Tag
The paper characterizes the resource and performance dynamics of LLM-based AI agents across tasks like question answering and coding, revealing bottlenecks and proposing optimizations that improve latency by up to 5.4×.