Since its launch two years ago, the Databricks Assistant has become an indispensable partner for data practitioners, helping them generate SQL and Python code, resolve errors, and receive contextual guidance directly within their workflows. Over that time, the AI landscape has advanced rapidly. The frontier has shifted from simple copilots and chatbots to agents that can reason, plan, and autonomously execute complex, multi-step processes.
Extending this paradigm to data requires more than fluency in code. Enterprise data agents must be aware of the context of your data, enable you to review and refine their work, and operate with the highest standards of governance. Databricks is uniquely positioned to deliver on this vision. With Unity Catalog providing unified policies, lineage, and business semantics, the platform is already the trusted foundation for data intelligence. Building on that foundation, agents can compress the time from question to insight without compromising on transparency, trust, or rigor. That is the future we are now bringing to the Databricks Assistant.
Bringing Agents to Databricks Assistant
We are proud to introduce the Data Science Agent, a major advancement that elevates the Databricks Assistant from a helpful copilot into a true autonomous partner for data science and analytics. Fully integrated with Databricks Notebooks and the SQL Editor, the Data Science Agent brings intelligence, adaptability, and execution together in a single experience. It is the first of a new generation of AI data agents available by selecting Agent Mode in the Assistant, and it will begin rolling out to customers in the coming days.
The Data Science Agent builds on everything you already do with Databricks Assistant today and massively accelerates your work when you hand it higher-level tasks. Here are just a few ways it can help your day-to-day:
- Exploring data: You can ask the agent to “perform exploratory data analysis on @table to identify interesting patterns”. You can provide additional guidance if you want to focus the exploration on a particular area. The “@” capability is an existing Assistant capability, making it easier to indicate to the Assistant the specific table you are referencing.
- Training and evaluating ML models: The agent can perform machine learning tasks, using MLflow capabilities as needed. For example, you can ask the agent to “train a forecasting model predicting sales in @sales_table”. You can then guide it to use specific model types or how much to focus on hyperparameter tuning.
- Fixing errors: People love the Assistant’s diagnose error button. In agent mode, the diagnose error capability can help you make additional updates and iteratively try the fix until the issue is resolved.
- Summarizing and explaining results: You can ask the agent to explain and summarize the results of your analysis or carry out further analysis.
- Finding relevant data: The agent can help you find the data you need to complete your task in Unity Catalog by searching tables you can access. Try to describe in detail what you are looking for, such as the column names or the type of data. The Data Science Agent will be more helpful for this if your tables and columns have descriptive comments.
Accurate, trustworthy responses
Our goal with the Data Science Agent is to deliver a data science and analytics experience you can trust, with answers that are accurate, relevant, and grounded in your organization’s data. This is a difficult problem, even for frontier AI models, which on their own don’t understand the semantics of your data, your business logic, or the way your teams work. The Data Science Agent bridges this gap by combining the reasoning power of AI models with the Databricks Data Intelligence Platform, ensuring results that are both reliable and context-aware. For example, it can search Unity Catalog to surface the right tables and notebooks and interpret results to suggest the best next steps, such as refining an analysis, training a model, or summarizing findings for stakeholders. By grounding agentic workflows in a governed context, the Data Science Agent turns raw automation into trustworthy acceleration.
Getting started
Workspace admins can enable the Assistant agent mode beta from the Databricks preview portal.
Once your admin enables agent mode, you’ll see a toggle in the bottom-right corner of the Assistant. Switch it to Agent, type your task, and let the agent take it from start to finish. For multi-step or more complex requests, we recommend trying out Planner for added transparency and control.
Using planner for more complex workflows
The agent’s planner capability helps you handle complex workflows by drafting a plan before execution. Toggle it on at the start of an Assistant thread, and the agent will propose detailed steps, asking clarifying questions as needed, then refine the plan based on your input. Once it looks right, click Continue, and the agent will execute it step by step, reviewing results with you along the way and summarizing the outcomes at the end.
The planner is especially valuable when the task spans multiple steps or requires careful orchestration. For example, in a churn investigation, you may want to guide the agent through dataset exploration, cohort analysis, and visualization. Or, when building an ML pipeline, the planner can help structure data cleaning, feature engineering, model training, and evaluation into a coherent flow.
Tool confirmation
You stay in the driver’s seat. Before running code, the agent asks for your approval. You can choose to:
- Allow once: approve a single execution
- Always allow for this thread: streamline work within the current Assistant conversation. This resets when you press the “+” at the top right corner of the Assistant panel.
- Always allow: give approval until you change the setting
In addition, the agent has built-in guardrails to help reduce unintended actions, such as accidentally dropping a table. That said, we still recommend reviewing generated code carefully, especially when it touches production data, important tables, or other sensitive operations.
On the horizon
Looking ahead, we’re investing in several improvements to make the Data Science Agent even more powerful:
- Broader context: Bring in additional context through MCP integration. This will provide the Assistant with new knowledge it doesn’t have today.
- Smarter memory: Assistant instructions are already used by the Data Science Agent, but we want the agent to make it even easier to update and curate your instructions
- Faster data discovery: the Data Science Agent can help you find the assets you need for your task. It takes a first step today with its ability to search tables and code, but we’re working on improving this area.
The Data Science Agent is just the beginning. Agent mode will grow to orchestrate entire workloads across Databricks. We’re building towards agent workflows for data engineering and beyond, all powered by the same trusted, governed foundation.
Try the Data Science Agent today 🚀
Check out our product page to learn more about Databricks Assistant, or read the documentation for more information on all the features.
Ask your admin to enable Databricks Assistant Agent Mode today, and start turning hours of work into minutes. This will give you more time for insights and less time for mechanics.
