We’re excited to introduce Agent mode in Genie areas. Our group has developed a robust agentic course of that iteratively plans, explores, and causes over your information to reply your enterprise questions. As a part of Databricks’ Week of Brokers, this weblog highlights one other means we’re altering how organizations work together with their information utilizing agentic AI.
This expertise unlocks a way more actionable degree of knowledge evaluation for everybody in your group. Now, anybody can get real-time insights into advanced enterprise questions like:
- Why did our churn price spike in Q3?
- How can we optimize our marketing campaign spend?
- What income affect ought to I anticipate if these two provide traces are interrupted?
How Agent Mode Works
While you ask a query in Agent mode, Genie doesn’t simply return a single question. It investigates the issue like an actual information analyst: planning an method, testing hypotheses, and iterating towards a proof.
For instance, think about a Genie house for buyer help. You discover a spike in reopened circumstances in December 2025 and ask: “What’s contributing to the spike in reopened help circumstances?”
Agent mode first confirms the spike, then explores potential contributors comparable to clients, merchandise, classes, or groups. It makes use of the enterprise context in your Genie house, together with Unity Catalog metadata and author-defined semantics, to deal with probably the most related potential contributing elements.
Agent mode then evaluates these hypotheses by executing a number of queries towards the underlying information. The semantics outlined in Genie’s data retailer train it the best way to produce correct queries. We additionally made Agent mode’s work clear so customers can at all times confirm its accuracy.
Throughout its evaluation, Genie constantly displays on the outcomes of every question and decides what to discover subsequent. On this instance, after testing a number of potential drivers of the spike, the agent decides that it ought to additional examine if seasonal patterns are contributing to the spike. This iterative cycle of speculation technology, querying, and reflection permits Genie to discover information extra completely and arrive at a well-supported clarification.
After finishing its evaluation, Genie generates a report of its findings. Following this instance, the report first quantifies the rise in reopened circumstances after which identifies the first contributors—specifically an increase in bug-related circumstances and efficiency from the L2 regional group. To help these conclusions, the report additionally contains visualizations and references to the underlying SQL for customers to evaluation.
Relying on the kind of query, Genie additionally gives actionable suggestions on what groups ought to deal with to enhance efficiency. Customers can then share these reviews immediately within the platform or obtain them as PDFs to simply distribute insights and collaborate.
Constructed for Questions of Any Complexity
Agent mode not solely unlocks superior enterprise investigations—it improves accuracy throughout all query sorts, from easy analytics to multi-step evaluation.
Even for easy questions, it takes small validation steps to make sure it understands the info earlier than responding. We’ve additionally tuned the agent to dynamically scale its reasoning to the complexity of the duty—transferring rapidly for easy prompts, and spending extra time planning and evaluating for deeper investigations.
The result’s quicker solutions for on a regular basis questions and extra rigorous evaluation when tackling advanced issues.
Customers interacting with Genie inside AI/BI Dashboards also can make the most of this new expertise. While you ask inquiries to Genie from a Dashboard, it leverages Agent mode by default.
Get Began In the present day
Workspace admins can now affirm Agent mode is enabled within the Workspace Previews web page. As soon as enabled, merely activate the Agent toggle in your Genie areas and ask your enterprise questions.
Agent mode is now out there, with way more to return together with API help and unstructured doc evaluation. Give Agent mode a strive in the present day — we are able to’t wait to listen to what you assume.
