“LangSmith provides an off-the-shelf evaluator that detects when an AI agent is stuck in a loop calling the same tools repeatedly, which indicates confusion or errors.”