Wayfound
  • Welcome to Wayfound
  • Key Concepts
  • Getting Started
  • Manager
    • Performance
    • Overview
    • Alignment
    • Meetings
    • Report Sharing
  • User Feedback
  • Agent Management Best Practices
  • Agents
    • The Agents Page
    • Definition
    • Guidelines
    • Connecting Agents
    • Connecting Actions
    • Connecting User Ratings
    • Connecting other Agent Events
  • Sessions
    • Recordings
    • Suggestions
    • Link Clicks
  • Visitors
  • Settings
    • Organizations
    • User Management
    • Integrations
      • HubSpot Integration
      • Salesforce Integration
    • Actions
  • API
Powered by GitBook
On this page
  1. Agents

Behavior

Behaviors let you control your agent's interactions with users with intuitive, natural-language prompts.

Last updated 2 months ago

Behaviors are only available for agents built in the Wayfound platform.

Wayfound offers behaviors as an easy and effective approach to prompt engineering. Behaviors allow you to break down long prompts into smaller components that you can test and iterate. By adding specialized behaviors your agent, you can ensure that it responds appropriately and delivers the best possible experience.

Agents come pre-populated with the following four behaviors:

  • You are a friendly AI Agent that helps users.

  • You should have a positive attitude and be helpful.

  • You should only use known information and not guess.

  • Only respond with information related to the Agent topic.

These behaviors are helpful for most applications, but they can be removed.

Updating and adding behaviors

You can update existing behaviors by clicking on them in the Behavior page.

To add more behaviors, click the + Add Behavior button below the list of current behaviors. You may add the following kinds of behaviors:

Visit Improving Agent Performance for tips on writing good behaviors.

Here are some more resources about prompt engineering:

If/then statement: Define specific actions for the agent to take in response to particular user inputs or situations.

Personality/tone guidance: Customize the agent's communication style, such as formal, casual, or empathetic.

Conversation flow: Guide how the agent conducts dialogues, including follow-up questions and topic transitions.

Call to action: Instruct the agent to encourage specific user actions, like contacting support or using certain features. Encourage the agent to provide the appropriate links for taking these actions.

Restrictions/limitations: Set boundaries on what the agent can say or do to ensure appropriate interactions.

Custom Command: Create unique behaviors tailored to your agent's specific purpose or domain.

As always, we encourage you to Test your agents when updating and adding behaviors. Wayfound makes this easy by offering a chat interface on the right-hand side of the screen. After you update or add behaviors, you can click reset at the top of the chat window to update the agent and test it. Testing your agents is a critical step to take before publishing them. Over time, we also recommend adding behaviors suggested by the Wayfound Suggestions tool.

OpenAI's

Anthropic's

Prompt Engineering Guide
Prompt Engineering Overview