Platform webhooks typically require a confirmation response within 3 seconds, while LLM inference usually takes several seconds or longer, making it necessary to introduce a message queue for asynchronous processing.