AI Assistant Systems, such as Gemini and Dola, have become integral to many applications due to their sophisticated task-handling capabilities. However, the resource demands of these systems are highly sensitive to user input, leading to unpredictable workloads that frequently cause latency spikes and Service Level Objective (SLO) violations. Traditional resource management methods, which primarily monitor request volume and infrastructure metrics, are inadequate for these environments because they overlook the critical impact of request \emph{content} on performance. This paper introduces AIMS, a content-aware resource management framework designed specifically for AI Assistant Systems. Unlike conventional approaches, AIMS analyzes request content to forecast the computational load, particularly the output token length of Large Language Model (LLM) inference, which is a key determinant of processing time. By predicting the resource consumption of incoming requests, AIMS enables dynamic and proactive autoscaling. Our analysis of a large-scale, production AI assistant system demonstrates that AIMS can significantly reduce SLO violations while improving overall resource utilization.