AI 的生产率悖论在于，例行任务的时间节省并不必然改善组织绩效。Ivan Zhao 指出，超过一半的 AI 推行问题源于数据和情境不足；与五年前相比，许多组织结构几乎未变。文章认为，真正的收益可能要等到企业从旧式“marching band”团队转向更灵活的“jazz band”结构，重新设计人类与 AI agents 的协作方式。

Glean 的 Work AI Institute 对 6,000 名数字工作者的调查显示，AI 每周为员工节省 11 小时，但只有 13% 的员工看到公司绩效改善。Rebecca Hinds 解释，部分收益被“botsitting”抵消：员工每周花 6.4 小时为 AI 提供情境、检查输出、重跑提示并清理错误。此外，近八成工作者每周切换多个 AI 工具，60% 会在不同工具中重复查询。

第三个问题是 AI 版“workplace theatre”：三分之一员工会淡化 AI 帮助，还要管理同事和老板对其工作的观感。短期内，组织需承认某些 AI 实验会失败；Nicolas Loeff 强调，技术边界不断变化。其他 workplace 数据包括 Natural History Museum 每年 710 万访客和 5.5 亿英镑扩建计划，以及 AXA 报告称 43% 的 18-24 岁人群可能有严重抑郁、焦虑或压力症状，六成曾向 AI 寻求心理健康帮助。

AI’s productivity paradox is that routine time savings do not automatically improve organizational performance. Ivan Zhao argues that more than half of AI rollout problems come from missing data and context; compared with five years ago, many organizational structures look almost unchanged. The article suggests that real gains may arrive only when companies move from old “marching band” teams to more fluid “jazz band” structures, redesigning collaboration between humans and AI agents.

A Work AI Institute survey by Glean of 6,000 digital workers found that AI saves employees 11 hours per week, yet only 13% see improved company performance. Rebecca Hinds explains that some gains are offset by “botsitting”: workers spend 6.4 hours weekly giving AI context, checking outputs, rerunning prompts, and cleaning mistakes. Nearly eight in 10 workers also switch among multiple AI tools weekly, while 60% repeat the same queries across tools.

The third problem is AI-style “workplace theatre”: one-third of workers downplay AI help while managing how colleagues and bosses perceive their work. In the short term, organizations must accept that some AI experiments will fail; Nicolas Loeff stresses that technology’s frontier keeps changing. Other workplace data include the Natural History Museum’s 7.1 million annual visitors and £550mn expansion plan, plus AXA’s finding that 43% of 18-24-year-olds may have severe depression, anxiety, or stress symptoms, and six in 10 people have asked AI for mental-health help.