Use an AI agent action
Summarize
Summary of Use an AI agent action
This action enables ServiceNow customers to integrate AI agents into Flow Designer workflows by using flow data to run an AI agent and configure its expected outputs for subsequent use within the flow. It is available as a Flow Designer Gen AI spoke action and requires the Now Assist AI agents plugin. Authorized users with theflowdesigneroradminrole can add and configure this action in a flow.
Show less
Key Features
- Inputs: Define inputs such as the AI Agent reference (must be created and configured beforehand), Support User (for supervised mode), Objective (text instructions for the AI agent), Execution Mode (Autonomous or Supervised), Wait for Completion option, Expected Outputs (dynamic objects to capture AI responses), Conversation Label, and Context Memory to guide AI behavior.
- Execution Modes:
- Autonomous mode: Runs the AI agent without user interaction.
- Supervised mode: Requires user input or approval, pausing the flow until the designated support user reviews and authorizes the AI agent’s work.
- Outputs: Provides several outputs usable elsewhere in the flow, including Execution Plan Record, Agent Output (expected output values), Agent Message, and Agent Status (success or failure of the AI run).
- Context Memory: Enables sending contextual information to the AI agent to influence its output generation.
Practical Application Example
An example use case involves the Incident resolution details AI agent, requiring the Now Assist for IT Service Management plugin and enabling the Resolution notes generation skill. In supervised mode, the flow pauses for a support user to interact via the Now Assist panel, authorizing the AI’s output. Once approved, the flow resumes and uses the AI-generated outputs, such as resolution details, to perform subsequent actions like sending email notifications.
Why It Matters and What to Expect
This action empowers ServiceNow customers to automate complex tasks by leveraging AI agents directly within flows, enhancing efficiency and accuracy. The supervised mode ensures control and compliance by involving human oversight when necessary. Customers can expect to seamlessly integrate AI-driven insights and outputs into their business processes, improving decision-making and operational workflows.
Use flow data to run an AI agent and configure the expected agent output for use later in the flow.
Roles and availability
Available as a Flow Designer Gen AI spoke action, which requires the Now Assist AI agents plugin. Users with the flow_designer or admin role can add an action to a flow and define configuration details.
Inputs
Provide a value for each input that your action needs. To add dynamic values, you can also select data pills using the pill picker.
- AI Agent
- Data type: Reference
The AI Agent that you want to use in this flow. You must install the Now Assist AI agents plugin and create and configure an AI agent. For information about installing the Now Assist AI agents plugin, see . For information about creating an AI agent, see .
- Support User
- Data type: Reference
The user who is contacted by Now Assist to review and authorize the AI agent's work when the AI agent runs in supervised mode. Select a user who meets the security criteria defined in the AI security controls. If the AI agents is configured to accept a dynamic user, select a user that has the roles necessary to run the AI agent. For more information about the security controls of an AI agent, see Define security controls for an AI agent. For information about turning on the Now Assist panel, see Activate the Now Assist panel standard chat.
- Objective
- Data type: String
The text directions sent to the AI agent.
- Wait for completion
- Data type: True/False
Option to pause the flow until the AI agent finishes running and provides output values.
- Execution mode
- Data type: Choice
Option to run an AI agent with or without user input or approval. In Autonomous mode, the action runs the AI agent without any user input or approval. In Supervised mode, the action requires user input or approval to run the AI agent.
- Expected outputs
- Data type: Dynamic Object
List of output objects that the AI agent is expected to return. Each expected output is an object that has a label, name, and a data type value. You can use the expected outputs to gather data from the AI agent and then use that data elsewhere in the flow.
- Conversation label
- Data type: String
The name that identifies the conversation about running the AI agent in the Now Assist panel. For information about activating the Now Assist panel, see Activate the Now Assist panel standard chat.
- Context Memory
- Data type: String
The context information that you want to send to the AI agent. This context helps determine what output values the AI agent produces.
Outputs
You can use these outputs as inputs for other items.
- Execution Plan Record
- Data type: Reference
The record containing the Objective, Execution Tasks, Messages, and Tools Executions run by the AI agent.
- Agent Output
- Data type: Dynamic Object
The object containing the list of expected output values generated by the AI agent.
- Agent Message
- Data type: String
The message returned by the AI agent.
- Agent Status
- Data type: String
The run status of the AI agent. The AI agent returns success when it runs or failure when it doesn't run.
Use Incident resolution details AI agent
This example requires installing the Now Assist for IT Service Management (ITSM) plugin and turning on the Resolution notes generation skill.
In this example, the Use an AI agent action is configured to use the Incident resolution details AI agent in a supervised mode. The objective lists data that will help the AI understand the resolution of an incident.
Since this example runs in supervised mode, the flow pauses until the support user interacts with the conversation generated in the Now Assist panel.
After the support user provides the necessary response, the flow continues running the action and provides it output values. The flow uses these output values to send an email message.