The act of supervising AI agents – known as ‘botsitting – is increasingly weighing down employees and organisations, taking up to 6.4 hours per week of employees’ time. Here’s how organisations can get that time back.
A study of 6000 knowledge-industry workers in Australia, the US and the UK has produced a remarkable finding: employees reported spending 6.4 hours per week – almost a full workday – supervising and reviewing the work of AI.
This labour, known as botsitting, consumed 37 per cent of employees’ total AI time, more than the 36 per cent they spent using AI to actually produce work, according to the study.
The consequences for employers include disenfranchised teams and reduced organisational productivity, says organisational psychologist Rebecca Hinds, Head of the Work AI Institute at Glean, in a co-authored Harvard Business Review article summarising the research.
“AI promised to ease some of the burdens [of digital technology],” the article reads. “Instead, in most of the more than 100 organisations we’ve studied, it’s making them worse.”
Speaking to AHRI, Hinds says businesses’ rapid embrace of AI has created an assumption amongst some leaders that the tools are more sophisticated than they really are.
“The biggest driver of botsitting is when an AI is asked to complete a complex task without the necessary context,” she says.
“You can hook up all your documents to an AI tool and it can still have very little understanding of the relevance and importance of those documents, and which ones matter more to which person.”
Employees soon find themselves botsitting: struggling to prompt an AI tool to produce a more contextualised result or manually revising the work they had outsourced to AI in the first place.
A certain amount of this review work will always be necessary, note Hinds.
“But a lot of it is unproductive and tedious, and over time, the effects can be profound.”
“People start to resent the work they’re doing because it’s no longer the higher-order work that they were promised AI would free up time for. t’s monotonous, it’s tedious and it doesn’t give them meaning or purpose.” – Rebecca Hinds, Head of the Work AI Institute, Glean
Botsitting as a potential psychosocial hazard
Hinds points out that organisations “pay twice” when they ask an AI to complete a complex task that it’s not cut out for.
“Once for the additional computer-processing power consumed and once for the manual labour required to steer the tool towards a useful result.”
While this cost burden might not yet be visible at an organisational level, the burden of botsitting on individual workers has already become apparent, says Hinds.
“We asked people to what extent they felt worn out by this technology, and the more time they spent botsitting, as opposed to productively working, the more worn out they felt.”
The study also found that, as well as contributing to burnout, botsitting can contribute to fundamental job dissatisfaction.
“People start to resent the work they’re doing because it’s no longer the higher-order work that they were promised AI would free up time for,” says Hinds. “It’s monotonous, it’s tedious and it doesn’t give them meaning or purpose.”
Employees also said the time they saved on a weekly basis by using AI wasn’t translating into a more balanced experience at work.
While Australian respondents spent 6.5 hours per week botsitting and said AI completed the equivalent of 10 hours of work per week for them, some said they were unsure how to reinvest the net time saved.
“The default is one of two things: do more of the same work, which often contributes to burnout, or pocket the time savings, because you’re already burnt out from botsitting,” says Hinds. “Neither outcome is good for the organisation.”
Getting to grips with botsitting
Hinds says leaders frequently make three miscalculations when rolling out AI tools for their teams:
- Underestimating how much management AI requires (i.e. not factoring in botsitting time);
- Mistaking data access for useable context (i.e. equating data with the intangible knowledge and perspectives employees bring to tasks); and
- Measuring the wrong things (i.e. pushing employees to hit volume targets using AI tools, rather than focusing on quality).
Together, these miscalculations can cancel out many of the productivity gains promised by AI, according to Hinds.
The best way to counter the detrimental impacts of botsitting, she says, is to adopt a more measured view of the abilities and limitations of AI in its current form.
“It’s very easy to take this technology at face value and expect it to magically transform workflows, but problems arise if the hard work that employees need to do to oversee this technology isn’t recognised too.”
“The more time they spent botsitting, as opposed to productively working, the more worn out they felt.”
She offers a three-point plan for leaders grappling with botsitting:
- Rethink existing performance metrics
Employees who feel pressure to hit productivity targets are more likely to use AI tools without care, which can lead to increased botsitting and, ironically, reduced productivity.“Employees whose organisations measure quality-type metrics are significantly more likely to say AI has improved their work than those whose organisations only focus on efficiency or productivity,” says Hinds.
- Define a process for dealing with AI failure
Expecting employees to know how best to coax optimal results from novel AI tools is unrealistic and can contribute to excessive botsitting.
“If employees don’t have clarity on what to do when an AI-tool session fails – whether to feed it context or to re-prompt it, for example – their productivity and work satisfaction will suffer.” - Think of AI as a junior partner
Framing AI as an assistant, rather than a fully-fledged employee, can help employees botsit more mindfully.
“The more that employees feel like AI is a junior teammate, with humans in the driver’s seat and humans with accountability, the more they realise that the technology isn’t perfect,” says Hinds.
“They then tend to think more critically about AI output, and also not to give up after they give the AI its first prompt.”
Whether future advances in AI capability will ease or exacerbate the botsitting load remains unclear, says Hinds. For now, she encourages leaders to return to organisational fundamentals when developing strategies for AI use.
“Well-documented processes and a culture of experimentation and innovation – all the staples of highly effective organisations – are great foundations for building an AI strategy that ensures the amplification AI brings about is positive.”
