📄 中文摘要
监控智能体的行为与传统软件不同，输入是无限的，行为是非确定性的，质量体现在对话中。为了有效管理智能体的表现，需要关注监控的内容、如何扩展评估过程，以及如何利用生产追踪数据为持续改进奠定基础。通过建立有效的监控机制，可以更好地理解智能体的行为模式，从而优化其性能和用户体验。生产环境中的实时数据分析将成为提升智能体质量的重要手段。
在生产环境中，你无法预测你的智能体将会做什么
出处: You don’t know what your agent will do until it’s in production
发布: 2026年2月26日
📄 English Summary
You don’t know what your agent will do until it’s in production
Monitoring agents differs significantly from traditional software due to the infinite nature of inputs, non-deterministic behavior, and the quality residing in the conversations themselves. Effective management of agent performance requires a focus on what to monitor, how to scale evaluations, and how production traces can lay the groundwork for continuous improvement. Establishing robust monitoring mechanisms allows for a better understanding of agent behavior patterns, leading to optimized performance and enhanced user experience. Real-time data analysis in production environments will become a crucial tool for improving agent quality.