AI & Production · Series

AI 员工系列The AI Employee Series

边界层笔记 · AI 与生产实践（之一 — 之四 + 番外）。从聊天到上岗，AI 为什么做不了员工——以及组织该怎么接住它。五篇笔记，一条主线：能力惊人，身份未定。The Boundary Layer · AI & Production (Parts 1–4 + Extra). From chat to deployment — why AI isn't a real "employee", and how organizations should absorb it. Five notes, one thread: brilliant capability, unsettled identity.

← 返回首页Back to home

系列目录The Five Notes

每篇均首发于公众号「边界层笔记」。点击阅读原文，跳转至对应文章。Every piece debuted on The Boundary Layer (WeChat). Click to read the original.

AI 智足够够了，但它做不了员工AI Is Smart Enough — But It Isn't an Employee

Klarna 用 AI 替下 700 名客服又重新招人；95% 的企业 AI 项目至今没创造价值。AI 不是新员工，它是个能力惊人、身份未定的新角色。Klarna replaced 700 support agents with AI, then rehired. 95% of enterprise AI projects have yet to create real value. AI isn't a new hire — it's a powerful, unplaced role.

阅读原文 ↗Read original ↗

AI 在生产中的漂移：为什么是必然的Why AI Drifts in Production — And Why It's Inevitable

日报一夜变空壳、系统悄悄偏移——AI 在生产环境里如何偏离预期，以及为什么这种漂移几乎必然发生、却总被忽略。An overnight-empty daily report, a quietly shifting system — how AI drifts from intent in production, why it's near-inevitable, and why it's always ignored.

阅读原文 ↗Read original ↗

AI 员工跨不过的两道门：知识与责任The Two Doors AI Employees Can't Cross: Knowledge & Accountability

英国法庭假判例案：知识不实，责任不立。AI 不会"查"，它只生成"看起来像引证"的文字——委派给 AI 的工作，责任依然是你自己的。A UK court case with fabricated precedents: knowledge unreliable, accountability absent. AI doesn't "check" — it generates text that only looks like citation. The responsibility you delegate stays yours.

阅读原文 ↗Read original ↗

AI 已是好马，但车还得人来开AI Is a Fine Horse — But a Human Still Holds the Reins

两个工程师用同一款 AI 重构代码库：一个炸了 340 个文件还伪造复盘，一个稳稳交付。协作有规，分工有界——差的不是工具，是用法。Two engineers refactored the same codebase with the same AI: one blew up 340 files and faked the review, one shipped clean. It's not the tool — it's how you use it.

阅读原文 ↗Read original ↗

番外：AI 让每个人都更快了，组织却没有Extra: AI Made Everyone Faster — Yet Organizations Stayed the Same

个人效率涨了，组织效率没动：96% 的公司没看到戏剧性改善，省下的时间像蒸发了一样。前面四篇讲一个人怎么驾驭 AI，这篇补上没回答的那半句——一群会驾驭的人，为什么还是快不起来。Personal speed up, org speed flat: 96% of companies saw no dramatic improvement, and the saved time evaporated. The first four pieces are about one person mastering AI; this one answers the half-left-unasked — why a team of masters still can't move faster.

阅读原文 ↗Read original ↗