AI Scheduled Tasks

Overview

AI Scheduled Tasks combine two things: a natural-language instruction and a repeating schedule. You describe the task the way you would ask a colleague to handle it, and DataWorks Data Agent repeatedly executes it in the specified environment on your chosen cadence, keeping the run history and output. The supported capabilities cover the conversational O&M, data query, and task inspection operations already supported by Data Agent—for example, rolling up yesterday's scheduled task runs, checking the production status of a given table, or sending on-demand notifications. Operations that require interactive input are not supported.

Core concepts:

• AI Scheduled Task: A record that contains a task name, a task description (natural-language prompt), an execution environment, and a schedule. Once started, Data Agent executes it automatically on the schedule.

• Execution Record: A single execution instance produced each time the task is triggered. It carries the instance ID, run status, start and end time, duration, and trigger type—the entry point for viewing results or troubleshooting.

• Trigger Type: Scheduled (automatically triggered when the schedule time arrives) or Manual (triggered from the task card or details header by clicking Run Now).

Difference from AI Inspection: AI Inspection targets standardized O&M health-check scenarios (baseline slowdown, workspace failures, change nodes, and so on), and its instruction is fixed by a template. AI Scheduled Tasks target custom scenarios where you write the task description yourself. They are ideal for turning routine actions that don't fit a template—such as rolling up a specific class of tasks from yesterday or notifying a particular business team on demand—into scheduled automation, with an upper bound aligned to Data Agent's capabilities.

Quick start

This section walks through the minimal loop of "create task → enable → run once to validate → view execution records", using Check whether yesterday's ODS synchronization tasks failed, every day at 09:00 as an example.

1. Go to the AI Scheduled Tasks page

   In the left navigation of DataWorks, choose AI Workbench > AI Scheduled Tasks to open the AI Scheduled Tasks page. Two tabs are shown at the top: My Tasks lists the task cards you have created; Execution Records aggregates the execution history across all your tasks. On first visit, the list is empty, and the Create Scheduled Task button is shown in the upper-right corner of the My Tasks tab.

2. Create a scheduled task

   Click Create Scheduled Task in the upper-right corner and fill in the drawer on the right, grouped by section:

   • Basic Information: Enter an easy-to-recognize Task Name (letters, digits, underscores, and hyphens only; up to 64 characters). Under Description, enter the natural-language instruction you want the AI to execute each time—for example, "Check yesterday's ODS-layer synchronization tasks, and summarize the failure reasons if any." The description is used as the prompt for the Data Agent.

   • Execution Environment: Choose the target Workspace the agent runs in (cannot be changed after the task is created). Set Resource Group to a general-purpose (Serverless) resource group.

   • Schedule Configuration: Schedule Time supports two frequencies:

     • Daily: Select a time (HH:mm). The task is triggered once a day at that time.

     • Hourly: Select a minute (0-59). The task is triggered every hour at that minute.

     The default is Daily at 08:00. In this example, choose Daily and set the time to 09:00.

   Click Create. The task is created in the Running state and is triggered automatically on the next scheduled time.

3. Run it once to validate the configuration

   On the My Tasks tab, locate the newly created task card. The card header shows the task name, the enable/disable switch, the schedule (for example, Daily 09:00), the creator, and the last modified time. Click the More menu in the upper-right of the card and choose Run Now to trigger a run immediately, without waiting for the next scheduled time. Use this to verify that the task description and execution environment behave as expected.

4. View execution records

   Click the task card to open the task details page and switch to the Execution Records tab. You can also switch to the top-level Execution Records tab on the list page to see the history for all tasks. Records are sorted by creation time in descending order. Key columns:

   Column	Description
   Instance ID	Unique identifier of the run. Can be copied.
   Status	One of Not Running, Waiting for Time, Waiting for Resources, Running, Failed, or Success.
   Trigger Type	Scheduled (triggered automatically by the schedule) or Manual (triggered by Run Now).
   Scheduled Time / Run Date	The schedule time and business date associated with this instance.
   Start Time / End Time / Duration	Actual execution time information from the agent. Each time column supports sorting.

   In the Actions column of a record, you can:

   • Run Log: Open a right-side drawer that streams the run log of the execution, so you can inspect the agent's process and output.

   • Rerun: Available for Failed or Success records. Triggers another run with the same parameters.

   • Terminate: Available for records in Running or Waiting states. Stops the current instance.

Daily task operations

After a task is created, you can perform the following actions from the task card or the task details header:

Changes made to scheduled tasks between 22:00 and 24:00, such as stopping or taking a scheduled task offline, take effect on the third day. Avoid performing operations on scheduled tasks during this period.

Besides the Execution Records tab, the details page also contains:

• Basic Information: Read-only view of the task's Basic Information, Execution Environment, and Schedule Configuration.

• Operation Log: A log of create, modify, start, and stop actions on the task.

Notes

• The task description is used as the prompt for Data Agent. Use clear natural language to describe what to do, which objects to act on, and what output is expected—including specific table names, task names, or workspaces—to avoid ambiguity.

• Only two schedule frequencies are supported today: Hourly and Daily. More complex schedules (for example, weekly or by business calendar) are not supported.

• Editing and deleting are only available after the task is stopped. The Workspace is a task-level attribute and cannot be changed after creation—create a new task if you need to switch workspaces.

• Each execution calls a large model and uses resource-group capacity, which consumes tokens and CUs. Choose a schedule frequency that fits your usage.

• AI Scheduled Tasks are not DataStudio cyclic tasks and do not participate in baseline assurance. To track a class of tasks with baseline assurance, use DataStudio cyclic tasks or the Baseline Slowdown Inspection template in AI Inspection.

Related capabilities

• For standardized baseline or workspace health checks, use AI inspection.

• For deep diagnosis of a failed instance with one-click O&M actions, see AI-powered O&M.