databricks. To begin, the company says a team enables the Genie plugin in its agent workspace and then asks business questions or assigns tasks in the normal workflow. The announcement positions the Genie MCP server as the alternative route for agents without a native plugin. It does not state a price, usage quota or quantitative rollout detail beyond describing the native integrations as available today.
What the archive adds
New Runtime's earlier coverage on October 8, 2026 described ChatGPT plugins becoming interactive interfaces beyond the conversation stream (Build richer, more interactive plugins for ChatGPT). This is a new integration event, not a restatement: Databricks is applying an assistant plugin surface to governed enterprise context across several products.
Trend signal
This supports Agent protocols become the interoperability layer. Databricks is offering native integrations for four named assistants while retaining MCP as the route for clients without one. That is the practical shape of an interoperability layer: a service can optimize selected connections without making its data and context available only through those vendors. The announcement does not show whether the fallback delivers equivalent capabilities in every client.
Why now
Databricks frames the release as a response to a growing enterprise problem: agents have more possible data sources to inspect, but not necessarily a reliable way to determine which definition, table or document should govern an answer. Its proposed solution is to build organizational meaning in Genie Ontology once and make it callable inside the assistant environments where employees already work.
Connections to other events
The release also meets New Runtime's October 8 coverage of ChatGPT plugins as interactive surfaces. That earlier development concerned how a plugin can occupy more than the conversation stream; Genie adds a particular enterprise payload to that surface: definitions, permissions, retrieved data, citations and, in MCP Apps-capable clients, visualizations. The connection is architectural rather than a claim that the two products share identical plugin behavior.
Why it matters
Enterprise-agent work often fails before the model’s reasoning becomes the issue. A capable assistant may retrieve a real number but apply the wrong revenue definition, consult an outdated document or expose a result outside the intended access boundary. Databricks is positioning Genie as a reusable control point for those problems: business meaning is defined centrally, retrieval is scoped to what the user may see, and outputs are accompanied by citations.
That approach could matter most for teams running the same workflows in more than one assistant. Instead of recreating metric definitions and retrieval rules in each ChatGPT, Claude or Copilot implementation, they can aim to maintain a common context layer. MCP gives that design a path beyond the current native-plugin list. But the integration does not remove the need for product and data owners. Ontology definitions, permissions, source freshness and query review remain part of the operational system.
What happens next
- Databricks expands Genie access to additional assistant clients through MCP rather than bespoke native plugins. (6 months, 65%). Watch for: named client integrations or MCP compatibility documentation.
- Enterprise buyers ask for auditable evaluation of citation accuracy, permissions and generated SQL before placing Genie in automated workflows. (12 months, 70%). Watch for: published controls, benchmarks or customer deployment details.
Limits and uncertainty
Databricks does not report independent measurements for answer accuracy, citation quality, latency, SQL reliability or business impact. The workflows are product examples, not measured customer case studies. No price, quota, deployment timeline, required Databricks edition or detailed setup burden is stated in the announcement.
It is also unconfirmed how consistently the plugin experience works across the named assistants, or whether the MCP route offers the same interactive visualizations and interaction model as a native connection. The company says Genie uses data the user is allowed to see, but does not detail permission propagation, audit controls or behavior when definitions conflict, sources disagree or a permitted dataset is incomplete. Citations make inspection easier; they do not independently establish that the interpretation, query or conclusion is correct.
Skill to build now
Map one recurring agent question to its operational definitions before connecting it to any assistant. Pick a narrow task such as weekly feature adoption review, write down the certified metric, source tables, allowed audience, update cadence and the human who approves an action. Then require the agent output to include the definition used, the cited source and the generated query or retrieval trail where possible. This creates a testable contract for an integration like Genie before it is trusted with broader business analysis.
Reader takeaway
Databricks is turning Genie into a governed business-context service that ChatGPT, Claude, Copilot Cowork and Grok Bot can call, with MCP available for other clients. The valuable change is not merely another plugin: it is the attempt to make permissions, definitions and cited retrieval portable across an evolving set of agents. Teams should evaluate that shared layer as carefully as the models above it.
Open question
Can Databricks demonstrate that a central ontology and cited SQL workflow improves decision quality across assistants, rather than only making their answers look more governed?