There was a guy I worked with at the post office. Name was Delgado. He’d been sorting mail for thirty-one years. Knew every route in the district by feel — which streets flooded, which dogs bit, which old ladies left water out on the porch in August.
When I started, he didn’t say much. Just watched me fumble with the trays for a week. Then one morning he said, “You’re stacking the flats wrong.” Showed me once. That was the training program. Thirty-one years of knowing compressed into four words and a gesture.
I thought about Delgado when I read that IBM is tripling its entry-level hiring this year. The same IBM whose CEO told Bloomberg in 2023 that AI would replace roughly 7,800 back-office jobs. Three years. That’s all it took to discover what Delgado could’ve told them for free: you can’t replace what you refuse to teach.
They’re calling it “Experience Starvation” now. Got a name for it and everything. Gartner coined it, which means it’s official enough to bill for. The idea is simple: when senior people use AI instead of mentoring junior people, the junior people never become senior people. The pipeline collapses. The ladder gets pulled up. And then everybody acts surprised when there’s nobody left who knows how to do the job.
New graduates now make up seven percent of new hires at Big Tech. Seven. Down from fifty percent before the pandemic. Half the people used to be new. Now it’s one in fourteen. The rest of the chairs are filled by AI or by mid-level people poached from competitors at a thirty percent salary premium — people who still don’t know where the bathroom is, let alone how the culture works.
Klarna figured this out the expensive way. Fintech company, Swedish, the kind of outfit that puts ping-pong tables in the break room and calls it innovation. They cut 700 customer service people and replaced them with an OpenAI-powered chatbot. Customer satisfaction dropped. Quality went to hell. By mid-2025 they were rehiring humans. Their CEO said, “We focused too much on efficiency and cost. The result was lower quality.”
No shit.
I’ve worked enough bad jobs to know that the people who run companies don’t understand what the people who work in companies actually do. They see a task list and think: this can be automated. They don’t see the judgment calls, the workarounds, the institutional memory that lives in somebody’s head because nobody ever wrote it down. They don’t see Delgado.
A Harvard study found that AI delivered forty percent quality improvements on tasks within its wheelhouse. But on tasks outside that narrow lane, the people using AI were nineteen percentage points worse than people working without it. The tool made them dumber. Not because the tool is bad, but because they trusted it past its limits. Like a guy with a flashlight thinking he can see the whole cave.
The AWS CEO said replacing junior workers with AI is “one of the dumbest things I’ve ever heard.” And this is a man who sells cloud computing for a living. When the guy selling the shovels tells you to stop digging, maybe put the shovel down.
IBM isn’t admitting they were wrong, of course. They’re “strategically recalibrating.” Their HR chief personally rewrote the entry-level job descriptions. Now junior developers do less coding and more “client engagement” and “AI output correction.” Which is a fancy way of saying: we need humans to fix what the machines get wrong, and we need them cheap.
The company line is: the jobs changed, the humans stayed.
What I hear is different. What I hear is: we burned the apprenticeship model, panicked when we realized nobody’s learning the trade, and now we’re hiring kids at entry-level wages to babysit our AI. We’re not investing in people. We’re investing in error correction. The junior employees aren’t being mentored. They’re being positioned as guardrails.
A professor at UC Santa Barbara said it plain: “Cleanup is always harder than prevention.” Three to five years from now, the companies that gutted their pipelines will face what he calls “a new nasty set of problems.” The kind that don’t show up in quarterly earnings but eat the foundation out from under you while everyone upstairs is celebrating the numbers.
Fifty-five percent of employers who laid off staff because of AI now admit they made the wrong call. That’s not a margin of error. That’s a majority of people who burned the house down and are now standing in the ashes wondering where they put the blueprints. Forrester says half of AI-attributed layoffs will be quietly reversed by end of this year. Quietly. Because nobody holds a press conference to say, “We were idiots.” They just post new job listings and hope nobody notices.
Delgado retired two years before I left the post office. Nobody replaced him. Not really. They brought in a kid who could work the new sorting machine but didn’t know which routes flooded. First heavy rain, half the district’s mail came back wet.
The machine worked fine. It was everything around the machine that fell apart.
That’s always how it goes.
Source: Corporate America Is Rethinking AI Workforce Needs, Led By IBM