The New Genie: From Data Q&A to Enterprise Agent

Context, not model size, drives business accuracy

Genie centers on the Genie ontology, an automatically generated map of an organization's data and business logic. It extracts snippets from tables, queries, dashboards, and connected apps, assigns each an authority score based on usage and authorship, and ranks them with an onto rank algorithm to resolve conflicts. Certified, broadly used sources outrank stray or outdated assets, so questions land on the right tables and the right definitions.

Answers include methodology and provenance. Genie recomputes metrics from authoritative sources, explains differences with stale or deprecated approaches, and respects data freshness. Customers report both higher accuracy on hard, context‑dependent questions and faster responses, since Genie avoids costly trial‑and‑error crawls across irrelevant data. GM, for example, runs more than 1,000 Genie agents across 9 million Databricks assets to retrieve information in moments rather than hours.

Connected to your real data, tools, and semantics

Genie routes across multiple connection patterns, each suited to a workload:

• MCP live queries for fast‑changing systems such as Jira, Gmail, Slack, or Salesforce, checked at query time for permissions.
• Managed search connectors that index unstructured content without teams managing pipelines, chunking, vector databases, or ACL propagation, entering private preview for select sources.
• Lakeflow Connect and Lakehouse Federation for structured systems of record, enabling text‑to‑SQL over dozens of data sources so results can be computed from live operational data.

Semantics are first‑class. The Unity Catalog glossary provides a certified, version‑controlled dictionary of business terms that flows bidirectionally with the ontology. Teams can import existing definitions from established ontology or semantic model vendors, promote inferred insights into curated terms, and export back out, building on current investments rather than starting over.

From answers to action: documents, skills, agents, and schedules

Genie moves beyond Q&A to get work done:

• Co‑authored documents that pull live data, generate interactive charts, restyle tone for audiences like investors or boards, and add sections such as cost‑benefit analyses on request. Documents can be shared by email, PDF, link, Slack, or Teams.
• MCP‑driven writes that act in systems such as Slack, Jira, GitHub, and Salesforce using the user's credentials, preserving source permissions and audit trails.
• Reusable skills that codify repeatable tasks and can be published to a Unity Catalog Skills Registry for organization‑wide reuse.
• Genie agents, formerly Genie Spaces, that bundle sources, instructions, skills, and memory into domain‑scoped coworkers for scenarios like renewals, launches, or finance KPIs. Subagent support is on the roadmap.
• Scheduled runs powered by Lakeflow Jobs, delivering reliable, recurring outputs. Proactive Genie Insights surface personalized, notable changes, then let users investigate why.

Genie meets users where they work. A customizable web app anchors the experience, a new Genie Mobile app for iOS and Android is in public preview, and integrations with Slack, Teams, and Copilot bring skills and scheduling into existing workflows. Other agents can call Genie through MCP, still benefiting from the ontology behind the scenes.

Governance and cost control without a new permissions model

Genie does not create a parallel permissions system. It enforces what already exists in connected sources. Live MCP connections validate access at query time, and indexed content stores and refreshes permissions within source API limits. Two employees can ask the same question and get different, correct answers based on what each is allowed to see.

Administrators gain a unified console to connect, configure, monitor, and restrict sources per metastore, including tool‑level controls such as allowing Jira comments but not ticket creation. Everything integrates with Unity AI Gateway for centralized control over tool access, monitoring, spend caps, and alerts. Pricing moves to pay‑as‑you‑go in DBUs with no per‑seat charges, a monthly free allowance of a little over ten dollars per user, and visibility through system tables. Unilever reports cutting finance risk analysis from days to minutes, already realizing roughly three million dollars in annual savings as they scale to hundreds of users.

Why Genie can be a daily driver for business decisions

Enterprise answers require enterprise context. Genie earns trust by grounding responses in a maintained ontology, acting through governed tools, and showing up in the systems people already use. The result is an AI coworker that explains its work, executes the next step, and operates inside the security perimeter organizations already trust.