Databricks turns Genie into an always-on agent with 91% task accuracy via Ontology layer
CautiousUse8597 · reddit · 2026-11-12
Databricks shipped a new Genie that shifts from Q&A to an agent owning ongoing work.
- Jobs instead of chats: hand it a job plus tools; it tracks progress, chases inputs, reacts to new data, and checks back for approval. Jobs run on schedules or event triggers (metric drops, repeated customer requests)
- Proactive: notices repeated work like weekly updates and offers to take over; one panel supervises parallel jobs with configurable human-in-the-loop limits
- Context layer: Genie Ontology (now GA) is a business knowledge graph agents build from dashboards, queries and notebooks, with governed Unity Catalog definitions taking precedence. In a Turing benchmark, Genie with Ontology hit 91% on enterprise tasks — 1.6x the strongest frontier agent — using under half the tokens (vendor benchmark)
- Genie Search: hybrid index over Drive/SharePoint/Jira/Confluence/Workday plus live MCP calls, enforcing per-source permissions
- Acts under your identity, fully traced, per-user budgets, can swap in cheaper open-weight models; plugins bring context into ChatGPT, Claude, Copilot and Grok. CEO Ghodsi uses it to surface bad news nobody tells the CEO of a 12,000-person company
Related event: Databricks launches autonomous AI coworker Genie with 91% accuracy(2 posts)→
More from coding & agent
- Developer removes Figma, Linear and VS Code from his dock as work goes headless — mobileraj · 2026-11-12
- Open-source music-skill lets Claude compose note by note — 7 pieces, 5 models, one prompt each — BandicootNo3152 · 2026-11-12
- MIT's Missing Semester adds an agentic coding lecture in its 2026 edition — moutsea · 2026-11-12
- DEATH MUST DIE: a game built with MCP and Claude Code officially launches — andrey_kurenkov · 2026-11-12
- Google Antigravity ships voice mode, cloud remote agents and /swarm in rapid-fire updates — gerardsans · 2026-11-12
- Stanford HAI's Loyal Agents initiative opens hackathon on human-agent research collaboration — StanfordHAI · 2026-11-12