会话追踪与成本控制助力诊断 AI Agent 故障
原标题：Session Traces and Cost Controls Help Diagnose AI Agent Failures
AI 摘要
StackGen 首席工程师 Sabith K Soopy 在 CNCF 成员文章中分享了生产环境中 AI Agent 故障诊断的实践经验，指出标准应用监控无法解释 Agent 为何循环调用、调用无效端点或虚报完成工作。文章建议使用 Langfuse 捕获嵌套会话追踪，结合硬性迭代上限、单工具调用限制和统计监控来控制成本，并将有界指标导出至 Prometheus。此外还介绍了 OpenTelemetry 生成式 AI 语义约定、LangSmith 和 Arize Phoenix 等互补工具，用于追踪数据的评估与回归测试。
正文节选
An agent can repeatedly call the wrong tool without triggering an availability alert. In a CNCF member post published on 4 August, StackGen principal engineer Sabith K Soopy described how session traces and cost controls help teams investigate repeated tool calls and unexpected spending. Drawing on months of operating agents in production, the post notes: "The hardest part isn’t building them; it’s understanding what they’re doing when they go wrong." Standard application monitoring indicates wh