Skip to content

Editing Agents

Learn how to modify and improve your existing AI agents to better serve your customers and achieve your business goals.

  1. Select Target Agent

    • Go to the Agents page
    • Choose the agent you want to modify
  2. Open Playground

    • Click on the selected agent
    • You’ll enter the testing playground
  3. Access Prompt Editor

    • Click Prompt Editor in the top right corner
    • This opens the agent configuration interface

You currently have 2 ways to edit an agent:

  1. In the code style (Code Editor)
  2. In right hand side “what are we automating section” by typing in input and requesting your desired changes (AI-Assisted Editor)

This is by far the quickest and most effective way to edit a prompt. The code editor is essentially the same as the document view, but sectioned off meticulously in XML tags, this is how the agent can read the documentation effortlessly knowing exactly what is what.

Advantages

  • Fastest and most precise editing method
  • Direct access to XML-structured prompt
  • Complete control over agent behavior

How to Use

  • Navigate through XML tags to find specific sections
  • Use Cmd/Ctrl + F to search for specific elements
  • Edit content between XML tags directly

To edit this prompt, for example if we want to edit the objection handle for “what’s my rate”. Click in the code, hit Command + F, or Control + F (on PC) and search the objection handle response id WhatsTheRate. When located on the page, you can delete the text between the tags.

Find the objection handler:

<Response id="WhatsTheRate">
Great question, @firstName. Because of the amount of volume we do
with the banks we get our clients 2-3% lower than any other dealer
and as low as 5.99% on used! Now your rate reflects your current
credit as well as the year, make, model, and mileage of the vehicle.
We partner with 21 national lenders for exclusive rate breaks. I just
need to build your file and send it to the banks so we can get an
accurate number for you. That way, we are not misleading you in any way.
Sound fair?
</Response>

Replace with updated response:

I would simply delete the text inside the tags and replace it with what I want the agent to say. Then click Save changes.

<Response id="WhatsTheRate">
Your interest rate will be a reflection of your current credit standing
as well as the year, make, model, and mileage of the vehicle. We partner
with 17 national lenders for exclusive rate breaks. I just need to build
your file and send it to the banks so we can get an accurate number for
you. That way, we are not misleading you in any way. Does that work for you?
</Response>
  • Use Cmd/Ctrl + F to search for specific response IDs
  • Look for XML tags like <Response id="ObjectionType">
  • Search for keywords from the response you want to edit
  • Use descriptive response IDs to make finding content easier
  • Always backup your current agent before making major changes
  • Edit only the content between XML tags, not the tags themselves
  • Test changes in the playground before deploying
  • Keep response lengths similar to avoid conversation flow issues
  • <Response id="..."> - Objection handling responses
  • <FAQ id="..."> - Frequently asked question answers
  • <Context> - Background information and persona
  • <Flow id="..."> - Conversation flow instructions

This editor is for more advanced editors who want to use AI to create and modify prompts. Proper prompt engineering techniques should be applied if using this. We recommend starting with the code editor to learn the prompt structure and best practices, then using the sidebar for guided changes.

When to Use

  • Complex prompt modifications
  • Adding new conversation flows
  • Restructuring agent behavior

Requirements

  • Strong prompt engineering knowledge
  • Understanding of agent structure and best practices
  • Familiarity with existing agent patterns

Process

  • Use the “What are we automating” sidebar
  • Describe desired changes in natural language
  • Review and refine AI-generated modifications
  • Test thoroughly before deployment

The document view provides an easy-to-read view of the current prompt.

The document view provides a human-readable format of your current prompt configuration. Use this for:

  • Understanding current agent structure
  • Planning modifications
  • Training team members on agent capabilities
  • Documentation and compliance review
  1. Use Cmd/Ctrl + F to search for the objection ID
  2. Locate the <Response id="ObjectionType"> tag
  3. Replace the content between the opening and closing tags
  4. Save changes and test in playground
  1. Find existing FAQ section in the code
  2. Copy the structure of an existing FAQ
  3. Update the ID and content for your new question
  4. Save and test the new response
  1. Locate the <Context> or persona section
  2. Update personality descriptions and communication style
  3. Review all responses to ensure consistency with new personality
  4. Test extensively to verify personality changes
  1. Find <Flow> sections in the agent code
  2. Modify conversation progression logic
  3. Update transition statements between conversation stages
  4. Test different conversation paths thoroughly

After making any edits:

  1. Save Your Changes - Always save before testing
  2. Test in Playground - Try various customer scenarios
  3. Check Edge Cases - Test unusual requests and responses
  4. Verify Consistency - Ensure all responses match your intended personality
  5. Monitor Performance - Watch real conversations for any issues
  • Document what you’re changing and why
  • Test changes on a copy of the agent if possible
  • Save screenshots of current responses for reference
  • Plan rollback strategy if changes don’t work well
  • Monitor conversation performance closely
  • Gather feedback from team members using the agent
  • Track customer satisfaction and response rates
  • Be prepared to revert changes if needed

If you need to make significant changes to agent behavior, conversation flow, or add complex new features, consider getting help from:

  • Senior team members with prompt engineering experience
  • Technical support for system-level issues
  • Training resources for best practices

If your edited agent isn’t performing well:

  • Review changes systematically
  • Test in playground with various scenarios
  • Compare performance to previous version
  • Consider reverting to previous version while troubleshooting

Effective agent editing requires patience and systematic testing. Start small, test thoroughly, and gradually build your skills with more complex modifications. Next, learn about Troubleshooting common agent issues.