Databricks has given its Genie AI assistant a promotion. Unveiled on November 11, 2026, the new always-on Genie no longer waits to be asked a question. Instead, it takes ownership of long-running enterprise work: running scheduled or event-triggered jobs, following up when information is missing, managing several assignments at once, and even delegating tasks to subagents. Users still decide which actions require approval, but the shift is clear. Genie is moving from a chatbot that answers to a coworker that delivers.
The launch, presented at Databricks’ DATA + AI LIVE event, is the company’s biggest bet yet that the next phase of enterprise AI is not better answers, but assigned work that gets done without a human in the loop. Thousands of customers already use Genie in production, according to Databricks, and General Motors has deployed more than 1,000 Genie agents across sales, engineering, supply chain and HR operations.
Learn more about Databricks’ unified data and AI platform, and read the original reporting from Unite.AI.
From Answering Questions to Owning Assignments
Genie has been evolving since its introduction in June 2024. The family grew to include Genie One for natural-language data analysis, Genie Agents for domain-specific work, and Genie Code for data engineering. Each step made the assistant more capable, but the interaction stayed the same: a person asked, Genie answered.
Always-on Genie breaks that pattern. A manager can now hand it a goal, such as keeping a weekly sales report current or monitoring a set of insurance claims, and the agent takes responsibility for the work around the analysis. It watches for changes, requests input when it is stuck, and acts as new information arrives. Databricks’ examples include routing insurance claims that drift outside underwriting guidelines and proposing marketing budget changes for approval.
In a demonstration reported by TechCrunch, Databricks CEO Ali Ghodsi showed Genie analysing employee satisfaction survey results, identifying where the company underperformed, evaluating proposed fixes, summarising poor sales pitches, and even preparing him for podcast interviews. The point of the demo was not the individual tasks. It was that Genie kept working on them without being prompted at each step.
As Mike Leone, an analyst at Moor Insights & Strategy, told TechTarget: “Genie stops waiting to be asked. Up to now, getting value out of Genie meant knowing the right question and when to ask it. Handing it a goal instead puts answers in front of folks who would never open a dashboard in the first place.”
Genie Ontology Goes Generally Available
None of that autonomy works without context, and that is where the second half of the announcement matters most. Genie Ontology, the knowledge layer first introduced in June, is now generally available. It automatically builds reusable knowledge snippets from dashboards, SQL queries and notebooks, while Unity Catalog Pages and Metric Views preserve explicitly governed definitions.
The idea is simple to state and hard to build: an agent cannot answer “which customers are at risk?” correctly unless it knows what the business means by “at risk” and which revenue figure finance actually trusts. Genie Ontology captures definitions, metrics and relationships across structured and unstructured data, wherever it lives, and makes that context available to agents. Its OntoRank system weighs source authority, certification, user relevance, popularity and recency to select context and resolve conflicting definitions.
Sanjeev Mohan, founder of analyst firm SanjMo, told TechTarget that the ontology is the capability that makes everything else possible. “Always-on Genie will get the headlines, but it only works if the agent understands the business,” he said. “By now, we understand that agents fail less from bad models than from missing context.”
An Ontology Manager for reviewing and curating the ontology is coming soon in beta, and external business-intelligence knowledge sources are in beta, initially supporting Power BI. Databricks says the ontology learns from how people actually query the data, so teams do not have to write every definition out by hand first.
Genie Search Unifies the Data Estate
Alongside the ontology, Databricks unveiled Genie Search, a governed way for humans and agents to search across an entire data estate. Genie Search is generally available for Google Drive, Jira, Confluence and Workday, with SharePoint in beta. It indexes content from those systems, enforces their permissions before returning results, and can combine the index with SQL queries and MCP connectors that retrieve live information and take actions in external tools.
That matters because so much of the context an agent needs lives outside the data warehouse: in documents, tickets and chats. Databricks product management told TechTarget that customers increasingly asked Genie questions depending on exactly that kind of information, and Genie Search was built to index it rather than leaving the agent to hop from app to app.
Organisational directories, also generally available and beginning with Workday, add reporting relationships and collaboration signals to help rank relevant knowledge. As Leone noted, one governed search also means one place to set permissions, and far fewer separate calls to pay for.
Genie Agents Add Documents and External Sharing
Genie Agents, the domain-specific members of the family, also gained new abilities. They can now analyse PDFs, slide decks and other documents alongside governed tables in Public Preview, returning page-level citations. Native MCP connectors are generally available, allowing agents to retrieve context and take actions such as creating tickets, updating records and posting messages.
Two sharing features are in beta. OpenSharing provides a governed agent preloaded with data, instructions and knowledge that can be used across Databricks accounts. Genie Clean Rooms let recipients ask questions without seeing the underlying data or proprietary logic, with conversation and output-size controls.
Genie One, the interface for this work, also creates slide decks, dashboards and interactive pages, with scheduled refreshes and shared skills for repeatable workflows. Databricks says Genie One is free for users through January 31, 2027, and that more detailed cost controls will be available before then.
Read the Databricks Genie One launch announcement for full details.
Governance, Approvals and the Turing Benchmark
Autonomy without guardrails would be a non-starter for regulated enterprises, and Databricks has paired the new capabilities with Unity Gateway controls. Traces capture the question, answer, cited objects, response time and actions, giving administrators a queryable record of everything Genie did. Administrators can restrict MCP tools, require approvals, set budgets and spending caps, and review usage. Genie acts under each user’s identity and can prepare an access request, but administrators and data owners retain approval authority.
For more on governing and rolling back rogue AI agents, see our coverage.
To back its context claims, Databricks released a benchmark developed with Turing. In the company’s evaluation, Genie passed 91% of task runs, compared with 58% for the stronger of two anonymised frontier agents, while averaging 431,829 tokens and 100.3 seconds per run. Each agent ran every task five times in the same workspace for a fictional UK energy supplier containing tables, notebooks, dashboards, saved queries and documents, including outdated and duplicated assets.
Databricks itself cautions that the test reflects an established deployment with accumulated business knowledge, and that differences may be smaller in new deployments or when table schemas alone are sufficient. That caveat is worth taking seriously: it defines the central enterprise requirement in this launch. Broader autonomy depends on governed, accessible institutional knowledge, with administrators deciding which jobs and actions Genie can take on.
Analysts also flagged gaps. Mohan argued that enterprises need to govern what agents do in production, not just what they read: a clear identity for each agent, limits on which actions it can take, human approval for consequential steps, and a full audit trail of its reasoning. Leone added that always-on Genie should be able to say when it is not sure of an answer, because “a wrong answer that shows up unprompted can reach a lot of people before anyone thinks to question it.”
What It Costs, and Who Should Care
Databricks has not put a single headline price on always-on Genie. Genie One is free for users through January 31, 2027, and Databricks says more detailed cost controls are coming before then, alongside the budgets and spending caps already available to administrators. The practical cost question for buyers is token usage: the benchmark’s average of more than 430,000 tokens per task run shows that long-running, context-heavy work is not cheap, and that is before the scheduled refreshes and background monitoring that always-on operation implies.
The launch lands in a crowded race. Snowflake, AWS, Google Cloud and Microsoft are all building their own agent and context layers, and IBM and Microsoft launched the Imagine Enterprise agentic platform with more than 6,000 prebuilt agents just this week. Databricks’ differentiator, in the analysts’ view, is the ontology: a context layer that learns definitions from real usage rather than requiring teams to document everything upfront. Competitors are chasing the same pattern, so that lead may not last long.
Related: IBM and Microsoft’s Imagine Enterprise agentic platform and Google’s Gemini 4 Carbon for long-horizon agentic work.
The Catch: Autonomy Is Only as Good as the Context
The honest summary of this launch is that Databricks has solved the easy half of the autonomous-coworker problem and is still working on the hard half. The easy half is capability: agents that can plan multi-step work, use tools, delegate to subagents and keep going. The hard half is trust: knowing the business well enough to act unsupervised, and proving it with audit trails, approvals and the ability to say “I don’t know.”
Genie Ontology and Genie Search are aimed squarely at the hard half, and the 91% benchmark result suggests the approach works when the ontology is mature. But the benchmark was run in a workspace with accumulated business knowledge, and Databricks’ own caution says results will differ elsewhere. Enterprises adopting always-on Genie should start with narrow, well-defined recurring work, keep approvals on for consequential actions, and treat the ontology as a living asset that needs curation, not a switch to flip once.
Conclusion
Databricks’ always-on Genie is a genuine step change in how enterprise AI is framed: from a tool you consult to a coworker you assign work to. The combination of Genie Ontology, Genie Search and governed agent actions gives it a stronger foundation than most rivals can claim today. Whether it earns the trust to run unsupervised will depend less on the model than on the quality of the business context behind it, and on the discipline of the teams deploying it. For data leaders, the message is to start building that context now, because the agents are ready to use it.
Related reading: Claude’s move into spreadsheets, meetings and shared whiteboards.
