Workers Lose Nearly Half Their AI Time to Cleaning Up Its Mistakes

Workers now spend about ninety minutes a day talking to or using artificial intelligence, roughly forty seven days a year, and nearly half of that time goes on troubleshooting errors or rewriting prompts. HR Dive reported the finding from a BambooHR survey of more than 1,600 full time salaried US workers. The rework alone is equivalent to twenty working days a year. At Metaintro, we watch this closely because the gap between how much time AI saves and how much it costs is becoming a real career skill rather than a technology question. Here is what the data shows.

What Did the BambooHR Survey Actually Find?

Respondents reported spending an average of an hour and a half each day using or talking to AI tools, which BambooHR calculated as about forty seven days a year. Nearly half of that time, equivalent to roughly twenty days annually, was spent on troubleshooting AI errors or on prompt iteration rather than on the work itself. Attitudes were nonetheless positive. Sixty five percent of employees said they felt confident and enthusiastic about using the technology at work, and 58 percent said they were motivated by the time savings they gained from it. Those two findings sit together more comfortably than they first appear. A tool can save meaningful time overall while still wasting a large share of the time spent operating it, in the same way that a car can be worth owning despite the hours lost to traffic. The useful question is not whether AI helps but whether the twenty days of rework is shrinking as people get better at it, or whether it is a permanent tax. Nobody has a reliable answer yet, because the tools change faster than anyone can run a clean comparison. What is measurable is that the rework exists at scale and that almost no organisation accounts for it when calculating what the technology saves.

Why Do Executives Spend Twice As Long With AI?

The survey found that people at vice president and C-suite level spent nearly twice as much time with AI as individual contributors did. That is a striking inversion of how workplace technology normally spreads, which is usually from the people doing the operational work upward.

There is a plausible explanation and an uncomfortable implication. The explanation is that senior roles involve more synthesis, drafting and summarising, which is precisely what current tools do well, while junior roles often involve execution inside specific systems where the tools help less. The implication is that the people setting AI policy are drawing on a materially different experience of the technology than the people who have to live under that policy. If your executive team finds AI transformative and your delivery teams find it a source of rework, both groups are describing their own reality accurately. That gap is worth naming explicitly in any internal conversation about adoption, and it explains a good deal of the friction described in why Gen Z workers are pushing back on AI at work.

What Is Workslop and Why Does It Spread?

Faulty or low quality AI output has acquired its own name, workslop, and the BambooHR figures suggest it is common enough to warrant one. The dynamic that makes it spread is straightforward. Producing plausible output is now nearly free, while checking it remains expensive and manual. Any system where the cost of creating something falls far below the cost of verifying it will accumulate unverified material. That has a direct consequence for how work gets reviewed. When a colleague sends you a document, the traditional assumption is that its existence implies effort and therefore some baseline of care. That assumption no longer holds reliably, and the reviewing burden shifts to the recipient. Teams that have not adjusted their review norms end up with the twenty days of rework distributed unevenly, usually landing on whoever is most conscientious. If you are that person, the correct response is to make the pattern visible rather than to absorb it quietly, because absorbed costs never get fixed. Raising it as a process question rather than a complaint about a colleague tends to work better, since the underlying issue is genuinely structural rather than personal. The judgement about when to reach for a tool at all is the first line of defence, and when you should not use AI at work covers where that line usually sits.

Is AI Replacing the Conversations People Used to Have?

This is the finding with the longest tail. Respondents spent an average of 41 percent of their working time with AI against 59 percent talking to people. More than a third said that knowledge transfer within their organisation now happens primarily through AI tools not through colleagues. BambooHR also found that 27 percent of respondents would rather ask AI than admit to a colleague that they needed help. Research from Workday published in May pointed the same way, finding that 33 percent of employees rarely or never spoke to co-workers during a given week beyond task related conversation, while 76 percent had used AI tools for advice and 52 percent for brainstorming. Respondents described AI as less judgemental and more available than human colleagues, which is an honest account of why the substitution happens. The career risk here is specific and easy to miss. Mentorship, sponsorship and the informal reputation that gets you considered for the next role are all built through exactly the conversations being displaced. A tool that answers your question without judgement also leaves no one who knows you asked it, and no one who remembers that you learned quickly.

How Much of This Is an Organisational Failure Rather Than a User Failure?

A considerable amount. Talent assessment firm Talogy reported in August that the technology has advanced at a pace that outstrips workers' capacity to judge its long term value, which is a polite way of saying that most organisations are deploying tools faster than they are building the judgement to use them. The result lands on individuals as rework and shows up in surveys as user frustration.

The picture inside HR functions illustrates it neatly. Research from Paylocity found that four in five HR professionals were actively managing at least one issue with their AI tools, while 43 percent said AI saved their recruiting teams the equivalent of a full working day each week. Both statements are true simultaneously. Real savings coexist with a persistent maintenance burden that nobody budgeted for and few people have in their job description. If you find yourself doing that maintenance, it is worth documenting the hours, because unnamed work does not appear in performance reviews and does not become anyone's priority to fix. Employers who ask staff to convert AI gains into more output not more capacity are making a choice, and Meta staff being told the productivity gains belonged to the company is a public example of it.

Does Any of This Mean AI Is Bad for Your Career?

No, and reading it that way would be a mistake. The data describes a technology in an awkward adolescence, not one that has failed. Sixty five percent of workers being confident and enthusiastic is a high number for any workplace tool, and the reported time savings are real even after the rework is deducted. What the data does suggest is which skills are becoming valuable. The twenty days a year lost to troubleshooting represents a gap between what the tools produce and what the work requires, and closing that gap is itself a capability. People who can specify a task precisely, spot a wrong answer quickly and know when not to use a tool at all are more valuable than people who simply use it often. That is a learnable skill, not a personality trait, and prompt writing for people who are not engineers is a reasonable place to start. Keeping the underlying domain knowledge sharp matters just as much, since you cannot catch an error in a field you no longer understand, and how to keep your skills ahead of AI at work covers that side of it.

Does the Wider Labour Market Back This Up?

It does, in a way that should reassure anyone worried the technology is about to remove their job outright. The Bureau of Labor Statistics employment projections published in August expect occupations most exposed to AI to grow just under 2 percent through 2035, slower than the 3.5 percent all occupation average but still growing instead of collapsing. The same projections show computer and mathematical occupations among the five fastest growing major groups at 7.3 percent. Set against the BambooHR findings, that produces a coherent picture. AI is not eliminating whole occupations at speed. It is redistributing work inside them, adding a maintenance and verification layer that did not previously exist while removing some of the drafting and searching that used to fill a day. The Occupational Outlook Handbook is the most detailed public source on how individual occupations are expected to change, and it is more useful for planning than any single survey. What the survey adds is the texture the projections cannot capture. A role can be projected to grow while the daily experience of doing it becomes measurably more frustrating, and twenty days a year of rework is the kind of frustration that drives people out of jobs that statistics say are perfectly healthy.

What Should You Actually Change on Monday?

Start by measuring your own number. For one week, note roughly how long you spend using AI tools and how much of that goes on fixing output. Most people are surprised in one direction or the other, and either result is useful. If your rework share is well above half, the problem is usually task selection not skill, which means you are reaching for the tool in situations where it was never going to help. Then protect the human channel deliberately. If a third of workers now get their organisational knowledge through tools not colleagues, the people who keep asking colleagues will accumulate relationships that the rest do not have, and in a slow hiring market those relationships are what produce opportunities. Ask one substantive question of a person each week that you could have asked a tool. It costs almost nothing and it is the only part of this that compounds. Make a habit of it with people outside your immediate team as well, because the colleagues who can vouch for you when a role opens are rarely the ones who already see your work every day. Finally, be careful about which tools you use for company work, since the boundary between convenience and a policy breach is thin, and the hidden risks of bring your own AI at work sets out where organisations tend to draw it.

What Does This Change About How You Present Yourself to Employers?

Employers have started asking about AI in interviews, and the weak answer is a list of tools. The strong answer describes judgement. Being able to say which tasks you hand to a model, which you keep, and how you verify what comes back demonstrates exactly the capability the survey shows most organisations are missing, and it distinguishes you from candidates who can only report enthusiasm.

It also pays to be honest about the failure modes. A candidate who can describe a specific occasion when a tool produced confident nonsense and explain how they caught it sounds like someone who has actually done the work, which is more persuasive than fluency. Interviewers hear generic enthusiasm constantly and have no way to test it. A concrete failure and a concrete recovery is testable, and it implicitly demonstrates that you understand the domain well enough to know when the output was wrong, which is the underlying capability being assessed. Research on how people really use AI at work suggests the gap between stated and actual usage is wide, so specificity reads as credibility. At Metaintro, we see AI capability written into more job descriptions every month, and the roles that pay for it are asking for discernment not familiarity.

Related Articles

• Most Workers Are Secretly Using AI on the Job and It Is Costing Employers
• Free and Low-Cost Ways to Learn AI Skills That Employers Trust
• AI Skills for Non-Technical Roles in Marketing, Sales, and Operations
• AI Skills for Managers Leading AI-Augmented Teams
• How to Show AI Skills on Your Resume and in Interviews
• The Reskilling Fix for the AI Skills Apocalypse Looks Like a Flight Simulator
• Output Used to Prove You Cared and AI Just Made It Meaningless
• Hidden Resume Prompts Are the Newest Way to Game AI Screeners and the Riskiest
• How to Use AI at Work Without Crossing Ethical Lines

People Also Asked

Q: How much time do workers spend using AI each day?

A: About an hour and a half, according to a BambooHR survey of more than 1,600 full time salaried US workers. That works out to roughly forty seven days a year. Nearly half of that time, the equivalent of about twenty working days annually, is spent troubleshooting errors or iterating on prompts rather than on the underlying work.

Q: What is workslop?

A: Workslop is the informal term for low quality or faulty output generated by AI tools that then has to be corrected by a person. It spreads because producing plausible material has become nearly free while verifying it remains manual and slow, so the checking burden shifts onto whoever receives the work rather than whoever produced it.

Q: Is AI reducing how much people talk to each other at work?

A: The evidence suggests it is. Workers reported spending 41 percent of their time with AI against 59 percent with people, and more than a third said organisational knowledge transfer now happens primarily through AI tools. Separate Workday research found 33 percent of employees rarely or never spoke to colleagues beyond task related conversation in a given week.

Build the judgement employers are actually paying for. Metaintro tracks roles across every sector where AI capability is becoming part of the job description, and matches them to what you already do in a single chat based thread. Create your free profile and see which roles value the skills you have been building by accident.