There is something almost comedic about watching the news cycle catch up to common sense. Companies fired people, deployed chatbots, watched customer satisfaction drop, and are now quietly rehiring. Sometimes the same people, often at lower salaries, occasionally offshore. The headlines call it a correction. I’d call it what happens when you let a narrative and the hype do your thinking for you.
The hype did the thinking
We’ve spent the last few years inside a very loud conversation about AI and jobs. Would it replace us? Would it free us? Would it make us obsolete or simply irrelevant? The fear was real, the hype was louder, and somewhere in the middle, actual decisions were being made by actual companies, with actual consequences for actual people.
And now, the data is starting to come in. Forrester Research found that 55% of employers regret laying off workers in the name of AI efficiency. More than a third spent more money on rehiring than they ever saved from the cuts in the first place. Klarna, which was held up as a bold AI success story, replaced 700 customer service employees, watched quality drop, and customer satisfaction decline, and ended up bringing humans back. The companies that were supposed to be showing us the future were quietly reversing course before anyone was paying close attention.
What the numbers actually say
The Oxford Economics put it plainly: companies were laying off workers based on AI’s potential, not its actual performance. The job losses were real. The capabilities meant to justify them, in many cases, weren’t there yet. And in the cases where AI did deliver (narrow, repetitive, high-volume tasks), it delivered exactly as well as anyone who had studied the technology would have predicted. Some processes can be automated. That part was never really in question. What can’t be automated so easily is everything that happens around those processes: the judgment calls, the pivots, the moments when something unexpected breaks the pattern.
Economists sometimes call this “so-so automation“, a term from Nobel laureate Daron Acemoglu, describing technology that displaces workers without delivering real productivity gains. It happens when the decision to automate is driven by pressure to look modern rather than a genuine understanding of what the technology can do. Yale’s Budget Lab analyzed U.S. labor market data from late 2022 through 2025 and found almost no measurable shift in the occupational mix. The mass displacement narrative was, statistically speaking, mostly noise.
What was real, and what the noise drowned out, was something quieter and, honestly, more troubling. Entry-level job postings dropped roughly 35% since early 2023. Not because AI replaced those workers, but because companies used the AI story as cover for post-pandemic headcount corrections. The technology became a convenient headline. And Gen Z workers, the generation that actually has the highest AI readiness of any demographic, were shut out of the entry-level roles they would have filled, cutting off the very people who could have helped these organizations use AI well.
And that, I think, is the part that deserves more than frustration, because the failure wasn’t really about AI at all.
The question nobody asked first
The companies that stumbled didn’t stumble because AI is incapable. Some things AI does remarkably well (we’ve been through that in Between Circuits and Purpose and Careers for a Soft Future). They stumbled because they skipped a step. The step where you actually think.
The question that should have come first, how does our work actually function, and where specifically would AI help?, got replaced by a different question: how do we signal that we’re using AI? Those are not the same question, and they don’t lead to the same decisions.
The space between the tasks
Real work, especially in knowledge environments, isn’t a clean list of tasks you can hand off one by one. It’s a web of context, judgment, institutional memory, and constant small pivots. You can automate the predictable parts. But most of the value lives in the space between the predictable parts. In noticing that something is quietly going wrong, in making a call the algorithm wasn’t trained for, in reading a room and responding to what’s actually happening rather than what was expected. That’s not a limitation AI will simply outgrow next year. It’s structural, and it matters.
I also want to acknowledge the opposite reaction, because it’s just as unhelpful. The backlash to these headlines has produced its own overcorrection. The confident declaration that AI is a gimmick, a bubble, a solution in search of a problem. We looked at the bubble dynamics in The Illusion of Endless Growth, and yes, the hype cycle is real. But “it didn’t work the way companies deployed it” is not the same thing as “it doesn’t work.” Treating those as equivalent is its own form of lazy thinking, just wearing a more skeptical outfit.
The more interesting question, the one harder to answer but worth sitting with, is what it means that so many organizations made significant decisions about people’s livelihoods based on a story they’d never stopped to interrogate. Not out of malice, mostly. Just without pausing to ask whether the narrative matched the reality of how their work actually got done.
Common sense, applied early
AI will keep improving. Some of what it can’t do today, it will eventually do. Probably. But the capacity to think carefully before acting, to ask the obvious question before making the call, that’s a human skill, and it’s one that was notably absent here. The lesson isn’t really about AI. It’s about what happens when the pressure to appear forward-thinking overrides the discipline of actually thinking.
Maybe the correction was inevitable. Maybe it’s even useful. If it slows down the reflex to automate first and ask questions later, something valuable came out of the mess. Not vindication, exactly. Just a reminder that common sense, applied early, is still cheaper than a Forrester report applied after the fact.