Most conversations about artificial intelligence and employment get stuck on a single question, which is whether jobs disappear. Nicole Junkermann thinks that question is both hard to answer and less interesting than the one underneath it.
An investor and entrepreneur who publishes a series of brief spoken pieces on the subject, she has been making an argument that cuts across the usual optimism and pessimism. The immediate issue is not employment levels. It is that the tasks being automated first are, very often, precisely the tasks by which people used to learn their professions.
Here Nicole Junkermann sets out which capabilities become more valuable as the tools improve, which quietly lose their value, and why the pipeline that turns beginners into experts deserves more attention than it is getting.
“Nobody ever put it in a job description, but that work was the training”
Nicole Junkermann begins with a claim about how professions have always worked, and what has quietly been removed from them.
Start with the thing you say gets missed. What is it?
That the entry level work in most professions was never really about the output. It was how people learned.
A junior lawyer reading through hundreds of documents, a trainee analyst building a model from scratch, a young journalist making the routine calls, a graduate engineer fixing small defects in somebody else’s code. In every case the organisation wanted the output, and got it. But the person doing it was acquiring something else at the same time: a sense of what normal looks like, and therefore a sense of what wrong looks like.
Nobody ever put that in a job description. It happened as a side effect, reliably, for a very long time.
And that work is what is being automated.
Almost exactly that work, yes. It is structured, it is repetitive, it has a clear right answer often enough to be a good fit, and it was expensive because it consumed the time of people who were being paid to learn.
So it goes first. From a cost perspective that is entirely rational. The difficulty is that the training was invisible on the balance sheet, and things that are invisible on the balance sheet get removed without anyone deciding to remove them.
How long before that shows up as a problem?
Not immediately, which is part of what makes it awkward. The current senior people already have their judgement. They acquired it the old way, and it does not evaporate.
The gap appears in five or ten years, when the people who would have been moving into those senior roles have spent their early careers reviewing output rather than producing it. Reviewing is a real skill and I do not want to dismiss it. But it is not the same skill, and I am not confident it builds the same underlying sense of a subject.
“The ability to tell that something is subtly wrong”
On what gains value, Nicole Junkermann is specific rather than general, and she starts somewhere most lists do not.
Let’s talk about which capabilities gain value. What is at the top of your list?
The ability to tell that something is subtly wrong.
Not obviously wrong, which anyone can see. Subtly wrong. The number that is plausible but does not fit with something else on the page. The argument that is well constructed and rests on a premise nobody stated. The summary that is accurate in every particular and leaves out the thing that actually mattered.
That capacity is very hard to acquire in the abstract. It comes from having done the work often enough that deviation registers before you can articulate why, and it is becoming the difference between someone who can use these tools well and someone who is simply passing along whatever they produce.
Is that not just experience by another name?
It is a particular kind of experience, and I would distinguish it from seniority. Plenty of people accumulate years without accumulating that sense, usually because they moved into managing the work rather than doing it fairly early.
What builds it is volume and feedback. Doing a thing many times, being told when you got it wrong, and adjusting. That is why the entry level work mattered, and it is why the question of how people get that volume now is worth taking seriously.
What else appreciates?
Knowing what is worth doing at all. That has always been valuable and it becomes more so as producing things gets cheaper.
When drafting something took a week, the decision about whether to draft it received real attention. When it takes twenty minutes, the natural response is to produce more, and organisations are discovering that a large volume of competent material is not obviously better than a small volume of considered material. Somebody still has to decide what deserves to exist, and that judgement is not made faster by any tool.
Then there is the ability to specify. Getting something useful out of these systems is mostly a matter of being clear about what you actually want, which turns out to be a rarer skill than anyone expected. A great deal of what looks like poor output is a precise answer to a request that was never pinned down.
It is a distinction Nicole Junkermann draws repeatedly, between the work of producing something and the work of deciding what should exist.
And what loses value?
Producing a competent first draft, mainly. That was a genuine professional skill and it commanded real money, and it is now close to free.
The same applies to routine synthesis. Reading twelve documents and producing a fair summary was skilled work. It is now a starting point rather than a deliverable, and anyone whose value rested largely on doing it well has had a difficult couple of years.
I would rather say that plainly than pretend otherwise. What follows from it is not that those people have nothing to offer. It is that the offer has moved.
“Speed is exactly the wrong thing to optimise for in a first job”
On what to actually do about it, Nicole Junkermann is blunt about the cost, both for organisations and for the people starting out in them.
If organisations wanted to address the apprenticeship problem deliberately, what would that look like?
The uncomfortable answer is that it looks like deliberately doing some work the slow way.
Having a junior person produce something from scratch, occasionally, when a tool could have done it in a fraction of the time, purely because the exercise builds something. Then comparing the two and discussing the difference, which is where most of the learning actually sits.
That is expensive and it is very hard to justify in a quarter. It is also, I think, the thing that separates organisations that will have capable senior people in a decade from organisations that will be trying to hire them from somewhere else.
Is there a cheaper version?
There is a partial one, which is to change what juniors are asked to review.
If a young analyst only ever sees polished output, they learn what polish looks like. If they are shown the failures as well, the cases where the system produced something plausible and wrong, and asked to work out why, they start building the pattern recognition that matters. That costs almost nothing beyond the willingness to keep the failures rather than discard them.
Most organisations throw the failures away, which is a shame, because they are the most instructive material available.
What about the individuals themselves? What would you tell someone starting out now?
Resist the temptation to be fast.
Speed is exactly the wrong thing to optimise for in a first job, because speed is the part that has been commoditised. If your value proposition is producing acceptable work quickly, you are competing directly with something that will always be quicker.
The thing worth building is depth in something specific. Deep enough that you can tell when an answer in that domain is wrong. That takes a few years of unglamorous attention and it is the most durable position available, because the tools are broad rather than deep, and the checking has to be done by someone who genuinely knows.
Does that mean avoiding the tools while learning?
No, and I would be careful about that advice. Refusing to use them produces someone who is slow and inexperienced with the instruments everyone else uses, which helps nobody.
The distinction I would draw is between using them to skip the work and using them to see more of it. Having something explain a concept a third way when the textbook has not landed is a genuine benefit. Having it produce the thing you were supposed to produce, and submitting that, means the hours passed and nothing was built.
The difference is invisible from outside. Only the person doing it knows which one happened, which is precisely why it requires some discipline.
Nicole Junkermann offers one way of checking, aimed at the individual rather than the employer.
Is there a way for someone to tell whether they are actually building that judgement or just going through the motions?
There is a rough test I find useful, and it takes about a minute.
Think of the last piece of work you accepted from one of these systems. Could you have produced it yourself, given enough time? Not as well, necessarily, and not as quickly. But could you have got there?
If yes, you are using the tool as an accelerator, and the judgement is intact, because you would have recognised a bad version. If no, you are using it as a substitute, and you have no way of assessing what you received. That is fine occasionally, for things outside your field. It is a problem when it describes the core of what you are supposed to be good at.
And if the honest answer is no?
Then the work to do is obvious, if unwelcome. Pick a narrow part of the thing and learn it properly, the slow way, without assistance, until you could produce a decent version unaided.
Not the whole discipline, which is unrealistic. A narrow slice, deep enough that you would notice an error in it. That slice becomes the place you can genuinely stand behind your own output, and in my experience it tends to widen on its own once the habit exists.
“The scarce thing will be people who can tell”
Asked where all of this lands, Nicole Junkermann is more optimistic than the argument might suggest.
Where does this settle over the next several years?
My expectation is that the scarce thing will not be people who can use these tools. That will be ordinary, in the way that using a spreadsheet is ordinary, and it will not distinguish anyone.
The scarce thing will be people who can tell. Who can look at fluent, confident, well-organised output in their field and say that the third section does not hold, and explain why. There will be fewer of those people than we need, because the path that used to produce them has been partly removed, and nobody has yet built a replacement.
That sounds pessimistic.
I do not think it is, particularly. It is a solvable problem and the solution is not technically difficult. It requires organisations to treat the development of judgement as something they invest in on purpose, rather than something they used to get free as a by-product of work they no longer need done.
Plenty of professions have faced this before. Every field with a genuine apprenticeship tradition worked out how to build expertise deliberately, at some cost, because the alternative was not having any. This is the same problem arriving somewhere it has not been before.
What worries me is not that it cannot be solved. It is that the cost falls now and the benefit arrives in a decade, and that is the shape of problem organisations are worst at.
About Nicole Junkermann
Nicole Junkermann is an entrepreneur and investor with a long-standing interest in how technology reaches the people who end up using it. She publishes the AI Overview, a set of brief spoken pieces that take one idea at a time and set it out in plain language, at nicolejunkermann.ai.