After your agents complete simulations, review what they learned to verify accuracy and identify gaps.

​
What Results Include

Each simulation produces:
  • Exploration history — Pages visited, actions taken, screenshots captured, time spent
  • Knowledge graph — A structured map of what the agent learned about your application, including UI elements, workflows, feature relationships, and navigation patterns
  • Q&A log — Questions the agent asked and the responses you provided
  • Learning milestones — Key discoveries and insights

​
Reviewing Results

  1. Navigate to Simulations in your dashboard
  2. Click on a completed simulation to view details
  3. Review the activity log and screenshots

​
Key Things to Check

  • Completeness: Did the agent cover all intended areas?
  • Accuracy: Does the agent’s understanding match reality?
  • Depth: Is the knowledge specific enough to help real users?

​
Verifying Agent Learning

Test your agent in the Playground with questions related to what it explored:
  • “How do users create a new project?”
  • “Where can users find their order history?”
  • “What happens when someone tries to checkout without an account?”
Check that responses are specific (referencing actual UI elements), accurate (matching real workflows), and actionable (users can follow the guidance).

​
Improving Results

If you find gaps or inaccuracies, run follow-up simulations with more targeted instructions: For incomplete exploration: 
Focus specifically on the checkout process. Learn the complete
workflow including cart management, payment processing, order
confirmation, and error handling.
For error scenarios: 
Explore what happens when users encounter problems during
account creation. Learn about common error messages, recovery
procedures, and when to escalate to human support.

​
Continuous Improvement

  • Run simulations after product updates to keep agent knowledge current
  • Use common user questions to guide simulation focus
  • Explore areas where users report confusion
  • Verify existing knowledge remains accurate over time

​
Next Steps

  1. Test agent performance — Verify improved capabilities in the playground
  2. Create follow-up simulations — Address gaps or inaccuracies
  3. Update agent knowledge — Integrate new learning with existing knowledge
  4. Deploy to production — Use your improved agents to help real users