📄 中文摘要
许多部署了大语言模型（LLM）驱动的代理的企业发现，随着 API 费用的上升，代理却依然频繁犯错，无法有效处理已知的边缘案例。代理的每次请求都是从零开始，导致在已解决的问题上不断消耗代币。问题的根源在于大多数代理框架将每次交互视为全新的开始，缺乏状态管理。通过使用开源工具如 OpenSpace，可以构建能够学习、降低成本并随着时间推移真正改进的代理。
为什么你的 AI 代理在烧钱，以及如何解决这个问题
出处: Why Your AI Agents Are Burning Cash and How to Fix It
发布: 2026年3月27日
📄 English Summary
Why Your AI Agents Are Burning Cash and How to Fix It
Many businesses deploying LLM-powered agents have noticed their API costs rising while the agents continue to make frequent mistakes and fail to handle known edge cases effectively. Each request from the agents starts from scratch, leading to continuous token consumption on problems they have already solved. The root cause lies in the fact that most agent frameworks treat every interaction as a blank slate, lacking state management. Utilizing open-source tools like OpenSpace can help build agents that learn, reduce costs, and genuinely improve over time.