Office workers have embraced AI tools at a startling clip. Yet a major new survey reveals they now devote nearly as much effort to coaxing those systems as to the output they generate. The hidden drag, dubbed botsitting, eats up more than a full workday each month for the average knowledge worker.
The Reality Behind the Productivity Claims
Employees report saving roughly 11 hours a week through automation. But those gains evaporate when they feed context into models, debug hallucinations, rerun failed prompts and polish sloppy results. Glean’s Work AI Index 2026 lays it out plainly. Workers spend 6.4 hours weekly on botsitting. That exceeds the time they spend coaxing useful work from the tools.
Of all hours logged interacting with AI, 37% go to supervision and repair. Only 36% produce finished output. The rest splits between learning the systems and building agents. And failure rates remain stubborn. More than one-third of AI sessions collapse outright. Workers restart or rework them from scratch. The Los Angeles Times captured the frustration in fresh coverage this week. Paul Leonardi, who co-authored the report, called the numbers striking. “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.”
Short. Blunt. Expensive.
The survey questioned 6,000 full-time digital workers across the United States, United Kingdom and Australia between December 2025 and January 2026. Researchers from Notre Dame, Stanford and UC Berkeley contributed. Eighty-seven percent already use AI at work. Seventy-five percent say it makes them personally more productive. Yet just 13% report their organizations perform significantly better because of it. The gap yawns wide.
Rebecca Hinds, head of Glean’s Work AI Institute, described botsitting as tedious, exhausting labor. It goes untracked, unrewarded and unbudgeted. “Workers now burn an average of 6.4 hours a week botsitting — most of a full working day, every week,” the report states. The Business Insider highlighted how this load pushes some toward the exit. Heavy botsitters prove 73% more likely to hunt for new jobs.
But that’s only half the picture. The report also tracks botshitting — the habit of shipping AI-generated material that employees haven’t fully verified, can’t completely explain or wouldn’t confidently defend. Sixty-nine percent admit to the practice. In organizations measuring only speed, that figure climbs to 74%. When leaders track quality too, it drops and self-reported work quality rises sharply. Eighty-three percent in quality-focused firms say AI improves output. Just 68% say the same where metrics fixate on volume alone.
And the fallout compounds. Layoffs tied to AI correlate with higher botsitting and botshitting rates. Workers left behind shoulder extra loads. Sixty-two percent in those settings actively job hunt. The cycle feeds on itself. CIO called it a vicious cycle that organizations ignore at their peril.
Executives push aggressive adoption. Many frame any hesitation as resistance. Yet the data shows reflexive AI use has become table stakes. Employees juggle multiple tools, switch contexts constantly and shoulder managerial duties their job descriptions never mentioned. One junior engineer in the report generated broken code with AI. A senior colleague stepped in to repair it. The junior then had to explain work he never truly produced. The invisible layer of human effort holds the system together. Remove it and confidence collapses.
Transformative companies approach the problem differently. They treat AI as an invitation to redesign jobs, not just accelerate old ones. They invest in context-rich systems that reduce prompting overhead. They measure outcomes beyond tokens generated or tasks completed. They train people when to ignore the model entirely. Those firms report stronger quality gains and lower frustration. The rest burn hours on maintenance that never appears in productivity dashboards.
Recent coverage reinforces the pattern. A Register article on UK-specific data found British workers average 5.8 hours weekly on similar tasks, with 90% required to use AI and many handling four or more tools. Brits appear to shoulder an even larger share of oversight relative to output. Discussions on X this week echo the fatigue. Engineers joke about trading cat-sitting rates for bot-sitting gigs. Others note the emotional tax of coaxing unreliable systems feels heavier than the marketing ever suggested.
So what now? The report urges leaders to build human infrastructure alongside the technical stack. That means clear policies on acceptable use, ongoing training, feedback loops that treat employee experience as data, and incentives tied to quality rather than raw volume. Without those changes the promised productivity wave stays trapped in endless correction loops. Workers grow resentful. Talent walks. Organizations celebrate adoption metrics while actual results stagnate.
The numbers don’t lie. AI automates more than a quarter of digital tasks today and could reach 35% within a year. Yet the tax on human attention remains steep. Companies that solve for botsitting will capture the real gains. Those that don’t will simply replace one form of busywork with another. The choice sits with leadership. The clock, however, keeps ticking in six-minute increments of debugging and context-loading. One prompt at a time.
WebProNews is an iEntry Publication