Is the veneer cracking on the golden promise of AI? A major investigation carried out by the Work AI Institute de Glean, in collaboration with researchers from Stanford and Berkeley, is throwing a wrench in the pond.
After surveying 6,000 office workers in the United States, the United Kingdom and Australia, the findings are stark: employees spend on average 6.4 hours per week to supervise, correct and debug AI tools.
This is the equivalent of a full day of work that evaporates each week into a hitherto invisible task. Welcome to the era of stability which cancels out productivity gains and degrades morale.
What is “botsitting”, this new invisible burden?
The stability is the term created by the report’s authors to describe all the work needed to make AI truly useful. It is about feed the tool with the right context, to meticulously check its results, to correct its sometimes gross errors and to rerun the queries until acceptable output is obtained.
It’s a you baby-sit syndrome robot, an unrecognized, unmeasured and above all unvalued task in the company. This shadow work is described as “ tedious » et « exhausting » by Rebecca Hinds, director of the institute.
Far from liberating humans, AI imposes a new layer of technological micromanagement. For every hour spent getting raw output from AI, employees spend almost another hour making it usable.
The human labor then transforms into permanent after-sales service for the machine.
Why doesn’t individual productivity improve that of the company?
This is the great paradox raised by the study. If 75% of employees believe that AI makes them more productive individually, only 13% of companies report seeing a significant improvement in their overall performance.
The answer is simple: the theoretical time savings are entirely absorbed by botsitting. The time “saved” on one side is immediately reinvested in monitoring the tool on the other.
Worse still, this situation gives rise to an even more dangerous phenomenon: the “ botshitting “. Exhausted by this constant supervision, 69% of AI users admit to having already submitted model-generated work without thoroughly checking it.
They disempower themselves, a psychological mechanism called “ moral disengagement “. This drop in vigilance on the productivity real exposes the company to critical errors, hidden under a veneer of technological efficiency.
What are the risks of botsitting for employees and businesses?
L’impact on the morale of the troops is devastating. An employee who spends a significant portion of his time botsitting has 73% more chance to actively look for a new job.
Frustration builds, resentment grows, and resumes get updated. Technology sold as liberating but in reality pushes talent towards the exit by confining them to verification tasks without added value.
The danger goes beyond simple overloading. Employees are asked to automate the parts of their job that they enjoy most, such as customer relations, to become simple chatbot supervisors.
This empties their job of its meaning and substance. For the company, the risk is twofold: drain of talent and an increase in costly errors due to botshitting.
How can businesses escape the botsitting trap?
The solution is not to deploy even more tools but to completely rethink their integration. According to the report, the companies that are doing well are not those that use AI the most but those that spend the most time on it. work “around” AI.
They invest in training, define clear quality standards and help employees develop their critical judgment. It’s about deciding in advance what should be delegated to a model and what should never be.
The key is not blind adoption, but operational discipline rigorous. Without this strategy, the hours “earned” thanks to artificial intelligence will continue to dissolve into the time lost cleaning up its mistakes, paying a high price with the departure of people who have had enough of playing babysitters.
