Botsitting Costs Workers 6.4 Hours a Week, What to Do About the Job Nobody Assigned You

Digital workers spend 6.4 hours a week botsitting to make artificial intelligence usable, according to a report from Glean covered by HR Dive. That is close to a full working day every week spent checking outputs, debugging mistakes and fixing confident but wrong answers. The term comes out of that same research, and the head of the company's Work AI Institute later set out the wider argument in Harvard Business Review with a professor of technology management at the University of California, Santa Barbara, defining botsitting as the work employees do to make AI useful and describing it as a hidden cost sitting alongside AI's promised benefits. At Metaintro, we track roughly 50 million job postings , and almost none of them describe this work, which is precisely the problem.

What Exactly Counts as Botsitting?

It is everything you do around the tool rather than with it. The research describes it as everything required to make AI usable, including checking outputs, debugging mistakes and fixing confident but wrong answers. In practice it also covers re-explaining context the model has forgotten, reformatting output that arrived in the wrong shape, and verifying claims a tool produced fluently and incorrectly.

The published summary of that Harvard Business Review piece opens with a consultant who advises government agencies on technology partnerships and who signed up for an AI scheduling agent expecting it to buy back his time. He pictured a clutter free calendar and an end to scheduling back and forth. What he got instead was a second job, managing the agent he had hired to manage his calendar.

That phrase, a second job, is the accurate description of the problem. The work is real, it is skilled, and it appears on nobody's job description. The authors are explicit that leaders need to account for this hidden labour, and the reason they have to say so is that most organisations currently do not.

How Much Time Are We Actually Talking About?

Nearly a full working day a week. The figure of 6.4 hours comes from a survey of 6,000 full time digital workers, split across 3,000 in the United States, 1,500 in the United Kingdom and 1,500 in Australia, conducted in December and January.

Put that in career terms rather than productivity terms. If you spend 6.4 hours a week on work that never appears in a performance review, then roughly sixteen percent of your working time is invisible to the people deciding your next raise. That is the single most important consequence of the finding, and it is fixable by you rather than only by your employer.

It also reframes the usual argument about whether AI saves time. The tools may well save more than they cost. The question that matters for an individual is whether the saving shows up as capacity you get credit for, or as a quieter, longer day. Our coverage of the coming burnout from managing AI agents and of the hidden cost of the AI workday tracks the same tension from different angles.

Why Does Botsitting Push People Toward the Exit?

Because unrecognised work is corrosive in a way that hard work is not. The research found that frequent botsitters, defined as those spending 40 percent or more of their time on botsitting activities, were 73 percent more likely than their peers to be actively hunting for another job.

That is a retention finding rather than a productivity one, and it should change how the problem is discussed internally. A team quietly absorbing hours of unaccounted oversight is not a team with an efficiency issue. It is a team with a flight risk, and the cost of replacing a skilled worker dwarfs the cost of adjusting how AI work is assigned and measured.

The second order effect is worse. The research describes workers who become exhausted by oversight and start cognitively offloading, writing that as the toggle tax increases workers "hand more of their thinking and judgment over to the machine. They start to cut corners. They stop checking outputs, verifying sources, and asking whether the AI's recommendations make any sense". The authors describe it as a slow surrender of agency rather than a single bad decision, in which workers first stop fully understanding the output, then stop interrogating it, and eventually stop feeling responsible for it at all.

What Happens When People Stop Checking?

They ship work they cannot defend, and the numbers are large. Nearly 7 in 10 AI users admitted to shipping AI generated work they had not reviewed, did not fully understand, or could not defend if asked. Heavy AI users, Generation Z workers, men and managers were the groups most likely to do it.

A separate survey found 29 percent of workers said they had turned in work they could not fully explain if asked how they did it. Blame follows the same curve. Twenty eight percent of AI users have blamed AI for bad outputs, rising to 41 percent among heavy users.

The career risk here is concentrated and personal. Being unable to explain your own deliverable is the fastest way to lose professional credibility, and it tends to surface at the worst possible moment, in front of a client or a senior stakeholder. Our reporting on teams getting less creative when they lean on AI and on what happens when AI use erodes ownership of work covers where that leads.

Is Oversight Actually a Skill Worth Having?

The evidence says it is the skill. A joint field study from the McCombs School of Business at the University of Texas at Austin and KPMG examined more than 523 early career professionals asked to use AI agents to complete workplace tasks. The finding is the most useful thing in this whole subject area.

Workers who excelled treated AI as a collaborator that needed direction, oversight and judgment, rather than simply delegating tasks to it. Those who fell short offered critiques that did not improve on what the AI produced alone, often chasing irrelevant issues or steering it the wrong way. Critically, the employees who fell short scored just as high on critical thinking, domain knowledge and AI literacy as the amplifiers who pulled ahead. The researchers concluded that the primary factor separating them was how they worked with AI, not what they knew.

One of the study's authors, a professor at the university, framed the aim as understanding what enables some individuals to consistently create value beyond what AI can produce on its own. A KPMG leader added that this is the most AI native generation entering the workforce, so if fluency with the tools is not what sets top performers apart, that says something about the entire workforce, and that the gap is coachable. Coachable is the key word, because it means this is available to you regardless of your starting point. Our guide to how workers who direct AI outperform those who delegate to it covers what that looks like in practice.

How Do You Make Botsitting Visible at Work?

Document the decisions rather than the hours. The research recommendation is specific and unusually actionable. Organisations should make this kind of judgement visible by having people explain why they made certain decisions and document why they accepted, changed or rejected AI's output, which turns an invisible skill into a coachable one and grades the process rather than the deliverable.

You do not need permission to start doing that for yourself. Keep a short running log of the AI assisted work you produce, noting what the tool got wrong, what you caught and what the consequence would have been if you had not. Three months of that log is the strongest possible evidence in a performance review, and it converts 6.4 hours a week of invisible labour into a documented contribution.

Then raise it as a workload question rather than a complaint. The framing that works is capacity based. If oversight of a given tool takes a predictable number of hours each week, that time has to come from somewhere, and naming the trade off is a manager's job to resolve rather than yours to absorb. Our guides to how to keep your skills ahead of AI at work and to workers who use AI earning more cover the upside of getting this right.

What Should Employers Do Differently?

Stop measuring adoption and start measuring judgement. The research is blunt that quantity heavy strategies, described as maximising AI usage for its own sake, are the wrong target, and that successful companies focus instead on setting context, defining what good looks like, and building judgement and other parameters around AI use itself.

There is a sector specific warning worth noting too. Human resources teams show higher than average AI adoption, with 90 percent of surveyed HR professionals reporting they used the technology, concentrated in lower stakes applications like content writing, coordination and administration. High adoption in a function that also sets policy for everyone else is a reason for that function to model good practice rather than only volume.

For anyone assessing an employer, this gives you a strong interview question. Ask how the organisation accounts for the time spent reviewing AI output, and whether anyone has measured it. An employer with an answer has thought about the actual workflow. An employer without one is likely to be counting the savings and not the costs, which means the difference lands on your calendar. Our reporting on workers and AI guardrails covers the governance side of the same question.

How Do You Actually Reduce the Hours?

Attack the causes in order of size. The largest single source of botsitting for most people is context, meaning the background a tool needs before it can produce anything usable. Re-pasting the same project details into every conversation is pure waste, and the fix is to build reusable context once, whether that is a saved brief, a standard prompt or a shared document the tool can draw on. This is unglamorous and it removes more hours than any change in tooling.

The second largest is the verification loop on tasks where errors are expensive. The honest answer here is that some tasks should not be delegated at all. Anything where you would have to check every line to be confident is a task where the checking costs more than the drafting, and doing it yourself is faster. Deciding that in advance, rather than after two rounds of correction, is what separates efficient use from expensive use.

Third is scope. Broad, vague requests generate broad, vague output that then needs shaping, which is the same failure that makes badly written work requests run late. Narrowing a request to one deliverable with a stated format and a stated audience cuts the rework substantially, and it is the same discipline that good delegation to humans requires.

Fourth is the honest audit. Keep a note for two weeks of every task where the AI assisted route took longer than doing it yourself. Most people find a clear pattern within a handful of examples, usually clustered in tasks that need judgement about a specific context the tool cannot see. Dropping those tasks from your AI workflow is the single fastest reduction available, and it costs nothing but attention.

What Does This Mean for Your Career Right Now?

Four moves. First, measure your own botsitting for two weeks. You cannot argue about a number you do not have, and the benchmark of 6.4 hours a week gives you something to compare against. Most people are surprised by the total.

Second, never ship what you cannot explain. Nearly seven in ten AI users admit to doing it, which makes the discipline of not doing it a genuine differentiator rather than a basic standard. If you would not be comfortable being asked how you reached a conclusion, the work is not finished.

Third, reframe oversight as a competency and put it on your resume in those terms. Directing, correcting and validating AI output is exactly what separated the strong performers in the field study from peers with identical knowledge and skills. Describe the workflow you own, the errors you catch and the standard you hold, because that is a hiring signal and almost nobody is writing it down.

Fourth, watch the retention signal in yourself. Frequent botsitters were 73 percent more likely to be job hunting, and if that describes you, the honest question is whether the problem is the tool or an employer who has quietly moved a day of unassigned work onto your desk each week. Both are fixable. Only one of them is fixable by staying.

Related Articles

• How People Really Use AI at Work
• The AI Skills Employers Actually Want
• Prompt Writing for Non Engineers
• When You Should Not Use AI at Work
• Using AI at Work Ethically
• Tasks Solopreneurs Should Never Hand to AI
• The AI Skills Managers Are Expected to Have
• How to Upskill for AI Without Going Back to School
• Workers Pushing Back Against AI at Work
• Shadow AI and What It Costs Employers

People Also Asked

Q: What does botsitting mean?

A: Botsitting is the work employees do to make AI useful. The term comes from research by an AI company surveying digital workers, and was set out in Harvard Business Review by the head of that company's Work AI Institute and a professor of technology management. In practice it covers checking outputs, debugging mistakes and fixing confident but wrong answers. Survey research puts it at 6.4 hours a week for digital workers, close to a full working day.

Q: How many workers ship AI output they have not checked?

A: Nearly 7 in 10 AI users admitted to shipping AI generated work they had not reviewed, did not fully understand, or could not defend if asked. Heavy users, Generation Z workers, men and managers were most likely to do it. A separate survey found 29 percent of workers had turned in work they could not fully explain.

Q: Does spending time on AI oversight hurt your career?

A: Only if it stays invisible. Frequent botsitters were 73 percent more likely to be actively job hunting, which reflects unrecognised workload rather than the skill itself. Field research found the workers who add the most value are those who direct and correct AI rather than delegate to it, and that this gap is coachable rather than a matter of prior knowledge.

Future proof your career by making the judgement work visible rather than absorbing it quietly. Metaintro tracks roughly 50 million job postings so you can find employers who are hiring for AI oversight and direction rather than just AI usage, and you can create a free account to get matched with them.