Databricks announces second-generation Genie

Databricks describes Genie as taking ownership of work, tracking progress, following up on missing inputs and contacting people when it needs a decision.[1] It can delegate tasks to multiple sub-agents so work proceeds in parallel.[1] The company says employees provide a goal and clear boundaries, after which Genie manages the task and contacts them when it needs input.[1] Genie can also run scheduled or event-triggered jobs and manage several assignments at once, according to Unite.AI.[2] Databricks says it runs in autonomous mode using its own compute.[1]

Genie Ontology captures company context such as definitions, metrics, people and processes; Ontology Snippets store reusable business context and update as underlying sources evolve.[1][3] Databricks says the Ontology ranks context using signals such as authority, certification, expertise, popularity and recency to resolve conflicting definitions.[3] Genie Search finds knowledge in documents, tickets and wikis alongside Databricks assets, subject to existing permissions.[3] Genie One can create slide decks, dashboards and interactive pages in a conversation, with scheduled tasks keeping artifact data current and skills applying brand standards.[4]

People can work with Genie through Genie One, email, Slack and Microsoft Teams.[5] Databricks says Genie Search supports more than 100 MCP connectors for live retrieval and actions, and can search an index in one pass rather than querying each source separately.[4][3] Native MCP connectors for Genie Agents are generally available, allowing agents to read from and act in tools including Jira, Slack and Google Drive.[6] Genie Agents can answer questions across governed tables and documents in Unity Catalog volumes with page-level citations, but remain in Public Preview.[6] Databricks says existing Unity Catalog permissions and Unity Gateway approval rules and budgets govern Genie’s actions.[1]

Databricks says thousands of customers use Genie in production, including Zillow, Unilever, adidas, Atlassian and AstraZeneca.[5] It says General Motors has deployed more than 1,000 Genie agents across sales, engineering, supply chain and HR operations.[6] In Databricks’ benchmark, Genie answered correctly in 91% of runs, 1.6 times the rate of the stronger frontier agent, while using less than half the tokens.[3] The company cautioned that those results reflect an established deployment where teams had accumulated business knowledge.[3] Analyst Mike Leone said goal-based work could surface answers to people who would not normally open a dashboard; he also warned that an unsolicited wrong answer could reach many people before it is questioned.[7]