At Databricks Data + AI Summit, Databricks announced Genie Agents as part of its wider Genie One suite.
At first, I had the obvious question: what does this actually do that Genie One didn’t already do?
The shift is that Databricks is moving beyond simple dashboard Q&A. Genie Agents is about helping business teams create governed AI coworkers that understand company data, repeat trusted workflows, and take action across business tools.
That is what makes it interesting. Most enterprises already have the data and infrastructure. Genie Agents is trying to make that data more useful in the flow of work, not just answering questions, but helping teams monitor, explain and act on what is changing.
Think of Genie Agents as domain-specific AI agents built on top of Genie, Genie One and the new Genie Ontology.
The older Genie Spaces were mainly about giving business users a governed way to ask questions over company data. Genie Agents takes that further. Instead of only answering questions, they can reason through tasks, use tools, work across systems, and be shared by teams.
According to Databricks, Genie Agents can support things like MCP connections, scheduled tasks, document generation, and writing back to external systems such as Slack or Teams. In simple terms, they are designed to move from “answer this question” to “help me run this workflow”.
The most interesting part is that a Genie conversation can be saved as a reusable agent. That agent keeps the sources, instructions, memory and behaviour from the original conversation, so other people in the team can use it again.
For example, imagine you ask the same three-part question every month before a leadership meeting. Instead of rebuilding that prompt each time, you could turn it into an agent. You and your team could then call that agent whenever needed and get a consistent, trusted output.
That is the real value: teams can get more leverage from their data without everyone needing to understand the underlying tables, dashboards or queries.
There are three related ideas here, and they are easy to mix up.
|
Concept |
What it means |
What you can test |
|
Agent Mode in Genie Spaces |
A preview feature where Genie behaves more like an analyst. It creates a plan, runs multiple SQL queries, tests ideas, and produces a report with citations, visuals and supporting tables. |
This is the most practical place to start if you want to test the experience inside Databricks. |
|
Genie Agents |
The broader direction Databricks announced at DAIS: reusable agents that can keep context from a conversation, connect to tools, schedule tasks, generate artefacts, write to external systems, and work across structured and unstructured business context. |
Some parts may depend on Genie One access, workspace settings, or Databricks account enablement. |
|
Agent Bricks |
A more developer-facing platform for building, evaluating, deploying and governing custom agents with models, tools, memory and production controls. |
Useful if you want to build more custom agentic systems beyond the Genie business-user experience. |
The simple distinction is Agent Mode is what you can test now, Genie Agents is the business-facing agent experience Databricks is building towards, and Agent Bricks is for teams building more custom agents from the ground up.
The interesting thing about Genie Agents is that the use cases can cut across almost every business team. Each team has different questions, but the pattern is the same: they want faster, trusted answers from the data they already have.
The main users are likely to sit across the whole organisation. A finance team may want a monthly variance agent. A sales leader may want a pipeline risk agent. An operations team may want an agent that monitors delays and explains what is driving them.
The rollout model is where it gets important. Data and BI teams will likely help operationalise the agents by making sure the right data, logic and permissions are in place. Data leaders will need to approve what gets shared across teams. Business leaders then become the end users, using the agents to get better outputs without needing to understand every table, dashboard or query underneath.
There are five main reasons Genie Agents deserved their own announcement slot at DAIS, separate from Genie One and Agent Mode.
1. Grounded in governed enterprise data
This is probably the biggest difference. Genie Agents are not sitting outside the Databricks ecosystem as a separate chatbot layer. They are built around Databricks’ existing governance model.
This matters because enterprise AI is not just about giving a model access to data. It is about making sure the model knows what data is trusted, who is allowed to see what, and which definitions should be used.
2. Understands business context, not just tables
Most AI agents can query data if you connect them to the right tools. The harder part is understanding what the data actually means. For example, “revenue”, “active customer”, “pipeline”, or “margin” can mean different things in different companies. Genie Ontology is Databricks’ attempt to capture that business context: metrics, definitions, relationships, calculations and trusted sources. This is what should make Genie Agents more useful than a manually wired agent that simply runs SQL.
3. Turn good analysis into repeatable workflow
This is the part I find most interesting. If a useful Genie conversation can be saved as an agent, then a good piece of analysis does not have to stay as a one-off interaction. It can become a repeatable workflow with the same sources, same logic, same instructions and same output style.
That is powerful because a lot of business work is repetitive. Monthly reviews, sales updates, risk checks, performance summaries and operational reports often follow the same pattern.
4. Connects into broader business tools
Databricks is also positioning Genie Agents as more than data Q&A. The idea is that they can connect with tools like Slack, Teams and Gmail, support scheduled tasks, generate documents, trigger alerts, use custom skills and connect through MCP.
That moves the experience closer to the actual flow of work, rather than keeping it trapped inside a dashboard.
5. It sits closer to the data platform than generic AI Agents
Claude, ChatGPT and Gemini are strong general-purpose assistants. But they do not automatically know your Unity Catalog permissions, certified tables, dashboard logic, metric definitions, lineage, row filters or column masks.
That is Databricks’ real commercial pitch: AI agents over enterprise data without rebuilding the governance layer from scratch.
Genie Agents feels like Databricks moving from BI into business automation. The value is not just that users can ask better questions. The bigger idea is that teams can turn repeated analysis, reporting and decision workflows into governed agents that understand the company’s data and context.
That's why this announcement matters. If Databricks can make this simple enough for business teams, while keeping governance, permissions and trusted definitions in place, Genie Agents could become a practical bridge between enterprise data and day-to-day work.