Testing Agents

The Playground is the fastest way to verify that your agents behave correctly before deploying them to production widgets. A structured testing approach helps you catch issues early.

Getting started

1. Navigate to the Playground page.
2. Select your agent from the dropdown.
3. Start chatting to test responses.

Testing strategies

Ask real customer questions

Type the questions your customers actually ask. If you have support tickets or chat logs, use those as inspiration. This validates that the agent gives helpful, accurate answers.

Test edge cases

Push the boundaries of your agent’s knowledge:

• Ask about topics the agent should decline to answer.
• Ask questions that are adjacent to your product but outside the agent’s scope.
• Try ambiguous or vague questions to see how the agent handles uncertainty.

Verify knowledge accuracy

When the agent references your documentation or product details, confirm the information is correct. Inaccurate responses erode user trust quickly.

Test all three modes

Each interaction mode serves a different purpose — test them individually:

• Tell — Does the agent give accurate text responses?
• Show — Does the agent navigate to the right pages and demonstrate the correct workflow?
• Do — Does the agent complete the action successfully?

Try different phrasings

Users ask the same question in many ways. Rephrase your test questions to make sure the agent handles variations:

• “How do I reset my password?”
• “I forgot my password”
• “Where is the password reset page?”

Iterating on your agent

After testing, refine your agent by:

• Updating instructions to correct behavioral issues.
• Adding or updating knowledge to fill gaps in the agent’s responses.
• Running additional simulations to broaden the agent’s understanding of your application.

Tip

Keep a list of test questions that represent common customer scenarios. Re-test after any instruction or knowledge changes to catch regressions.