Overview
At DAIS 2026, Databricks presented an integrated vision of the Genie ecosystem: three interconnected products designed to help enterprises extract actionable insights from their data and take autonomous action across business systems.
The ecosystem addresses a fundamental challenge: business context enabling data-driven decisions is scattered across dozens of systems. Previous-generation data agents failed because they probed those systems iteratively and inefficiently. Genie’s unified approach combines accurate data access with business context understanding.
Genie One: Data-Smart AI Coworker
Genie One is a conversational AI assistant that evolved from Genie’s analytics roots to provide comprehensive coworking capabilities. It connects across entire data ecosystems through Lakehouse federation and Lakeflow Connect.
It deploys across multiple surfaces: web, mobile (iOS/Android), Slack, and Microsoft Teams. Capabilities include scheduling, alerts, monitoring, document creation, and custom skills. Native integration with everyday business tools (Gmail, Slack, Teams) and MCP support ensure compatibility with existing AI workflows.
Example use case: “Sales leaders can ask Genie to prepare a daily brief for all their customer meetings, combining the context from their calendar, email, and hard data.”
Genie Agents: Autonomous Domain-Specific Assistants
Genie Agents is the evolution of Genie Spaces (over 1 million created by customers). Agents take autonomous multi-step actions without human intervention, reasoning over both structured data (tables, views) and unstructured data (documents, files).
They are created from simple natural-language prompts, are shareable and customizable across teams, and enable domain experts to scale their expertise at the organizational level. Foot Locker described the impact: “Genie Agents are transforming how we lead… reshaping the way our business interacts with data and makes the decisions that matter most.”
Genie Ontology: Automatic Context Layer
Genie Ontology is the foundation enabling superior AI performance. It automatically extracts knowledge from tables, queries, dashboards, pipelines, and connected applications, organizing it into a living graph of business operations and data semantics.
The system implements authority-weighting similar to PageRank: evaluating source definitions based on author authority, usage frequency, connection to certified assets, and freshness. Source-native permissions are enforced automatically throughout.
Performance Benchmark
Databricks’ internal benchmark of 28 real-world data analysis questions showed:
- Genie accuracy: 84.5% correct first-attempt answers
- Strongest general-purpose coding agent: 52.4%
- Weakest competitor: 25%
- Speed advantage: Genie delivers 2× faster performance than the strongest coding agent
Key Points
- Genie One available on web, mobile, Slack, and Microsoft Teams
- 1M+ Genie Spaces created by customers (now evolving into Genie Agents)
- Genie Agents create autonomous agents from natural-language prompts
- Genie Ontology learns business semantics automatically without manual curation
- PageRank-style authority weighting for business definitions
- 84.5% accuracy in benchmark vs. 52.4% for the best general coding agent
- 2× speed advantage over the strongest competitor in the same benchmark
- Source-native permissions automatically enforced throughout the ontology
- Uplight and Foot Locker already in production with measurable benefits
Why It Matters
Business analysts have long dreamed of “just asking the data.” Previous attempts failed because agents didn’t understand context: what “a good quarter” means for this specific company, how “active customer” is defined in the billing system context, or which table is the authoritative source for revenue.
Genie Ontology addresses this at the root: rather than each agent having to learn the business from scratch, the ontology automatically builds and maintains that knowledge from all available signals — historical queries, certified dashboards, existing documentation — and makes it available to all agents.
The PageRank-like authority weighting is particularly clever: it doesn’t rely on humans to curate what’s “correct,” but instead surfaces definitions that are most used, most connected to certified sources, and most recently maintained — the same signals that indicate credibility in the real world.
