What is Action?
A single step within a workflow that performs a specific task. Actions can be AI-powered (like text generation, classification, or extraction) or utility-based (like data transformation, API calls, or conditional logic).
Think of it this way
"If a workflow is a recipe, an action is a single instruction like 'chop the onions' or 'simmer for 10 minutes.' Each action does one thing well, and you combine them to create the full dish."
Example
A 'Generate Text' action that creates product descriptions based on input specifications.
Marketer Use Cases
Text generation action to create dozens of ad copy variations from a single brief
Classification action to automatically categorize incoming customer feedback by sentiment and topic
Extraction action to pull key details (company, role, pain points) from sales call transcripts
Key Concepts
Input/Output
Every action accepts inputs (from previous steps or workflow inputs) and produces outputs that can be used by subsequent actions.
Configuration
Actions are configured with parameters that control their behavior—prompts for AI actions, field mappings for integrations, or conditions for logic.
AI Actions
Leverage language models to generate text, extract structured data, classify content, summarize documents, or answer questions.
Utility Actions
Perform operations like filtering data, transforming formats, making HTTP requests, or adding conditional branches.
Benefits
- •Modular design allows mixing and matching actions to build any workflow
- •Pre-built actions reduce development time for common operations
- •AI actions bring intelligence to automation without ML expertise
- •Reusable actions ensure consistency across workflows
Action vs Function
Actions are similar to functions in programming—they take inputs, perform operations, and return outputs. However, actions are configured visually and can be AI-powered without writing code.
Challenges & Solutions
Challenge: Action output doesn't match expectations
Solution: Use preview/test features to validate action output with real data before deploying
Challenge: AI action inconsistency
Solution: Refine prompts, add examples, increase specificity, or use structured output formats like JSON
Frequently Asked Questions
Can I create custom actions?
Yes, you can create custom actions using code blocks or by wrapping API calls, then reuse them across workflows.
How do I choose the right AI model for an action?
Consider the task complexity—simpler tasks work well with faster, cheaper models, while complex reasoning may require more capable models.