Microsoft 工程师 Victor Dibia 描述，使用 Anthropic 的 Claude Code AI agent 来写程式和修正程式，并没有让他感到更自由或更不忙。这个 agent 并没有像资浅同事一样分担工作，反而让他一直盯著萤幕，像是在「照顾」它的工作，并担心自己不能放心把视线移开。他的经验反映出 AI agents 更广泛的一项承诺：它们可以处理电子邮件、排程和笔记摘要等行政工作，但也会改变监督它们的人类的工作性质。

文章说，这种模式已经可从使用数据和产品动能中看出来。OpenClaw 在 1 月的一篇部落格文章报告，一款能代表使用者向同事和朋友发讯息的 AI 助手，在单一周内有 2 million 次访问，而 Anthropic 预期其更广泛的 Claude Cowork agent 会超过受欢迎的 Claude Code 产品。研究人员把这种现象称为「AI brain fry」，并认为 AI 工作中最耗脑力的部分不是使用工具，而是监督工具，这会让劳工感觉自己更像分心的经理，而不是高效率的操作员。

BCG 在 1 月针对大型公司中的 1,488 名美国全职员工进行调查，发现当 AI 取代繁琐工作时，确实可以减轻倦怠，但如果劳工还必须管理这些工具并承担更多总工作量，这些好处就会被抵销。结果是，许多使用者的心理疲劳更高，尽管各公司仍因 AI 强大的生产力吸引力而积极采用。文章的核心警告是，AI agents 也许能提高产出，但若没有新的界线，它们也可能加剧压力，并为监督它们的人带来一种新的过劳形式。

Microsoft engineer Victor Dibia described how using Anthropic’s Claude Code AI agent to write and fix code did not make him feel freer or less busy. Instead of acting like a junior delegate, the agent kept him glued to his screen, “babysitting” its work and worrying he could not safely look away. His experience reflects a broader promise of AI agents: they can handle administrative tasks such as emails, scheduling, and note summaries, but they also change the nature of work for the humans supervising them.

The article says this pattern is already visible in usage data and product momentum. An OpenClaw blog post in January reported 2 million visits in a single week for an AI assistant that can message colleagues and friends on a user’s behalf, and Anthropic expects its broader Claude Cowork agent to outgrow the popular Claude Code product. Researchers are calling the phenomenon “AI brain fry,” arguing that the most mentally taxing part of AI work is not using the tools but overseeing them, which can make workers feel more like distracted managers than efficient operators.

BCG surveyed 1,488 full-time US workers at large companies in January and found that while AI can reduce burnout when it replaces tedious work, the gains are offset when workers must also manage the tools and take on more total work. The result was higher mental fatigue for many users, even as companies remain eager to adopt AI because of its strong productivity appeal. The article’s central warning is that AI agents may boost output, but without new boundaries they can also intensify stress and create a new form of overwork for the people monitoring them.