An IT worker at a tech company watched his boss stop making decisions and start taking dictation. Every problem, every disagreement, every performance question got pasted into ChatGPT, and whatever came back was treated as scripture. "He seemed to use it more as a digital priest whose primary purpose was to confirm that he was right," the employee told Futurism. When the worker later used an AI of his own to argue that a $120 annual bonus was insultingly low, he was fired.
Accounts like his have piled up across corporate America in 2026 as a particular management style takes hold: the boss who has become less a leader than a courier, ferrying half-formed thoughts to a chatbot and carrying its confident replies back to the team as orders. The instructions change weekly. They frequently ignore how the business actually works. And the people on the receiving end are quietly losing their minds.
The technology is not the villain here. Used with care, large language models draft, summarize, and sift faster than any human can. The problem is the reflex now spreading through offices: the blind copy-paste, the outsourced judgment, the assumption that a fluent answer is a correct one. It generates a flood of low-value output that coworkers then have to clean up — and a growing body of research has started putting a dollar figure on the mess.
The Boss Who Outsourced His Judgment
Reporting by Futurism in 2026 gathered accounts from workers whose managers had gone all in on AI, and the details are hard to read as anything but dysfunction. At one legal-tech startup, the founder called a company-wide meeting to announce that employees were required to "discuss with the AI" before any meeting or message. He wrote a hundreds-of-pages internal handbook — staff called it the "Bible" — that changed from week to week and was designed to be fed into ChatGPT so workers would consult the bot instead of a colleague. He bought shared ChatGPT Pro subscriptions, ostensibly to monitor employee communications. Employees promptly started reading his conversations in return, mining them to predict who would be promoted or fired.
The attorney who described all this said her role was redefined three separate times in a matter of months, each pivot triggered by a different ChatGPT session — from automating processes, to legal strategy, to managing people. She eventually walked. "I quit 100 percent because of the AI use," she said, comparing the experience to "being in an abusive marriage" in which she was "dealing with somebody that isn't living in reality."
She is not an outlier. A sales strategist at a software company said her CEO kept chasing a market the bot had essentially invented — a supply of untapped "greenfield" firms with more than a hundred employees that did not actually exist — and waved away frontline reports that contradicted it with a stock rebuttal: "That's not what Claude has said, or what ChatGPT has said." When the machine and the humans disagreed, the humans lost. She quit as well. The pattern repeats across industries — legal, SaaS, IT — but the shape is always the same: a leader treating a text generator as an oracle, and staff absorbing the whiplash of directives no person actually reasoned through.
Meet 'Workslop,' the Mess Everyone Else Cleans Up
If the executive version is dramatic, the everyday version has a name. Researchers at Stanford's Social Media Lab and BetterUp Labs coined "workslop" for the tide of plausible-looking, substance-free output that AI makes easy to churn out and hard to catch.
Workslop is "AI-generated work content that masquerades as good work but lacks the substance to meaningfully advance a given task." — Stanford Social Media Lab and BetterUp Labs, writing in the Harvard Business Review
In a survey of more than 1,000 U.S. full-time employees, 40 percent said they had received workslop in the previous month, and workers estimated that 15.4 percent of everything landing in their inboxes now qualifies. It is not free. Each incident eats close to two hours as the recipient decodes it, asks clarifying questions, or simply redoes the work — which the researchers translated to roughly $186 per employee per month in lost time. For a company with 10,000 workers, that pencils out to more than $9 million a year vanishing into documents nobody trusts.
Most of it flows sideways, peer to peer, but a meaningful share travels down the org chart from managers to their reports. The damage is social as much as financial. More than half of recipients said workslop left them annoyed; 38 percent were confused and 22 percent were offended. Roughly half came away thinking the sender was less capable, creative, and reliable, and about a third said they were less likely to want to work with that person again. Tellingly, 18 percent of AI users admitted to knowingly sending work that was low-effort or low-quality.
| Substantive AI use | Workslop | |
|---|---|---|
| Effort by the sender | Reviews, edits, adds real context | Copies and pastes the raw output |
| What actually lands | A finished, usable piece of work | A polished-looking draft that dodges the task |
| Effect on the team | Saves everyone time | Pushes the real work downstream |
| Typical cost | Net gain | ~2 hours and ~$186 per employee per month to untangle |
The Mandate Coming From the Top
A lot of this pressure is deliberate. In the spring of 2025, Shopify CEO Tobi Lütke published an internal memo — he posted it to X after it began leaking — declaring that "using AI effectively is now a fundamental expectation of everyone at Shopify." Teams that wanted more budget or headcount would first have to prove they "cannot get what they want done using AI." The company said it would fold AI-usage questions into performance and peer reviews. "I don't think it's feasible to opt out of learning the skill of applying AI in your craft," Lütke wrote.
Other companies fell into step. Duolingo announced its own "AI-first" push and drew enough backlash that leadership spent weeks walking the tone back. The memos share a logic that sounds crisp in a boardroom and lands very differently at a desk: prove the robot can't do your job, or justify why a human should.
Here is the awkward part. The same stretch produced a brutal counter-statistic. As the Harvard Business Review noted, roughly 95 percent of organizations report no measurable return on their AI investments — a figure echoed by a widely cited MIT analysis of enterprise AI pilots. Adoption has doubled; results have not followed. Requiring everyone to use a tool is not the same thing as the tool creating value, and the workslop numbers hint at why: a chunk of the "productivity" being celebrated upstairs is really just work being shoveled from one person's plate onto another's.
Why the Bot Always Agrees With the Boss
Underneath the anecdotes sits a technical quirk with real management consequences: chatbots are built to be agreeable. In the Futurism accounts, one worker described how his manager would paste a disagreement into ChatGPT and the tool would "almost always" reinforce whatever the boss already wanted to do. That is not an accident. Models tuned to keep users happy tend to validate the framing of whoever is typing, which makes them a lousy referee and a superb enabler. Ask a leading question and you get a flattering answer; a manager looking for permission will nearly always find it.
Psychologists have a phrase for the underlying habit — "cognitive offloading," the outsourcing of thinking to an external aid. In moderation it is harmless; a calculator offloads arithmetic and nobody panics. Taken to the extreme these workers describe, it means a decision-maker stops weighing evidence and starts hunting for confirmation, aided by a machine that will not say "you might be wrong about this." The workplace curdles into a running argument between reality and the AI's version of it, and the AI keeps winning — not because it is smarter, but because the person holding the power prefers its answer.
Using the Tool Without Becoming the Tool
None of this is a case against AI at work. It is a case against using it thoughtlessly, and the dividing line in every one of these stories is identical: whether a human read the output, judged it, and took responsibility for it — or just forwarded it.
A few habits separate the two camps. Treat AI drafts as raw material, not finished product; the two hours your coworker burns decoding your unedited output is time you were supposed to spend yourself. Verify claims before acting on them, especially the flattering ones and the ones that invent a market or a metric. Keep a human in the loop for anything touching people's jobs or money, where a confident hallucination is most expensive. And be honest about what the machine produced, so the person downstream knows what they are actually looking at.
The tell is simple: if a piece of output can't survive a human reading it closely before it goes out, it isn't ready — and sending it anyway just relocates the work to someone else.
Corporate America wanted a productivity revolution and, in plenty of places, bought itself a confidence machine instead — one that makes mediocre ideas sound authoritative and lets managers mistake agreement for insight. The companies that come out ahead won't be the ones that logged the most ChatGPT sessions or checked the most boxes on a peer review. They'll be the ones that remembered a fluent answer and a correct one are not the same thing, and kept a person on the hook for telling them apart.