Artificial intelligence (AI) is delivering on its promise to save corporate workers hours of tedious labor, but a growing hidden tax is emerging: Employees are spending nearly as much time managing and correcting the tools as they are benefiting from them.
A new study published by the Work AI Institute reveals a stark paradox in the global AI rollout. While the technology saves digital workers an average of 11 hours per week, employees must reinvest more than six hours back into “botsitting,” the tedious task of verifying AI outputs, correcting errors, and rewriting prompts.
The research surveyed 6,000 digital workers across the United States, the United Kingdom, and Australia, combining survey responses with anonymized workplace data from the AI platform Glean.
The findings highlight a massive disconnect between individual efficiency and corporate profitability. While 75% of respondents reported a personal productivity boost, a meager 13% of organizations have seen significant business gains or revenue growth.
“Most people don’t realize the amount of time that they’re spending working on the tools to get the time savings that they’re professing,” said Paul Leonardi, co-author of the study and professor of technology management at the University of California at Santa Barbara.
The report exposes what it calls a “thick, mostly invisible layer of human labor holding the whole thing together.” According to data, for every hour a worker spends generating useful output from AI, they spend another hour making it usable. Of the total time dedicated to AI interactions each week, 37% are swallowed by botsitting, while only 36% goes to actual content production.
This friction is driven by frequent system failures. Workers reported that more than a third of all AI sessions fail completely, requiring a total restart or substantial rework.
This reliance on automated assistance is also eroding accountability. The survey found that 41% of workers admit to submitting AI-generated work they could not explain if pressed.
The report highlighted a critical breakdown involving a junior software engineer who integrated thousands of lines of AI-generated code before logging off. The code contained flaws that a senior engineer, already facing tight deadlines, had to meticulously untangle, while the junior employee was unable to explain the logic behind the automated output.
“Botsitting puts a number on verification debt. The labor of checking and correcting AI output is the cost that determines how much autonomy these tools actually earn, and most organizations have been treating it as invisible,” said Mitch Ashley, vice president and practice lead of Software Lifecycle Engineering and AI-Native Software Engineering at The Futurum Group. “Teams that leave correction as manual review keep paying a tax that eats the time savings they claim. The real exposure arrives when fatigue wins and people stop checking, because that is when unverified output ships into production.”
The issue is compounded by aggressive corporate mandates. Over the past six months, Silicon Valley firms have aggressively pushed employees to maximize AI usage. However, unbridled adoption has led to massive waste. Uber Technologies Inc., for instance, reportedly exhausted its entire annual AI budget within just four months without shipping a single usable feature.
Leonardi suggests the root problem is structural, treating entry-level staff like supervisors.
“We’re essentially expecting individual contributors to act as managers,” Leonardi said. “They’re managing these AI agents… but we’re not taking into account all of the work that actually goes into managing.”
