Skip to main content
The Respond step follows your instructions to perform reasoning on the contents of the agent’s memory and create a text response that is shown to the user of the agent. The response is also added to the agent’s memory so that later steps can reference it to help accomplish the agent’s objective. Use the Respond step any time you want the user of an agent to see the agent’s output on the screen. It is important to include a Respond step immediately before write actions like Draft Gmail, Create a Google Doc, and others. This allows the actions to use exactly the content you specify in order to perform their write operations.

​
Configuration

In addition to the natural language Task instructions which inform the reasoning performed by the Respond step, you can also modify the Response style. Response style has three options: Factual, Balanced, and Creative.

​
Factual (default)

Minimizes “randomness” in LLM responses
  • Example use cases:
    • Information retrieval
    • Data analysis

​
Balanced

Provides accurate but flexible responses
  • Example use cases:
    • Report generation
    • Draft messaging with data

​
Creative

  • Provides diverse, expressive responses
  • Example use cases:
    • Brainstorming
    • Marketing content generation