The Botsitting Burden: Why AI is Creating More Work and Higher Turnover

A new report finds that workers spend over six hours a week 'botsitting' AI by fixing its mistakes and providing necessary context. This hidden, unrewarded labor is creating a productivity paradox where individual tools fail to improve overall company performance. As a result, frustrated employees are significantly more likely to quit, prompting a need for better organizational standards regarding AI integration.

Key Points

• Workers spend nearly a full working day (6.4 hours) each week supervising, debugging, and cleaning up after AI tools.
• A productivity paradox exists where individual AI usage is high, but organizational performance gains remain low at only 13%.
• Botsitting is often tedious, unrewarded, and unmeasured, leading to a 73% increase in the likelihood of employees seeking new employment.
• AI is frequently automating the enjoyable parts of jobs, such as relationship building, while leaving humans with the exhausting task of bot supervision.
• Successful organizations focus on the 'work around AI' by providing context and defining what 'good' work looks like rather than just increasing deployment.

Sentiment

The overall sentiment is skeptical and concerned, and it generally agrees with the article's warning. The community sees botsitting as a real hidden cost that can degrade morale, trust, and quality, especially when adoption is mandated or measured only through output. The most positive comments treat AI as useful under skilled, selective control rather than as proof that broad organizational performance gains are automatic.

In Agreement

• AI often removes the enjoyable parts of work and leaves humans with review, correction, context feeding, and accountability.
• The burden is worse when coworkers or managers send unverified AI-generated work, because recipients must inspect everything and trust erodes.
• First-draft speed is not the same as finished value; architecture, testing, quality, rework, and business impact still determine whether productivity improved.
• Forced AI adoption can damage autonomy, motivation, pride in craft, and mental health, which can erase apparent gains.
• Management incentives favor visible output metrics and AI adoption narratives while ignoring the hidden work around AI.

Opposed

• Some workers report that agents make them substantially faster at concrete tasks such as scripting, queries, infrastructure work, boilerplate, and parallel project execution.
• Supervising an AI agent can be understood as normal delegation, similar to managing a junior employee and making decisions about tradeoffs.
• The current frustration may be a transitional cost while workers and toolmakers learn better workflows.
• Programmers and other knowledge workers may be losing a rare privilege of enjoyable work rather than experiencing something entirely new in labor history.
• Automation of service and administrative work may be necessary for broader productivity gains, even if it reduces fulfillment for some roles.