Where does a senior engineer come from?
The only answer I know of is unglamorous. They were a junior one for years, on real work, with somebody more experienced telling them what they had got wrong. Nobody has found a way to produce the judgment without the years, and I think that is because most of what we call judgment is a memory of things that went badly because of you. Not a memory you can acquire from reading about somebody else’s.
Here is the part I find genuinely odd, and it is an accounting observation rather than a sentimental one. That training was never something anybody bought. A firm hires a junior analyst and pays her a salary. In exchange it gets a year of work — models rebuilt, numbers checked against the source, a deck reformatted at eleven at night — roughly what the salary was for. But the same arrangement also produces, at no extra charge, a person who in six years will be worth several times what she costs now and hard to hire from outside at any price. Two things came out of one transaction. Only one was ever priced.
Anything with no price is invisible the moment somebody does the arithmetic. Suppose a tool arrives that checks the numbers and rebuilds the model at a tenth of the cost. The comparison in that meeting is one priced thing against another, and it is not close. The second output, the one nobody was charged for, appears on neither side of the sheet. So it goes first, because the work carrying it is exactly the work worth automating: repetitive, well-specified, tedious. Junior work and automatable work are close to the same list.
You might think a firm that sees this coming would simply pay for the training on purpose. Some will. But notice what changes when training stops being a byproduct and becomes a purchase. The firm carries the whole cost, and the person it trained is free to take that value anywhere, including to a competitor who paid nothing towards it. Under the old arrangement this barely mattered, because the training rode along free on work you needed done anyway. Now it is a visible line item whose return is partly captured by the rest of the industry — the kind of cost nobody volunteers for.
So my worry is narrow: what used to happen for free now needs somebody to choose to pay for it, and that choice is easy to defer. The saving is immediate and lands on this year’s numbers. The shortage arrives in a decade, in somebody else’s org chart.
Now let me argue against myself, because the counter-case is a good one.
Apprenticeship was a dreadful teacher. Enormous amounts of junior work taught nothing — chasing a missing signature for three days, retyping figures from a PDF, sitting in a meeting you were not allowed to speak in. None of that built judgment, and we should not write a eulogy for drudgery just because skill happened to be standing nearby. Strip the waste out and a junior might spend those two years reading real incident reports, being questioned by something that never gets bored, and reviewing what an AI produced. Reviewing is a real teacher; editors learn by editing. And crafts have relocated before without much being lost — typesetting had a long apprenticeship and is gone, while the ability it served, making a page readable, moved into design and software.
Where the counter-case falls short, I think, is on one thing. Reviewing output teaches you to recognise what is wrong in front of you. I am not convinced it teaches you to anticipate what is likely to go wrong, because that seems to come from having been the person who caused it, with your name on the change. Recognition and anticipation feel like different abilities, learned different ways. This is reasoning from how it felt to learn my own work rather than evidence, and it is the sort of intuition that turns out sentimental half the time.
Two fields already behave as though the distinction is real. Surgical training still puts a trainee’s hands on a patient under supervision, because watching has never been found to be enough. Aviation went further into automation than almost anyone, then noticed that pilots who rarely flew by hand were losing the feel for it, and regulators began encouraging airlines to give crews deliberate manual practice — practice that produces nothing, exists only to keep an ability alive, and must be paid for.
So here is what I think. The skill will not vanish; the next generation can probably be taught differently and perhaps better. That was never the hard part. The hard part is that the old arrangement required nobody to choose to fund it, and the new one does, with the bill landing on whoever moves first.
What follows is narrow but real. The training juniors used to absorb from the boring work will not reappear on its own once the boring work is gone, and no line in your accounts will report it missing. It has to be bought deliberately, at a cost you can see, for a return you cannot. The fields that already do this are the ones where being wrong is expensive and slow to show itself. Software has mostly not had to think that way, and I suspect it is about to.