Testing AI agents means moving beyond exact string checks into intent, context, RAG retrieval, API execution, and user-friction validation.