Use Start a workflow to have an AI Agent launch a workflow from a selected application. Choose the form fields the agent can prefill and decide whether it needs a response from the workflow.
Accessing the Tools Page
Open the AI Agent and select Tools in the top navigation. From here you can add, edit, enable, disable, or remove tools.
Adding the “Start a workflow” Tool
Select Add a tool → Start a workflow.
Choose the application and select the target under Approval form / Workflow form.
Review the tool name. Select Add to create it or Add and configure to set up inputs and outputs.
General Settings
Under Details, confirm the name, description, tool type, application, and workflow form. Set Credentials to use to End user credentials for the caller’s access or Specific user for the selected user’s access. Check that the execution identity can start the workflow.
Inputs – Optional fields used to prefill workflow data
The Inputs list starts empty. Select + Add input to choose which workflow form fields the AI Agent can prefill.
Select inputs
The picker shows each available field’s Input name, internal ID, and Input type. Check the fields you want the agent to prefill, then select OK. Use the internal ID to distinguish fields that share a label.
Inputs table configuration
Each selected field will appear in the table with the following columns:
Column | Description |
Input name | Workflow form field to prefill |
Input type | Field type inherited from the workflow form |
Fill using | Choose AI-generated input or Custom value |
Value | Required only when using a Custom value |
Notes:
Prefilling is optional. The tool can prefill only the fields listed under Inputs. To stop prefilling a field, remove its input row.
For each row, choose Dynamically fill with AI to extract a value from the conversation or Custom value to use your own mapping. Configure Value for custom mappings. Compare the input options
Completion
Set whether the AI Agent waits for data from the workflow before continuing.
After running
Waiting for response waits for the workflow to return data at a Respond to AI node.
With the non-waiting option, the agent continues after starting the workflow and the workflow runs in the background. Use this when the agent can proceed without the workflow result.
Relationship with Respond to AI
If the calling agent needs data, put a Respond to AI node on the workflow path that should send it, and use Waiting for response. The node returns data when the workflow reaches it. For example, an early node can return a request ID while the approval continues. Configure the Respond to AI action
Outputs available to the agent and other tools
The tool starts with no outputs. Add mappings for the values the agent or a later tool needs, such as a workflow instance identifier or a response message.
Managing outputs
Select Add output, give it a name, choose the matching type, and select Customize to map its value. You can remove an unused output and add it again later.
Summary
In a test workflow, map two form fields, such as Asset ID and Inspection reason. Start it through the agent with a disposable example. Open the new workflow instance and check that both fields contain the expected values.
If the agent needs a result, set Waiting for response and map an output from the workflow. Check that the expected value arrives when the instance reaches Respond to AI.
If the agent should continue without waiting, test that it can proceed and find the workflow instance independently to verify it started.

