Databricks turns Genie into an autonomous enterprise AI coworker

Databricks says its revamped Genie assistant passed 91% of task runs on a new benchmark, against 58% for the stronger of two anonymized frontier agents. The company unveiled the new Genie on November 11, 2026, describing it as an autonomous AI coworker for continuous, long-running enterprise work. The claim matters because Databricks is betting that the next phase of enterprise AI isn't smarter chatbots — it's agents that own the work around an analysis for hours or days at a time.

What does the new Genie actually do?

The new Genie can run scheduled or event-triggered jobs, follow up on missing information and manage several assignments at once. Users decide which actions require approval. Databricks says thousands of customers already use Genie in production, and the new version is designed to take responsibility for the surrounding work: monitoring changes, requesting input and acting as new information arrives.

The company's own examples ground the pitch. Genie can route insurance claims that move outside underwriting guidelines. It can propose marketing budget changes for approval rather than executing them blindly. In both cases, the assistant prepares the decision and a human stays in the loop.

How does Genie find the right context?

At the center of the release sits Genie Ontology, 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. That distinction is the point: automatic extraction for speed, governed artifacts for trust.

Databricks' OntoRank system resolves the conflicts that inevitably arise. It weighs source authority, certification, user relevance, popularity and recency to select context and settle competing definitions of the same business term.

Genie Search is also generally available, covering Google Drive, Jira, Confluence and Workday, with SharePoint in beta. It indexes content from those systems and enforces their permissions before returning results. The index can be combined with SQL and MCP connectors, which retrieve live information and take actions in external tools.

Organizational directories — generally available and starting with Workday — add reporting relationships and collaboration signals to help rank relevant knowledge. Not everything ships at once, though:

• External BI knowledge sources are in beta, initially supporting Power BI.
• Ontology Manager, for reviewing and curating the ontology, is coming soon in beta.

What's new for agents and document work?

Alongside the core assistant, Genie Agents can now analyze PDFs, slide decks and other documents alongside governed tables in Public Preview, returning page-level citations. Native MCP connectors are generally available, letting agents retrieve context and take actions such as creating tickets, updating records and posting messages.

Two partner-facing capabilities arrive in beta. OpenShipping's sibling, OpenSharing, provides a governed agent preloaded with data, instructions and knowledge across Databricks accounts. Genie Clean Rooms take the opposite approach: recipients can ask questions without seeing the underlying data or proprietary logic, with conversation and output-size controls in place.

The scale of deployment is already visible at one customer. Databricks reports that General Motors has deployed more than 1,000 Genie agents across sales, engineering, supply chain and HR operations — the clearest illustration of the multi-department model the company is pursuing.

How much control do administrators keep?

The autonomy comes with Unity Gateway controls. Traces capture the question, the 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.

Identity stays with the humans. Genie acts under each user's identity and can prepare an access request, but administrators and data owners retain approval authority. That split — the agent drafts, the owner decides — runs through the entire design.

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 that date.

How credible is the benchmark?

To support its emphasis on context, Databricks released a benchmark developed with Turing. In the company's evaluation, Genie passed 91% of task runs versus 58% for the stronger of two anonymized frontier agents, while averaging 431,829 tokens and 100.3 seconds per run. Each agent ran every task five times in the same workspace: a fictional UK energy supplier containing tables, notebooks, dashboards, saved queries and documents, including deliberately outdated and duplicated assets.

The setup was designed to reward accumulated, governed business knowledge — exactly what Genie Ontology is built to supply.

Databricks itself flags the caveat, and it's a significant one. The test reflects an established deployment with accumulated business knowledge; differences may be smaller in new deployments or when table schemas alone are sufficient. A vendor-run benchmark with anonymized competitors always deserves scrutiny, and Databricks doesn't pretend otherwise.

Why the knowledge layer is the real story

That limitation 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. The benchmark gap — 91% against 58% — is less a statement about raw model capability than about what structured, certified context does for any agent working inside a company.

For Databricks, the strategic logic is straightforward. The company already owns the data platform where much of that context lives: Unity Catalog, dashboards, queries, notebooks. If autonomy is gated on governed knowledge, the vendor that hosts the knowledge holds the advantage. For buyers, the question flips: the value of an autonomous coworker will track the quality of the ontology underneath it, not the model on top.