AI is being taught how to do your job. Will artificial intelligence replace us, or just change the way we work?

There is a new boom industry in the American economy, and its raw material is you. As CBS News reported on 60 Minutes this weekend, a growing business now pays working practitioners — music producers, accountants, lawyers, nurses — to teach artificial intelligence the tricks of their trades. The people being hired for this work are not futurists or researchers. They are the people whose jobs the machines are being prepared to do.

One of them is Robbie Hiser, a music producer who spent years developing the instincts that separate a working professional from an enthusiastic amateur: when a bassline crowds a vocal, when a mix needs air, when a take that looks wrong on paper sounds right in the room. Hiser trained AI for Mercor, one of the companies in the emerging pipeline that connects domain experts with the labs building systems like ChatGPT, and he was paid to hand over some of those instincts. His view, as he told 60 Minutes, is that there will always be a place for creatives like him — even after teaching the machine a few of the tricks of his trade.

That sentence holds the entire tension of the next decade of work. The people best qualified to teach AI to do a job are, by definition, the people currently doing it. Some of them are being paid handsomely to help train their own would-be successors. But the evidence so far suggests the transaction is more complicated — and more human — than the headline version of the story.

The Machine Is Not the Expert

The uncomfortable truth of modern AI is that raw capability was never the bottleneck. Today’s models can already write, summarize, code and compose to a passable standard. What they lack is the tacit layer of expertise — the thousands of small judgments a professional makes without noticing, the context a veteran reads in seconds that a novice cannot see at all. That layer cannot be scraped from the open internet, because by definition it was never written down.

That is the gap companies like Mercor exist to close. Expert training — in which practitioners demonstrate, evaluate, correct and grade a model’s output in their specialty until its work starts to resemble the professional’s — has become a paying labor market precisely because the tacit layer is expensive and hard to get. Labs can buy compute by the acre, but judgment has to be extracted, one correction at a time, from the people who hold it. It is a striking inversion: the more an economy automates expertise, the more valuable the last holders of that expertise become, at least for as long as the extraction lasts.

Replacement or Reconfiguration

Every wave of automation arrives with the same warning — this time the machine takes the whole job — and the historical record keeps declining to cooperate. The power loom did not end textile work; it moved it up the stack. The spreadsheet did not eliminate accountants; it eliminated the adding-machine clerk and raised the value of the analyst. The compiler did not run out of programmers to serve; it multiplied how much software each one could ship. Automation, so far, has reconfigured occupations far faster than it has deleted them.

The AI version of that reconfiguration is visible inside the 60 Minutes piece itself. The music producer teaching the machine is not, in that moment, being replaced by his student; he is being paid as its tutor. The economics are uncomfortable but not senseless: a practitioner with rare judgment can now sell that judgment twice — once to clients, and once to the labs. The risk lands later, and on someone else: the beginner who once would have learned the trade by doing the first-draft work the machine now handles.

The Entry-Level Squeeze

That is where the damage is most plausibly concentrated, and it is the quiet part of the AI-jobs debate. The tasks today’s models handle best — research memos, first drafts, routine code, basic edits, summarization — are precisely the tasks organizations have historically assigned to newcomers. Economists scanning hiring data have found early signs of exactly that: softer demand for the bottom rungs of the professional ladder, even where demand for experienced people who can supervise, verify and correct machine output remains solid.

For anyone entering a knowledge trade, the implication is that the apprenticeship must now be seized rather than assigned. The credential that opens doors matters less than demonstrable judgment — a portfolio of decisions, corrections and shipped work that shows an employer exactly what the machine cannot yet do without you.

What This Means For You

If you are early in your career: treat AI fluency as table stakes, not a differentiator — assume every peer applicant has it. Spend your energy getting close to final judgment: review, editing, client contact, verification. Those are the seats automation is still a long way from taking, and they are where promotions now begin.

If you are a mid-career expert: your tacit knowledge is now a sellable product, and there is a genuine market for it through platforms like Mercor. Teaching the machine can be a side income, a strategic window into what the labs are building, or a moat around your own role. Whichever it is, get the terms in writing and know exactly which parts of your craft you are licensing.

If you manage people: redesign roles around the new division of labor — machine first drafts, human final judgment. Measure output and quality rather than hours, because the hours required to produce a competent first draft just collapsed.

And the headline question — replace or merely change? The honest answer from the current evidence is both, unevenly. Some occupations will compress; almost all will reconfigure. The workers with the most leverage are the ones who do what Hiser did: look at the machine honestly, price their own expertise, and decide deliberately which parts of the trade they are willing to teach.