The Agentic Interaction step combines the capabilities of the User Interaction and LLM Interaction steps into a single, powerful component. It enables you to create dynamic, conversational workflows that can gather information, execute tasks, and respond naturally to users—all within one step.
Overview
The Agentic Interaction step is ideal for scenarios where you need to collect one or more pieces of information through natural conversation, rather than rigid, sequential prompts. It's particularly useful for:
-
Gathering multiple data points in a single, fluid interaction
-
Handling complex user inputs that may require clarification or follow-up
-
Creating human-feeling experiences for business-critical flows
The step includes three main sections:
-
Prompt & Styleguide: Define what the interaction should accomplish and how it should behave.
-
Completion Criteria: Specify when the interaction should end.
-
Outputs: Define the variables the step should produce.
Key Components
Prompt
The Prompt defines the task or goal for the agentic interaction. Write clear instructions describing what the step should accomplish.
Example:
Collect the user's shipping information for their order. You need their full name, street address, city, postal code, and country. If any information seems incomplete or unclear, ask for clarification.
Prompt Tips:
-
Describe the overall goal of the interaction
-
List the information you need to collect
-
Include any validation rules or constraints
-
Specify how to handle edge cases or unclear responses
Styleguide
The Styleguide controls the tone and behavior of the interaction. Use this to ensure the conversation aligns with your brand voice and user expectations.
Example:
Be friendly and professional. Use short, clear sentences. If the user seems frustrated, acknowledge their concern before continuing. Do not use emojis.
Completion Criteria
Define when the agentic interaction should complete and hand control back to the workflow. Be explicit about the conditions that signal the interaction is finished.
Example:
Complete when all required shipping fields have been collected and confirmed by the user, or when the user explicitly asks to cancel.
Context Access
The Agentic Interaction has full awareness of the conversation and workflow state:
-
Context Variables: Access variables set at the conversation level.
-
Tool Variables: Access variables defined within the current tool.
Use the / command palette to inject these values into your prompt, styleguide, or completion criteria.
Outputs
The Outputs section defines the variables this step will produce. Each output includes:
|
Field |
Description |
|---|---|
|
Name |
The variable name used to reference this output in subsequent steps |
|
Type |
The data type (e.g., string, number, boolean, object) |
|
Description |
Instructions for when and how this variable should be set |
|
Required |
Whether the interaction must set this variable before completing |
Example Configuration:
|
Name |
Type |
Description |
Required |
|---|---|---|---|
|
|
string |
The user's complete name as provided |
Yes |
|
|
string |
Full address including street, city, postal code, and country |
Yes |
|
|
string |
Any special delivery instructions the user mentions |
No |
Tip: Write clear descriptions for each output. The agentic step uses these descriptions to understand when and how to populate each variable.
When to Use Agentic Interaction
|
Use Case |
Recommended Step |
|---|---|
|
Collecting a single, simple input |
User Interaction |
|
Processing data without user involvement |
LLM Interaction |
|
Gathering multiple inputs through natural conversation |
Agentic Interaction |
|
Complex flows requiring clarification or follow-up |
Agentic Interaction |
|
Business-critical data collection with validation |
Agentic Interaction |
Best Practices
-
Be specific in your prompt: Clearly describe the goal and the information needed. Vague prompts lead to unpredictable behavior.
-
Define clear completion criteria: Ambiguous completion conditions can cause the interaction to end prematurely or continue indefinitely.
-
Use required outputs thoughtfully: Mark outputs as required only when the workflow cannot proceed without them.
-
Leverage the styleguide: Consistent tone improves user experience, especially for customer-facing workflows.
-
Keep it focused: If your agentic interaction is handling too many unrelated tasks, consider splitting it into multiple steps for better maintainability.