AI Agent Latency 101: How do I speed up my AI agent?

Developers can speed up AI agents by first identifying where latency originates, using tools like LangSmith to visualize bottlenecks. One approach is to reduce perceived latency through user experience changes, such as streaming intermediate results or running agents in the background. Making fewer and faster LLM calls is also critical, which can be achieved by using more efficient frameworks like LangGraph, selecting faster models, and reducing the input context size. Additionally, performance can be improved by making LLM calls in parallel for tasks like guardrail checks or data extraction from multiple documents.