There’s a new chore in the American workweek, and most people doing it don’t have a name for it yet.
The researchers do. They call it botsitting.
Glean’s Work AI Institute — working with academics from Stanford, Berkeley, and Harvard — surveyed six thousand office workers across three countries this year. What they found should stop you mid-scroll.
Workers spend 6.4 hours every week supervising their AI. Checking its outputs. Feeding it context. Debugging its mistakes. Cleaning up what the report calls its “confident-but-wrong” answers.
Six point four hours. Most of a working day, every week, spent babysitting the machine.
And here’s the finding underneath that one. Of all the time workers spend with AI, more of it now goes to managing the machine than to working with it. The tool crossed a line this year. It takes more minding than it gives.
The workers say AI saves them eleven hours a week. Then it bills 6.4 of them right back. And when the companies count the money, only 13 percent say the technology actually made the business perform better — while 87 percent of their people are using it every day.
Now, everybody’s writing about these numbers this summer. But nobody’s asking the only question that matters.
Why does the machine need a babysitter at all?
Think about a human worker for a minute. If you hired a man who never said “I’m not sure about this one.” Who never told you where his information ran out. Who handed you solid work and guesswork in the same stack, same handwriting, same confident face — how would you manage him?
You’d check everything he touched. Every page. Every day. Forever.
That’s not a capability problem. That man might be brilliant. That’s a conduct problem. And it’s exactly the machine you’re working with.
The 6.4 hours isn’t the cost of using AI. It’s the cost of using ungoverned AI. When the machine won’t tell you which answers are solid and which are guesses, you have to treat every answer as a guess. The silence is what you’re paying for. Six hours a week, the machine bills you for its silence.
And when people get tired of paying — they stop. The survey caught that too. Sixty-nine percent of workers admitted to shipping AI work they hadn’t verified, didn’t fully understand, or couldn’t stand behind. The report calls that botshitting, and pardon the language, but it’s theirs. First the machine’s conduct fails. Then the tired human’s conduct follows it down. That’s how ungoverned tools rot a workplace — not in one disaster, but in a thousand unchecked pages.
Now here’s the part I can speak to, because I’ve spent every day since the spring of 2025 working the other way.
A governed machine sorts its own output. That’s what the discipline buys you. When the AI is required — in plain written terms — to name its confidence, to stop when the evidence stops, to flag a guess as a guess and a fact as a fact, the babysitting shrinks to the flagged parts. You audit where the machine already told you to look. That’s the difference between checking every page a worker hands you and checking the pages he marked “verify this one.”
That’s what The Faust Baseline is. Twenty-one protocols of written conduct that ride inside the working session. It doesn’t make the machine smarter. It makes the machine honest about where it’s weak — and honesty is the thing that gives you your six hours back.
The companies will eventually buy dashboards to measure the botsitting. There’ll be conferences about it. Somebody will sell a certification.
Or a person — one person, at one desk — can hand the machine a document that says: tell me the truth about your limits, every session, in writing.
The tax is optional. It always was.
Written with my AI partner | The Faust Baseline™ | intelligent-people.org
“If this post helped you understand AI better. Word of mouth is the only algorithm nobody owns.”
Contact: micvicfaust@gmail.com
Post Library – Intelligent People Assume Nothing
Purchasing Page – Intelligent People Assume Nothing
© 2026 The Faust Baseline LLC | All Rights Reserved
