When AI Creates More Work: Inside the Hidden Cost of Botsitting

The 11-Hour Promise vs. the 6-Hour Tax

A large-scale survey from the Work AI Institute — drawing on 6,000 digital workers across the US, UK, and Australia — found that AI saves workers roughly 11 hours per week. That’s a genuinely impressive number.

But buried in the same data: workers spend over six of those hours checking outputs, fixing errors, rerunning prompts, and generally managing the bot like a distracted intern who needs constant supervision.

Do the math. The net gain shrinks fast.

What Botsitting Actually Looks Like

It’s not dramatic. That’s the point.

It’s re-prompting because the first output missed the tone. It’s fact-checking a summary that sounds confident but isn’t. It’s hunting down the right internal document so the AI stops hallucinating context it doesn’t have.

The Work AI Institute report put it plainly: there is a “thick, mostly invisible layer of human labor holding the whole thing together.”

Of total time spent interacting with AI each week, 37% goes to botsitting and only 36% to actually producing work. The rest? Overhead, setup, and restarts. More than a third of AI sessions fail outright — requiring a full do-over.

The Manager Problem Nobody Planned For

Here’s the structural issue, articulated well by Paul Leonardi, professor of technology management at UC Santa Barbara and a co-author of the study:

“We’re essentially expecting individual contributors to act as managers. They’re managing AI tools, AI agents — and we’re not taking into account all of the work that actually goes into managing.”

This reframing matters. Companies didn’t hire more managers. They handed management responsibilities to people who weren’t trained for it, didn’t ask for it, and aren’t being compensated for it.

The result is a workforce quietly absorbing a new cognitive load while the org chart stays the same.

Personal Gains, Organizational Gaps

The disconnect between individual productivity and business outcomes is where this trend gets genuinely strange.

75% of individual workers report a productivity boost from AI. Only 13% of organizations say they’ve seen significant business gains from AI adoption.

That gap isn’t a rounding error. It’s a structural failure.

Productivity gains are pooling at the individual level — in saved drafts, faster research, quicker first passes — but not converting into revenue, shipped products, or measurable growth. Uber reportedly burned through its entire 2026 AI budget in four months without shipping a usable feature. That’s an extreme case, but it rhymes with a broader pattern.

The Accountability Blind Spot

There’s a subtler cost hiding underneath the efficiency numbers.

The survey found that 41% of workers sometimes deliver AI-generated work they couldn’t explain if asked. The report illustrates this with a junior engineer who pasted thousands of lines of AI-generated code before going to bed — only for a senior engineer to spend hours untangling a bug the junior couldn’t diagnose or even describe.

When workers offload judgment to the bot, they also offload understanding. That’s fine until something breaks. Then it’s very much not fine.

The Honest Takeaway

AI tools are genuinely useful. The productivity gains are real. But the current moment is best understood as a transition tax — a period where the tools are capable enough to create dependency but not reliable enough to eliminate oversight.

Botsitting isn’t a bug in your workflow. It’s a feature of this phase.

The smarter move isn’t to use AI less. It’s to choose tools where the botsitting cost is low, the failure rate is transparent, and the output lands close enough to done that your six hours go somewhere better.

Observe the tools carefully. The ones worth keeping are the ones that don’t need babysitting.