📄 中文摘要
在与 AI 代理工作时，角色的重要性不可忽视，独特的视角能够塑造输出，超越单纯的上下文。然而，每次新的上下文窗口都从零开始，代理需要重新理解代码库，这不仅是个不便，更是结构性问题。每个会话的空白起点导致代理重复发现已知信息，扫描已读文件，推断已告知的约定。随着代码库复杂性的增加，这一问题愈发严重，无法扩展。因此，如何为代理提供持久的理解成为关键。
代理记忆只是一个你忘记索引的数据库
出处: Agent Memory Is Just a Database You Forgot to Index
发布: 2026年3月24日
📄 English Summary
Agent Memory Is Just a Database You Forgot to Index
The importance of personas in working with AI agents is highlighted, as distinct perspectives shape outputs in ways that raw context cannot replicate. However, a significant limitation arises: each new context window starts from scratch, requiring the agent to rebuild its understanding of the codebase every time. This is not merely an inconvenience but a structural issue. Beginning each session with a blank slate leads to agents rediscovering known information, scanning previously read files, and inferring conventions already communicated. As the complexity of the codebase increases, this problem worsens and does not scale. Thus, providing agents with persistent understanding becomes crucial.