AI Will Take Your Job (Just Not How You Think)

Illustration: giant robot in a business suit casually wheeling away an office chair while a human stares in confusion, flat vector, warm reds/oranges

The latest report says AI replacing jobs will vaporize 40% of white-collar work by 2030. Last month's report said 50%. Next month's will say 60%, presumably because the reports are being written by the AI that's doing the replacing. Somewhere between the doomsday keynotes and the LinkedIn fortune cookies ("AI won't take your job, but someone using AI will"), the actual story got lost. It's weirder, funnier, and much more annoying than either version.

The Apocalypse Keeps Getting Rescheduled

Let's start with the body count of predictions. In 2016, Geoffrey Hinton said we should stop training radiologists. A decade later, radiologist employment is fine, salaries are up, and the AI tools are… assisting radiologists. The robot didn't take the job; it took the tedious part of the job and handed everyone a bigger workload with better software.

The greatest hits of the AI replacing jobs era read like a blooper reel. Klarna's CEO bragged in 2024 that its AI assistant did the work of 700 full-time customer service agents. Glorious. Then in 2025 Klarna quietly started hiring humans again, because customers apparently enjoy talking to entities that have experienced childhood. IBM announced in 2023 that it would pause hiring for thousands of back-office roles that AI could do; the roles got filled anyway, slowly, by people. Duolingo declared itself "AI-first" in 2025 and learned that "AI-first" is corporate for "we'll be explaining this in a town hall for the next year."

Every prediction fails the same way: it models a job as a list of tasks, automates the list, and declares victory. Jobs were never lists of tasks.

What Actually Gets Eaten: The Intern Layer

Here's the pattern nobody puts in the keynote. AI doesn't replace jobs. It replaces the entry-level tasks inside jobs. The junior copywriter who wrote 40 product descriptions a day? That work now belongs to the great flood of machine-made mediocrity. Tier-one support macros, meeting summaries, the "draft a polite no" email — all gone to the machines.

Which means the career ladder lost its bottom rung. You used to learn the craft by doing the boring stuff for two years. Now the boring stuff is done by a model, and you're expected to arrive fully formed, like Athena, but with student debt. The cruelty of the current moment isn't mass unemployment — it's that getting your first job got harder while keeping your tenth got easier. The AI replacing jobs story is really an AI replacing on-ramps story, and nobody knows how to fix that with a keynote.

The New Jobs Are Weird

Every automation wave invents job titles that sound like pranks. Ours include:

Slop Inspector

Someone has to read the 10,000 AI-generated product descriptions before they go live, because the model keeps describing the toaster as "a revolutionary paradigm in breakfast." That someone is now a full-time role with a straight face and a red pen.

Prompt Plumber

The vibe-coded internal tool works until it doesn't, and then a human descends into the prompt chain with a wrench. This pays surprisingly well, which tells you everything about the current state of the art.

Model Babysitter

RLHF annotators, red-teamers, the people who teach the model that no, it should not help with that. The AI replacing jobs narrative never mentions that the replacement itself employs a small nation of contractors clicking "thumbs down" on bad outputs for a living.

Why the Charts Are Always Wrong

Three reasons the spreadsheet prophets keep missing:

Jobs are bundles, not tasks. "Copywriter" is 10% writing and 90% knowing that the client hates the word "synergy" because of a 2019 incident. Models do the 10% brilliantly and the 90% not at all. What gets automated is the task; what survives is the bundle — plus the accountability for when the bundle goes wrong.

Jevons paradox eats predictions. Make copywriting ten times cheaper and the world doesn't need ten times fewer copywriters — it demands a hundred times more copy. We saw it with spreadsheets (accountants multiplied), with ATMs (bank tellers multiplied), with every "this will kill the industry" technology since the loom. Demand expands to absorb the cheap thing, then invents expensive new things on top.

Someone has to sign. The last 5% of any consequential job is liability — a human name on the decision. Models can't be sued, fired, or meaningfully blamed, which turns out to be a load-bearing feature of employment. The signature is the moat.

So Who Should Actually Worry?

Be honest about where you sit on the automatability curve. If your job is 90% one repeatable task, done alone, with no relationships and no accountability — the yes-man assistant is already drafting your farewell speech. Pure data entry, first-draft factory work, rote triage: the machines are genuinely good at this now, and "genuinely good" is all the CFO needs to see.

But if your work involves judgment calls, weird edge cases, clients who need hand-holding, or consequences when things break — congratulations, you're not being replaced. You're being given a tireless intern who occasionally lies to you with perfect confidence. Manage accordingly: verify everything, trust the vibes never.

The Real Plot Twist

The AI replacing jobs discourse assumes a fixed number of jobs, like there's a pie and the robots are eating slices. But the pie keeps growing new slices — AI trainer, agent wrangler, synthetic data janitor — most of them jobs that didn't exist five years ago and sound fake today. The economy doesn't have a job shortage; it has a dignity shortage, a training shortage, and a shortage of anyone willing to explain to a 22-year-old how to get rung one back.

The robots aren't taking your job. They're taking the worst parts of everyone's job, breaking the ladder, and handing the survivors a shovel. It's not the apocalypse. It's just restructuring, with better branding.