Automating tasks from Task Mining

  • Release version: Australia
  • Updated June 17, 2026
  • 4 minutes to read

    • Automation Center can receive task recordings from Task Mining, decompose them into discrete automations, and generate an AI agent in AI Agent Studio that executes those automations using AI Desktop Actions.

    About task automation

    Automation Center reads the task recording summary, breaks it into modular automations, and generates an AI agent — without requiring the user to manually author agent instructions or AI Desktop Actions tools from scratch.

    This feature currently supports only Excel-to-browser interactions, where data is copied from a Microsoft Excel file and it is entered into a form in a browser.

    Important:

    This feature uses generative AI to produce its output. AI-generated content, including decomposed automations and agent instructions, may be inaccurate or incomplete. Review and validate AI-generated output before activating an agent.

    Key benefits

    Automating tasks from Task Mining provides the following benefits:

    • Reduces the effort required to automate repetitive desktop work by generating agent instructions from recorded task data.
    • Automatically identifies and categorizes steps as UI-based or background interactions, reducing manual analysis by the technical user.
    • Technical users can refine AI-generated automations before creating an agent so the output reflects the actual task requirements.
    • Creates a reusable AI agent that any authorized user can invoke from the Now Assist panel without accessing Automation Center or AI Agent Studio directly. The agent can also automatically run if a trigger is set and it meets the trigger criteria.

    Task automation workflow

    The following graphic describes the end-to-end workflow for automating tasks from Task Mining:

    Figure 1. Automating tasks from Task Mining
    Workflow diagram showing the end-to-end process from Task Mining recording to AI agent invocation in the Now Assist panel
    1. A repetitive task in Task Mining is recorded (Business user). When the recording is complete, the user submits an automation request. The task timeline and a summary of user interactions are saved as an attachment to the automation request record.
    2. The automation request appears in Automation Center. A technical user reviews the request. There is no automated notification that informs the Automation Center user about the creation of the automation request in Task Mining.
    3. The automation request in Automation Center is opened (Technical user) and Generate automations is selected on the Automations tab. Automation Center reads the attachment and uses AI to decompose the task into discrete automations.
    4. Each decomposed automation is categorized as one of the following:
      • UI block: An on-screen interaction performed in a browser, such as entering data into a ServiceNow form. Automation Center generates a AI Desktop Actions UI block for these automations using screenshots from the task recording.
      • Non-UI block: A background interaction with an application that does not require on-screen control, such as reading data from a Microsoft Excel file. Automation Center creates a tool with instructions for the existing background application handler.
    5. The decomposed automations are reviewed (Technical user). If the output is unsatisfactory, the user can provide additional instructions and regenerate. Because the decomposition is AI-generated, results may vary between runs.
    6. When satisfied with the decomposed automations, the technical user selects Create agent. Automation Center sends the automation details to AI Agent Studio, which creates an agent pre-populated with instructions, a role description, input parameters, and AI Desktop Actions tools.
    7. The technical user reviews the generated agent in AI Agent Studio, corrects any inaccurate instructions, and tests the AI Desktop Actions tools. UI block anchors — the UI elements that AI Desktop Actions uses to identify where to click or type — may require manual adjustment because Task Mining and AI Desktop Actions capture anchor data differently.
    8. After testing and validation, the technical user activates the agent and assigns it to the appropriate roles. Authorized users can then invoke the agent from the Now Assist panel by describing the task they want to perform. You can also schedule the agent to run when some triggers are met. The agent will run the task as scheduled.

    Limitations and requirements

    Consider the following when using this feature:

    • Windows requirement: Task Mining recording and AI Desktop Actions agent execution require a Windows machine. Automation Center configuration and agent creation can be performed on any supported operating system.
    • Supported scenario: This feature currently supports only Excel-to-browser interactions, where data is copied from a Microsoft Excel file and it is entered into a form in a browser.
    • AI-generated output variability: Decomposed automations and agent instructions are generated by AI. Each run may produce different results. Review all AI-generated output before creating or activating an agent.
    • UI-block anchor adjustment: AI Desktop Actions UI blocks require anchors that identify the UI element to interact with. Anchors captured from Task Mining recordings may not match the format AI Desktop Actions expects and may require manual correction in the AI Desktop Actions editor.
    • No cross-product notification: Automation Center users don't receive a notification when a Task Mining user submits an automation request. The business user and the technical user must coordinate directly.
    • Plugin dependencies: Both Automation Center and AI Desktop Actions must be installed on the target instance before you can create an agent. Now Assist for Automation Center must be installed to use this feature.
    • AI skill dependencies: User task step summarization skill must be activated. For more information, see Activate skills for Now Assist for Automation Center.