The Savings Have Arrived
Thirteen hours a week, per worker — that's the productivity gift Glean's new Work AI Institute survey says AI has handed the modern knowledge worker. Six thousand of them, actually: three thousand in the US, fifteen hundred each in the UK and Australia, surveyed this past December and January. Eighty-seven percent of them use AI. Seventy-three percent say it makes them more productive. The dashboards glow green. One presumes champagne was involved somewhere in a Glean marketing meeting.
And then, dear reader, there is the botsitting.
Six Point Four Hours
That's the other number in the study, and it's the one nobody put on a slide. Six point four hours a week — not saved, spent — on the unglamorous, untracked, unrewarded labor of keeping the AI functional. Feeding it context it should already have. Debugging outputs that fail for reasons no one can quite articulate, because the whole system runs on probability rather than logic, which is its own special kind of maddening. Cleaning up messes. Stitching together tools that, despite years of vendor promises, still do not talk to one another.
Half the gains, consumed by guarding the gains. The researchers named this botsitting. The term is exact, and — sources within the Work AI Institute Community confirm — not meant kindly.
Polished Nonsense
Thirty-six percent of AI sessions fail outright, requiring a full restart. And sixty-nine percent of workers admit to shipping AI-generated output they could not explain or defend if a manager asked — call it botshitting, a coinage so precise it belongs in a dictionary supplement no one asked for. Forty percent, give or take, hand in work that is, in the study's own memorable phrase, polished nonsense: finished-looking, confident-sounding, and hollow the moment anyone pokes at it.
Even now, the individual worker feels great about all this. Faster. Sharper. More capable. It's the organization that isn't feeling it — only thirteen percent report any real improvement in actual performance. Somewhere between the worker's inbox and the org chart, the gains evaporate. They evaporate, it turns out, directly into the botsitting hours.
Nobody Measured the Second Part
Rebecca Hinds, who runs the Work AI Institute, put it about as plainly as a research director is allowed to: coordination costs and misalignment between human and AI labor are eating the gains at the aggregate level. Her diagnosis for why — most companies are running what amounts to AI sprawl, a pile of tools that don't share context and don't know what each other is doing, which means a human has to be the glue. Every time. Manually. Forever, or at least until someone builds the thing Glean is, not coincidentally, selling.
And yet — as if this were not enough — the study found that workers doing the most botsitting are also the ones most actively job-hunting. Digital exhaustion, plus a quiet loss of faith that their employer has any real transformation plan beyond a town hall and a Slack emoji. Hinds has a name for the gap between the town hall and the plan, too: performative theater — executives narrating an AI revolution while making cuts that suggest they don't believe their own narration.
The machine is fast. Managing the machine is slow. Nobody put the second part on the org chart, which is how you end up with a workforce that is simultaneously more productive and more tired, and a company that is neither.
Thirteen hours saved. Six hours spent keeping the saver functional. Net: seven.
