AI 助手的上下文记忆缺口：用户被迫反复复制粘贴与重新介绍

来源证据

用户的上下文散落在十几个工具中，只能靠自己的记忆和耐心维系；每次与 AI 协作都要重新解释项目、粘贴对话、提醒它人物身份，最后一段仍要手工完成。

across a dozen tools, and the only thing holding it together is your own memory and your patience You re-explain the project, paste the thread, repetitively remind it who this person is, and then do the last mile by hand anyway. the bottleneck moved, and most products haven't I don't think I'm alone in reading it that way Garry Tan has stressed upon it this year telling founders that the leverage sits in your context rather than in the model same weights, same window, yet wildly different output