The Hidden AI Tax: Enterprise Workers Are Spending 6.4 Hours a Week Cleaning Up After AI Tools

AI is saving workers time — but then demanding much of it back. A new survey of 6,000 full-time digital workers by Glean's Work AI Institute, co-authored by researchers from Stanford, UC Berkeley, Emory University, and several other institutions, quantifies a productivity paradox quietly eating into enterprise AI ROI.

What the Research Found

The survey found that 87% of digital workers are using AI, and the technology is automating more than a quarter of their workload — saving approximately 11 hours per week per worker. But employees lose roughly 6.4 hours per week (one-third of their work week) to what researchers term "botsitting": the unrecognized labor required to make AI actually function.

Botsitting includes:

• Feeding AI context that it doesn't already have about the organization
• Supervising outputs and checking for errors
• Debugging failures across multiple disconnected tools
• Repeating prompts across different AI systems that don't share information

Two New Failure Modes

The study names a related behavior: "botshitting" — submitting AI-generated work without fully verifying or understanding it. 69% of users admit to doing this, and 41% have delivered work they could not explain if asked. Another 28% have blamed AI for mistakes they themselves caused.

Workers using multiple AI agents are significantly more likely to botshit, because agents can spiral through many steps before an error surfaces — by which point cleanup is expensive.

What High-Performing Organizations Do Differently

The report finds that the companies pulling ahead are not using AI more intensively — they're spending more time on the work around AI:

• Setting context before models run
• Defining what "good" looks like in advance
• Building judgment about when not to use AI at all
• Continuously revisiting governance and policy

Only 13% of respondents say AI has significantly improved their company's performance — a signal that volume of AI usage is not the right metric. Quality of framing, governance, and skill development are.

Read the full article on CIO