

How issues work
After Bluejay evaluates a simulation conversation, it finds where the agent went wrong and matches each failure to similar failures from your other simulation conversations. Every simulation issue is an Agent Failure: the agent had a failure based on its conversations (transcripts). Tool and phone provider errors become issues only in production.Work through an issue


1
Open the top issue
Start with the issue that hits the most conversations, and read a few of its conversations to see what the agent did.
2
Find the cause
Click Investigate to ask Bluejay AI what to change, or Fix to copy a prompt for your coding assistant.
3
Track the work
After your fix, add a custom metric that checks for the failure, and rerun the simulation.