AI Agents can give customers wrong answers or mishandle requests due to misconfigured instructions — and these problems are often not found out until they already happened.

The problem: testing only works one scenario at a time, by hand, in the Test AI Agent panel. There's no way to check a new agent handles every scenario without typing each one out, and no easy way to know if an edit to an existing agent broke something that used to work. Manual testing doesn't scale, so agents launch, and get updated, without being fully tested.

Make it significantly easier to catch these problems before customers do — when setting up a new AI Agent or after editing an existing one. This should include:

Users have to go through a slow, manual process to think up and type out every test scenario before launching a new AI Agent or when checking an edit to an AI Agent's instructions.