Workers are supervising the supervisor instead of doing the work itself
Across Britain's workplaces, a quiet paradox has taken hold: the tools deployed to liberate workers from repetitive labor have instead created a new form of it. Nearly six hours each week, employees now spend watching over artificial intelligence systems — catching errors, verifying outputs, preventing embarrassments — a practice already earning its own name, 'botsitting.' It is a reminder that technology, however powerful, does not transform organizations on its own; the human work of learning, adapting, and building trust in new systems cannot be skipped, only deferred — and deferred at a cost.
- British workers are losing the equivalent of a full working day every week not to distraction or inefficiency, but to supervising the AI systems meant to make them more efficient.
- Only one in four employees feels adequately trained to use the AI tools their employers have already deployed, leaving most workers in a state of low-level uncertainty about whether they're using these systems correctly.
- Companies moved fast on adoption — pressured by competition and vendor promises — but skipped the slower, unglamorous work of building real workforce readiness around these tools.
- The productivity windfall executives anticipated has not arrived; instead, organizations are paying a hidden tax in lost hours without recognizing it on any balance sheet.
- Until training catches up to deployment and workflows are redesigned around what AI can actually do well, the efficiency gains will remain a promise rather than a reality.
Something unexpected has taken root in Britain's offices: a new form of unpaid, unacknowledged labor. Workers across industries are spending close to six hours every week watching over AI systems — verifying their outputs, catching hallucinated facts, preventing errors from reaching clients. The practice has earned a name, 'botsitting,' and it quietly undermines the very productivity case that justified AI adoption in the first place.
The six hours are not trivial. For engineers, analysts, and creators, that is a full working day each week redirected away from strategy and craft toward the supervision of a tool that was supposed to reduce supervision. The opportunity cost is real, even if it rarely appears in any formal accounting.
Deepening the problem is a stark training gap. Only one in four UK employees reports feeling genuinely prepared to use the AI tools their companies have handed them. Organizations moved quickly on deployment — the competitive pressure was real, the technology seductive — but moved slowly on the harder work of teaching people to use it well. The result is a workforce left to figure things out alone, spending hours in low-level uncertainty rather than confident productivity.
This is not a story about technology failing or workers resisting change. It is a story about the mismatch between how fast a tool can be deployed and how long it actually takes an organization to learn, adapt, and build the right processes around it. Deployment takes weeks; genuine integration takes months or years — and most companies haven't done that work yet.
The cost of that gap is being paid every week, invisibly, in hours spent watching machines rather than moving work forward. Until training catches up and organizations build honest clarity about what AI is actually for, the promised efficiency gains will keep arriving just slightly out of reach.
Across Britain's offices and tech hubs, a peculiar new job has emerged—one that didn't exist five years ago and pays nothing extra. Workers are spending nearly six hours each week doing what's come to be called "botsitting": watching over artificial intelligence systems, checking their outputs, catching their mistakes, making sure they haven't hallucinated a fact or invented a citation or sent a client something embarrassing.
It's a form of labor that reveals the gap between what companies promised when they deployed AI and what's actually happening on the ground. The technology was supposed to accelerate work, to let people focus on the thinking while machines handled the grunt. Instead, many workers find themselves in a new kind of grunt work—the work of supervising the supervisor.
The six-hour weekly figure is substantial. That's a full working day, every week, spent not on the tasks AI was meant to augment but on the task of making sure AI doesn't break things. For a software engineer, a designer, a content creator, or an analyst, six hours represents real opportunity cost. It's time not spent on strategy, on creation, on the work that actually moves a business forward.
What makes this particularly acute is the training gap. Only one in four employees across the UK workforce report feeling fully prepared to use the AI tools their companies have handed them. Companies have moved fast on deployment—the technology is seductive, the competitive pressure is real, the vendors are persuasive—but they've moved slowly on the unglamorous work of actually teaching people how to use it well. The result is a workforce that's been given a powerful tool and told to figure it out, which means spending hours each week in a state of low-level anxiety, double-checking outputs, wondering if they're using the system correctly.
The irony is sharp. AI was supposed to deliver a productivity windfall. Companies invested in it, restructured around it, made announcements about it. But the payoff hasn't materialized the way executives imagined. Instead of workers doing more in less time, workers are doing the same amount of work while also supervising an AI system that's supposed to be helping them. The math doesn't work. The promise hasn't been kept.
This isn't a story about AI being bad or workers being Luddites. It's a story about the gap between the speed of technological change and the speed of organizational learning. Companies can deploy a new tool in weeks. Teaching a workforce to use it well, building the processes and safeguards around it, understanding what it's actually good for and what it's not—that takes months, sometimes years. And most organizations haven't done that work yet.
The question now is whether they will. The six-hour weekly cost of botsitting is a tax on productivity that companies are paying without realizing it. Until training catches up to deployment, until organizations clarify what AI is actually for and build the workflows around it, that tax will keep getting paid. Workers will keep watching the machines, waiting for the efficiency gains that were promised to arrive.
A Conversa do Hearth Outra perspectiva sobre a história
So these six hours a week—is that time spent actively fixing things the AI gets wrong, or is it more like... nervous checking?
Both, honestly. Some of it's catching real errors before they go out. But a lot of it is the anxiety of not knowing if you're using it right. If you haven't been trained, you don't trust the output, so you verify everything.
And the companies know this is happening?
Not really, or not in a systematic way. It's invisible labor. It's not on a timesheet. Workers just absorb it into their day and feel like they're working harder, not smarter.
Why haven't companies trained people?
Speed. They wanted to move fast, show they were modern, get competitive advantage. Training is slow and boring and doesn't look like innovation. So they skipped it.
So the AI is actually making work slower?
Not slower, exactly. But it's not making it faster either, which is what was promised. The productivity gain is being eaten by the supervision cost.
What would actually fix this?
Real training. Clear guidelines about what the tool is and isn't for. Building workflows that account for the fact that AI needs human oversight. Treating it like a serious change, not just a new software rollout.