Creating AI agents for AI Desktop Actions

  • Release version: Australia
  • Updated November 2, 2025
  • 2 minutes to read
    • Summarize
    Summarized using AI
    This content was generated using new OpenAI-powered functionality. Results are provided on an as is basis and are not guaranteed to be accurate or complete.

    Summary of Creating AI agents for AI Desktop Actions

    In the ServiceNow agentic ecosystem, AI agents are sets of large language model (LLM) instructions and tools designed to perform specific tasks autonomously or semi-autonomously across web and desktop environments. These agents mimic human intelligence to execute repetitive desktop actions in legacy systems, thick client applications, and web applications without APIs or backend integrations. They generate dynamic execution plans, coordinate subtasks, process human input during execution, and collaborate with users for decisions requiring human judgment.

    Show full answer Show less

    Key Features

    • Agent Creation: Use AI Agent Studio to create AI agents by defining a clear name, description, role, and the steps the agent must complete. Supported LLMs can be configured, along with third-party access and long-term memory management.
    • Tool Integration: Add desktop action tools to AI agents to enable automation capabilities. Tools must have clear names and descriptions to ensure the agent selects the appropriate tool for executing tasks. Both defined desktop actions (for desktop and web tasks) and adaptive desktop actions (for web-based tasks) can be added.
    • Security Controls: Configure who can access and invoke the AI agent through Access Control Lists (ACLs), and define the data the AI agent can access once activated. These controls ensure secure operation and data privacy. Note that trigger conditions are not supported; agents must be manually triggered within the AI Desktop Actions application.
    • Channel Selection and Activation: Activate AI agents for use within Now Assist for Virtual Agent. Agents can engage users across selected communication channels, with customizable processing messages to indicate active engagement.
    • Default Agents and Workflows: The Web Automation Agent and associated agentic workflow are provided out-of-the-box for adaptive desktop actions upon installation of AI Desktop Actions.

    Practical Impact for ServiceNow Customers

    By creating AI agents tailored to your organization’s desktop and web automation needs, you can significantly reduce manual repetitive tasks, improve operational efficiency, and enable collaborative human-AI workflows. The ability to integrate desktop action tools empowers these agents to work across systems lacking traditional integration methods. Security controls ensure that AI agents operate within defined access and data boundaries, maintaining compliance and governance standards. Activation within Now Assist allows seamless user interactions through familiar channels, enhancing user experience and productivity.

    Create an AI agent in AI Agent Studio to mimic human-like intelligence while executing desktop actions for repetitive tasks in web and desktop environment.

    Overview of AI agents for AI Desktop Actions

    In the ServiceNow agentic ecosystem, an AI agent is a set of large language model (LLM) instructions and tools that can perform specific tasks. The Now Assist AI agents can perform specific tasks and functions, often using natural language instead of traditional code. For more information, see Create an AI agent.

    AI agents process instructions, generate execution plans, and run desktop actions autonomously and semi-autonomously across legacy systems, thick client applications, and web applications lacking APIs or backend integrations. AI agents can interpret your goal and map them to one or more desktop actions via metadata (capabilities, inputs, and outputs).

    Use AI agents to do the following tasks for your organization:
    • Generate a dynamic execution plan for desktop or web-based tasks
    • Coordinate with other AI agents to complete subtasks
    • Process human input during task execution when required
    • Collaborate with users to resolve issues that require human judgment

    For adaptive desktop actions, an AI agent named Web Automation Agent and agentic workflow named Web Automation are provided by default when you install AI Desktop Actions.

    1. Define the specialty of an AI agent

      Write a clear name and description of the AI agent, define role, and list of steps this AI agent must complete, define supported LLMs, enable third-party access, and manage long-term memory. The LLM analyzes the specific wording that you use to understand the specialty of this AI agent.

    2. Add a desktop action to an AI agent.

      Add a desktop action as a tool to the AI agent to enable desktop and web automations. Tools provide your AI agents with the capabilities necessary to complete their tasks. Providing your AI agents with the appropriate tools help with the robustness and quality of their performance. An AI agent selects a tool based on the tool’s name and description, which must be clearly written.

    3. Define security controls for an AI agent
      Define security controls for who can access the AI agent and what data the AI agent has access to. The Define security controls step is divided into two parts:
      • Define user access: Creates an ACL that determines which users can discover or invoke the AI agent.
      • Define data access: Defines the data that the AI agent has access to once it’s invoked.

      Trigger conditions are not supported for AI agents that execute desktop actions. You must manually trigger these agents from the system where the AI Desktop Actions application is installed.

    4. Select channels and status for an AI agent

      Activate the AI agent to use in an assistant in Now Assist for Virtual Agent and set the processing message. This AI agent can engage with users who initiate an interaction when it’s available for use in channels. Select channels where you want this AI agent to be available to engage with users.