Will AI take your job? What has actually happened so far
Translators review machines now, illustrators lost clients, electricians boomed. What AI really automates, whose ladder is thinning, and what a normal person should do about it.
In 2023, the CEO of a Fortune 500 company could say "AI will transform our workforce" and mean nothing by it. By 2026, the sentence has bodies attached. Language-service firms reshaped themselves around machine translation, with human translators increasingly hired to review machine output instead of translating from scratch — same industry, different job, usually lower rate per word. Stock photo illustrators watched clients generate images for pennies. Customer support floors got thinner as chatbots absorbed the routine tickets. And in the same three years, AI companies could not hire fast enough, electricians and data center builders entered a boom, and "AI engineer" became the most demanded title in software.
Both things are true at once, which is exactly what makes this hard to think about. So skip the slogans — "AI will take all jobs" and "AI creates more jobs than it destroys" are both bumper stickers — and look at what actually happens.
What does AI actually automate?
Tasks, not jobs. Almost no job is one task. A paralegal reads documents, drafts summaries, manages filings, talks to anxious clients, and catches the thing that looks wrong. AI is now startlingly good at the first three, useless at the last two. So the job does not vanish; it gets recomposed. One paralegal with good tools does what took three, which is wonderful news for that paralegal and terrible news for the other two — and quietly worse news for the junior role that would have learned the trade by doing the tasks the machine now does.
That is the honest pattern across fields: the entry rungs get automated first, because entry-level work is precisely the routine, checkable, pattern-shaped work models handle best. Several studies through 2024 and 2025 found early signs of exactly this squeeze — hiring slowdowns concentrated in junior white-collar roles most exposed to AI, while senior roles held steadier. The ladder still exists. Its bottom rungs are getting thinner.
Which work is actually exposed?
Forget the manual-versus-office instinct; it points the wrong way now. A plumber under a sink, a nurse lifting a patient, an electrician in a crawlspace — those are among the hardest jobs on earth to automate, because the physical world is messy and robots remain expensive and clumsy. What is exposed is work that happens entirely on a screen, produces text, code, images, or decisions with patterns in them, and gets checked by someone else anyway: translation, routine copywriting, basic illustration and design, tier-one support, junior coding, form-heavy law and accounting, transcription.
Exposed does not mean eliminated. It means the machine does the first draft, and the human role migrates to judgment: deciding what to make, checking whether it is right, taking responsibility for it, and handling the cases that do not fit the pattern. The tasks that survive share a shape — accountability, taste, trust, physical presence, and blame. Someone must be sued when it goes wrong, and it will not be the chatbot.
Is this time different from every other panic?
Partly. The reassuring precedent is real: ATMs did not end bank tellers, spreadsheets did not end accountants — they ended certain tasks and the professions reorganized upward. Two hundred years of automation panic have mostly resolved into more jobs, different ones.
Two things are genuinely new, though. Speed: past transitions took a generation, long enough to retire out of the old job; this one is measured in product cycles. And target: this is the first automation wave aimed at cognitive work, the thing white-collar workers believed was safely theirs. Nobody can honestly tell you how it nets out, because the answer depends on how good the models get, and the people building them do not know either. Anyone selling certainty about 2030 employment — utopian or apocalyptic — is selling.
What should a normal person actually do?
Use the tools, seriously and now. Not because the slogan "you will be replaced by a person using AI" is deep, but because it is directionally right: in every observed rollout, the gap that matters is between workers who fold these tools into their day and workers who do not. Competence with AI is quickly becoming what spreadsheet competence was in 1995 — unremarkable to have, disqualifying to lack.
Then move your own weight toward what machines are bad at: owning outcomes, dealing with people, making judgment calls with incomplete information, physical skill, and being the person others trust when it is ambiguous. If you are early in a career, be aware the bottom rungs are crowded and automate-able — the fastest route up is becoming the person who wields the machine rather than the person who competes with it. If you are a parent, teach the child taste and judgment, not just output; output is what got cheap this decade.
And keep some humility handy, in both directions. The typists' pool disappeared, and the world filled with jobs no typist could have imagined. Something similar is likely on the far side of this — but likely is not guaranteed, and the crossing is being made at highway speed with no map. The only bad strategy is standing still and hoping it is hype.