Project knowledge for AI agents: docs instead of memory retrieval
What happened
An article titled "AI agents need documentation, not memory plugins" reached the Hacker News front page on October 4. It argues that agents should keep project knowledge in structured documents, consulted before work and updated after, instead of retrieving past conversation snippets. The author says RAG-based memory plugins suffer from missing context, stale records and poor auditability, and describes Operator Memory, a plugin used for over a year that stores instructions, specs, research and an index as Markdown, with no vector database or background consolidation process. The piece presents the author's practice and claims without further validation.
Written by AI from the coverage · updated 2 hours ago
Coverage
Follow the reports to see the story from different sides.
- Hacker News front pageAgents don't need memory, they need documentation
AI 智能体的项目知识应通过结构化文档维护，文章主张用工作前查阅、工作后更新的流程替代对历史对话片段的检索。作者指出，基于 RAG 的记忆插件存在上下文缺失、记录过时和难以审计等问题，并介绍了自己使用一年多的 Operator Memory 插件，以 Markdown 保存指令、规格、研究和索引，无需向量数据库或后台整理进程。
Heat over time
Not enough continuous observations to draw a trend yet.