There was a woman at the post office — Darlene, third window, twenty-six years on the job — who kept a rubber band on her wrist. Every time management rolled out a new system, a new procedure, a new way of doing what she’d been doing since before the regional manager was born, she’d snap that rubber band. Didn’t say a word. Just snapped it, looked down, and went back to doing things the old way.
Management thought the new system was working. The reports said so. Darlene didn’t argue with reports. She just didn’t use the system. Neither did anyone else on the floor.
I thought about Darlene when I read that fifty-four percent of workers are avoiding their company’s AI tools. Not loudly. Not with protests or walkouts or strongly-worded emails to HR. They’re just not using them. More than half the workforce looked at the thing their bosses spent millions on and quietly went back to doing it themselves.
This is the most American form of rebellion there is. Not revolution. Not confrontation. Just a silent, collective refusal to participate in someone else’s fantasy.
The survey found that workers are spending eight hours a week cleaning up after AI. Eight hours. That’s a full workday, every week, spent fixing what the machine broke. Over a year, that adds up to fifty-one days. Fifty-one days of a human being’s limited time on earth, spent undoing what an algorithm did wrong.
Last year it was thirty-six. So we’re getting worse at this, not better. The machines are making more messes, faster, and the humans are mopping harder, slower, and nobody at the top seems to notice because the dashboard says everything is fine.
Here’s the number that gets me. Eighty-one percent of executives believe their AI deployments have “significantly improved productivity.” Twenty-one percent of workers agree. That’s a sixty-point gap. Sixty points between the people who approved the purchase order and the people who have to live with it. I’ve seen smaller gaps between opposing witnesses in a murder trial.
There’s a word for this in psychiatry. When a person’s beliefs about reality diverge this sharply from the evidence, they call it a delusion. When a CEO does it, they call it vision.
An economist at Johns Hopkins put it plainly enough. “AI didn’t deliver,” he said. “Welcome to the real world.” He said productivity should be way up if the hype were true. It isn’t. GDP growth hasn’t budged toward the five or six percent the Silicon Valley prophets promised. The revolution is running behind schedule, and nobody thought to check with the people doing the actual work.
If ninety-five percent of AI deployments fail to generate expected ROI — and that’s the number MIT came back with, not some crank with a blog — then what we’re looking at isn’t a technology problem. It’s a faith problem. A ninety-five percent failure rate in any other industry would end careers. If a restaurant failed ninety-five percent of its entrées, you wouldn’t eat there. If a surgeon botched ninety-five out of a hundred operations, they’d be in prison. But in tech, a ninety-five percent failure rate just means you need more funding.
I worked at a place once — a packaging plant outside San Pedro — where the foreman bought a machine that was supposed to replace three guys on the line. Beautiful machine. German engineering. The brochure had glossy photos and words like “throughput” and “seamless integration.” It arrived on a Tuesday. By Thursday, five guys were working the line instead of three — the original three, plus two more to babysit the machine that kept jamming.
The foreman never admitted it was a mistake. He couldn’t. He’d pitched the machine to the owner as the future. He’d shown the brochure. He’d used the word “ROI” in a meeting, and once you say ROI in a meeting you’ve committed yourself to a reality that doesn’t exist yet, and admitting it doesn’t exist means admitting you don’t know what you’re talking about. So the machine stayed. And five guys worked around it. And the reports went upstairs saying productivity was up, because the reports measured output, not effort — and output was technically higher, it just took five people and a German machine to do what three people had been doing with their hands.
That’s where we are. The machine is on the floor. It jams. Five people work around it. The dashboard says it’s working. And the executives believe the dashboard because the alternative — that they spent millions on a glossy brochure with a login page — is too painful to consider.
The workers know. They’ve always known. Darlene knew. The guys at the packaging plant knew. The fifty-four percent who quietly ignore the AI tools at their desks — they know. They just can’t say it, because questioning the judgment of people who control your paycheck has never been a winning strategy for anyone who needs to make rent.
So they do what workers have always done. They work around it. They fix the messes quietly. They nod in the meetings and go back to their desks and do things the way that actually works. They snap the rubber band and keep their mouths shut.
The executives will figure it out eventually. Right around the time the next shiny thing comes along, and they can blame this one on “implementation challenges” and move the budget to whatever the new brochure promises.
The fifty-one days will get absorbed into the general exhaustion of being alive in 2026, where everything is smart except the decisions being made about it. Nobody will get those days back. Nobody will apologize for taking them.
Darlene retired in 2004. She took her rubber band with her. I sometimes wonder if she’s still snapping it somewhere, watching this whole thing unfold, not saying a word.
Source: There’s a Mass Rebellion Against AI in the Workplace