一个能够预测下一个词的系统不仅能写出优质代码、提供战略建议，还能对人类问题给出令人惊讶的共情回应，但企业仍常把AI当成普通企业软件，硬塞进既有流程、KPI和IT管理。文中称这是一种深层战略失误，“去怪异化”使其独特力量被抹平，最终退化为下一轮办公室自动化浪潮。
“去怪异化”让企业沿用正常软件部署逻辑，比如要求每周90%的员工使用新工具，于是AI常被用于会议转录、产出大量额外备忘录或PPT等“工作垃圾”。当研究显示AI可提升30%生产率时，管理层容易直接算出应裁减30%人力，但更关键的是：当单个程序员可写出百倍代码时，组织如何重构，什么新产品与新市场会被打开。
更好的框架是“领导层—人群—实验室”三位一体：高层提出愿景并示范使用，让全体员工在安全边界内试错，再由技术与非技术人员组成的实验室把发现持续转化为新工作流与制度化能力。若激励不足，员工会理性隐匿AI使用（担心惩罚、担心收益被公司吞并，甚至悄悄减少90%工作量），管理层就看不见真实影响，最终只能依赖演示而非机制性学习，走向自动化与裁员而非重新定义组织可能性。
A system that can predict the next word can also write strong code, offer strategic advice, and respond with notable empathy, yet companies still treat AI as ordinary enterprise software by forcing it into existing workflows, assigning KPIs, and placing it under IT management. The text argues this is a major strategic mistake because de-weirding smooths away AI’s transformative edge and reduces it to another office automation wave.
That de-weirding mindset applies normal rollout rules, such as requiring 90% of employees to use a new package each week, so AI often becomes limited to compliance tasks like meeting transcription and huge volumes of low-value memos or slides. When studies claim 30% productivity gains, leadership may conclude 30% workforce reduction, but the harder question is what organizational redesign, new products, and new markets become possible when one programmer can now write a hundred times more code.
A stronger model is Leadership, Crowd, and Lab: executives set direction, empower employees to experiment, and sustain technical and non-technical teams that turn discoveries into new workflows and institutional knowledge. Without the right incentives and risk tolerance, employees hide AI use because of fear, uncertain sharing of gains, or quietly doing 90% less visible work, creating a visibility gap that drives firms toward automation and layoffs instead of exploring what becomes possible now but is hard to predict a year ahead.
Source: The IT department: Where AI goes to die
Subtitle: Ethan Mollick on why employers should treat AI as what it is: weird
Dateline: 4月 01, 2026 03:21 上午