Almost everyone who is good at their job got there the same unglamorous way. They started at the bottom and did the work nobody else wanted. They pulled the research nobody had time to pull, wrote the first drafts that would be torn apart and rewritten, formatted the deck, checked the numbers, sat in the corner of meetings they did not yet understand and took notes they were not sure mattered. 

It felt, at the time, like the price of admission, the dues you paid before the real work began. We were wrong about that. The dues were the real work. That grunt work was an apprenticeship in disguise, and we learnt the craft by doing the small things badly and then, slowly, a little better.

I have been thinking about this because the layer of work I am describing is precisely the layer that Artificial Intelligence (AI) now does best. The research summary, the competent first draft, the cleaned-up data, the tidy version of the messy notes. These are no longer the tasks you hand to the nervous 23-year-old to cut their teeth on. 

They are the tasks you hand to a machine, which does them faster, more cheaply, and often, if we are honest, better than the nervous 23-year-old would have. From a purely operational view, this is wonderful. The work gets done. The senior people are freed up. Nobody has to wait three days for a draft that needs heavy correction anyway.

But there is a quiet paradox sitting underneath all of this, and it troubles me. The work we are most eager to automate is the same work that used to make people good. The bottom rung of the ladder is being sawn off, and we are still expecting people to somehow arrive at the top. 

We have built a world where a young professional can produce a polished output on their first day without ever having developed the understanding that the polish is supposed to signal. They can hand you something that looks like the work of someone with five years of judgement, while carrying none of it. And the danger is not that the output is bad. The danger is that it is good, and that nobody, including them, can see what is missing.

What is missing is the reps. The reason the old grunt work mattered was never the output it produced. It was what it did to the person producing it. Writing a hundred mediocre drafts is how you eventually develop the instinct for a good one. Sitting with a messy dataset until it makes sense is how you learn to smell when a number is wrong. 

You cannot shortcut your way to judgement, because judgement is not information. It is the residue of having struggled with the problem yourself, enough times, that you no longer have to think consciously about it. When the machine does the struggling for you, it keeps the residue. You get the clean answer and lose the education that was hidden inside producing it.

So what does the early-career professional actually do, if the path the rest of us walked is closing behind us? My honest answer is that you have to learn deliberately what we were able to learn by accident. When AI hands you a draft, the worst thing you can do is accept it and move on, because you will have produced something without becoming anyone. 

The better move is to treat the output as the start of the work rather than the end of it. Interrogate it. Ask why it made the choices it made. Try to write the thing yourself first and then compare. Redo the parts that feel too easy. Use the tool as something to argue with, not something to hide behind. The practice is still available to you, but for the first time in history you have to choose it on purpose, because nobody is forcing you to do the hard version anymore.

And this places a real responsibility on those of us who are further along, because the old way we trained people no longer works. We used to mentor simply by handing down the work, trusting that the doing would teach. We cannot do that now, because the doing is exactly what we have given to the machine. 

If we are not careful, we will preside over a generation that never got the experiences that made us, and we will wonder, a few years from now, why the bench feels so thin. The answer will be that we automated away their education and called it efficiency.

The work of mentoring now has to be more intentional than it has ever been. It means deliberately recreating the struggle we were lucky enough to have forced on us. Handing juniors the messy, ambiguous problem rather than the clean, well-defined task. Asking them to defend their thinking out loud, not just deliver the output. Slowing them down on purpose, sometimes, so they actually wrestle with something before the tool resolves it for them. 

None of this is efficient in the short term, and that is exactly the point. We have to choose to be a little less efficient with the next generation so that they have something to draw on when efficiency is no longer enough.

The ladder is not gone. But it is no longer automatic, and that changes everything. It used to be that simply showing up and doing the unwanted work was enough to grow. Now growth has to be built deliberately, from both sides, by the young professional choosing the hard version of the work and by the rest of us refusing to let the tool rob them of the very experiences that once made us worth listening to.