The Playground lets you chat with an agent, but there is no way to save a set of test questions with expected answers and re-run them after every change. AI Data Platform should add test sets: store questions, the tool calls expected, and key phrases the answer must contain, then run them in one click and show pass or fail before deploy or redeploy. While building the Shipment Routing Advisor today, I had to retype the same questions (S1001 classification, S1005 reroute) after every edit. A saved test set would make agent changes safe and repeatable.
