A director I worked with had a Monday habit that looked like efficiency. He would pull his team's weekend updates into a chatbot, ask it for a summary, and forward the result to his own boss before his coffee had cooled. It read well, clean paragraphs and confident numbers, the kind of tidy that makes a busy executive nod and move on. Then one Monday the summary carried a shipment volume that was off by a factor of ten, and nobody upstream noticed until a planning meeting three days later had already built a staffing decision on the wrong figure. He had not lied to anyone. He had done something more ordinary and far more common. He had passed along a document he never actually read.

There is a name now for what he produced. Researchers at BetterUp Labs, a leadership-development firm, working with Stanford's Social Media Lab, called it workslop: AI-generated output that looks finished but carries no real thinking, which forces the person who receives it to supply the thinking that got skipped.1 In their September 2025 survey of 1,150 U.S. desk workers, forty percent reported being handed workslop in the previous month, and each instance took close to two hours to untangle.1 For each worker who receives it, they put the hidden tax at about 186 dollars a month, and they estimate that across a ten thousand person company, once you account for how many are actually hit, it adds up to somewhere near nine million dollars a year.1 BetterUp sells coaching, so hold the dollar figures loosely, though the underlying mechanism is hard to wave away.

The cost does not vanish, it moves

What makes workslop expensive has less to do with the flawed output itself and more to do with where the work travels after it leaves the sender's hands. When a leader generates a report and forwards it without reading it as though he wrote it, he has not erased the labor of judgment. He has relocated that labor onto whoever opens the file next, usually a person with less context and less standing to push back. Polish is what makes it dangerous. A rough draft announces that it needs a second look, while a fluent one implies that it is done, and busy people tend to believe the implication.

Speed at the top becomes friction at the bottom. The leader saves ten minutes and spends two hours of someone else's, and the organization never sees the transfer recorded on any ledger.

I watched a version of this at a manufacturer running dozens of production lines. Their teams were using an AI assistant reactively, the way people first approach a search box, typing a question and pasting back whatever came out. The exposure was straightforward. An automated alert or a summarized metric could travel into a real production decision without a single human weighing whether the number made sense. On a floor with that many lines, a plausible wrong figure does more harm than an obvious one, because a plausible number gets acted on before anyone thinks to question it.

When it comes from the corner office

The newer wrinkle is who tends to be sending it. A 2026 survey by Zety, a career-services company, found that fifty-five percent of employees had received workslop from a supervisor.2 Of those, eighty-five percent said it damaged their trust in that leader, and seventy-four percent said it lowered their confidence in the person's work overall.2 Sit with what that does to a team. The shortcut a leader takes to look responsive gets read one floor down as a statement about standards. Nobody there reads it as efficiency. They read it as a leader who could not be bothered to check, and they start applying the same arithmetic to everything else that arrives from above.

This is why I treat unreviewed AI output as a question of accountability rather than a question of technology. A model can draft. It cannot own. The instant a leader forwards something with his name on it, he has vouched for it, whether or not he read a line. Ownership is the one part of the exchange a machine cannot supply, and it happens to be exactly the part a forwarded-but-unread document strips out.

A discipline worth the name

Banning the tools would be foolish, and it would not hold anyway. What works is a sequence I walk through with the leaders I coach. Generate, and let the model produce the draft. Review, and read that draft as though you will have to defend every line of it in a room, because at some point you will. Own, and send only what you would put your signature beneath. The review step is where a leader's judgment actually lives. Treating it as busywork stapled onto the real work gets the job backward, since it is the one piece the machine was never going to handle for you.

None of this argues against using AI. I run most of my own writing through a model before a human, me, decides what ships. The point is smaller and harder to sit with. The value a leader adds was never the production of pages. It was the judgment about whether the pages were right and whether they belonged in front of anyone at all. A tool that makes pages effortless does not retire that judgment. It leaves that judgment as the only thing still worth a leader's name. Your people are already using AI, so that is not the real question. The real question sits closer to home: before the next summary goes out under your name, did you read it?