Skip to main content
Last updated on

Genie Code task for jobs

Use the Genie Code task to run an autonomous agent as part of a job. Genie Code works from a natural-language prompt and runs with auto-approve.

Beta

This feature is in Beta. Workspace admins can control access to this feature from the Previews page. See Manage Databricks previews.

Overview​

The Genie Code task type launches a new Genie Code chat with your prompt and produces a response. Genie Code reads data, calls tools, and acts on your prompt without requiring additional input or approval. After the run completes, you can open the chat and continue to interact with Genie Code.

Use a Genie Code task when you want to automate complex analysis, data operations, or decision-making as part of a scheduled job. For example:

  • Summarize overnight job results and email a report.
  • Analyze incoming data and flag anomalies.
  • Investigate a Jira ticket and propose a fix.
  • Generate a weekly compliance audit.

Requirements​

  • Your workspaces must have access to Genie Code's agentic capabilities. See Requirements.
  • Your task prompt must comply with Genie Code usage policies.

Configure a Genie Code task​

note

The jobs UI displays options dynamically based on other configured settings.

To add a Genie Code task:

  1. Navigate to the Tasks tab in the Jobs UI.
  2. Click Plus icon. Add task.
  3. Enter a name into the Task name field.
  4. In the Type drop-down menu, select Genie Code.

Configure the prompt​

Enter a natural-language prompt in the Prompt field. The prompt describes what you want the agent to do when the task runs. For example:

Check the most recent results in the batch_errors table. If error_count is above 5, create a summary of the errors by type and severity, then post a summary to #data-ops Slack channel.

(Optional) Configure parameters​

To reuse one prompt across runs with different values, reference job parameters in the prompt. Job parameters are pushed down to the task automatically, so you do not configure parameters on the task itself.

  1. Define job parameters as key-value pairs on the job, using Edit parameters in the Job details sidebar. See Configure job parameters.
  2. In the prompt, reference a parameter with {{name}}, where name is the job parameter's key.

For example, with a job parameter named region:

Summarize yesterday's sales for {{region}} and flag anomalies.

When the task runs, Databricks replaces each {{name}} placeholder with the parameter's value. A placeholder with no matching parameter is left unchanged. A job parameter's value can be a static value or a dynamic value reference, such as an upstream task's value. See Dynamic value references.

Finalize job configuration​

  1. (Optional) To configure retries, run-duration or streaming-backlog thresholds, or notifications, see Advanced task settings.
  2. Click Save task.

To edit, clone, disable, or delete this task, see Configure and edit tasks in Lakeflow Jobs.

How Genie Code job tasks work​

Autonomous execution with auto-approval​

When a Genie Code task runs, the agent executes your prompt autonomously. Auto-approval is always on for job tasks, so you do not confirm tool calls during the run. The agent uses an AI classifier to check each action against your prompt and blocks risky operations that fall outside the intended scope.

To understand how auto-approval works and its limits, see Approve tool actions.

warning

Auto-approve is a productivity feature, not a security boundary, and you cannot disable it for a job task. Because the task runs unattended, be deliberate about what production data, sensitive workspaces, or shared resources the prompt can reach.

Reading upstream task outputs​

The agent can read the outputs of upstream tasks in the same job as part of its work. Describe the upstream data you want it to use in your prompt, and the agent retrieves the relevant task outputs during the run.

Genie Code settings and MCP connections​

The task inherits your Genie Code settings and any custom Model Context Protocol (MCP) connections you have configured. The agent has access to the same tools and data sources you would use in an interactive Genie Code session.

Output​

Each task run produces a continuable Genie Code conversation thread. When the task completes, the Run output displays a link to the conversation, the task prompt, and the latest response. Click the link to open the thread and review what the agent did, ask follow-up questions, or continue the work interactively.

Limitations​

  • Auto-approval cannot be disabled: Job tasks always run with auto-approval on. You cannot turn it off or review actions before they execute.

Additional resources​

On this page