Behavioral verification for AI agents
Blog
Test Outcomes, not Trajectories
An experiment with the τ²-bench airline suite shows how different tool calls can lead to the same correct outcome.
We use cookies to see how people use this site. Privacy policyDecline