The Ladder Lost Its Bottom Rung

Every expert I have ever worked with was made the same way. They spent years doing work that was slightly beneath them, under someone who could tell when it was wrong.

The junior analyst who built the model nobody used. The associate who read the whole contract so the partner could read the two clauses that mattered. The engineer who was handed the bug nobody wanted, in a corner of the system nobody understood, and came out the other side understanding it. None of that work was glamorous. Most of it was, strictly speaking, replaceable. And all of it was the apprenticeship that produced the person who could later be trusted with the thing that was not replaceable.

That work is now the easiest work in the world to hand to a machine. Which means we are quietly removing the bottom rung of the ladder that every senior person in every field climbed, and we have not decided what replaces it.

The work was never the point

Here is the thing organizations get wrong about junior work. They price it by its output. The first draft, the research memo, the test coverage, the reconciliation. Measured that way, a system that produces the same output in seconds is an obvious upgrade, and the junior role looks like a cost that technology has finally made unnecessary.

But the output was never the main thing the junior work produced. The main thing it produced was the junior. Doing the reconciliation by hand for two years is how someone learns what a reconciliation is for, which numbers lie, and what a wrong one feels like before the spreadsheet confirms it. The memo was a byproduct. The judgment was the product, and it was built slowly, out of thousands of small, low-stakes acts of being wrong in front of someone who knew better.

Remove the acts and you have not removed a cost. You have removed the factory.

What the machine actually is

I want to be precise about the tool, because the temptation is to make it either a villain or a savior, and it is neither.

Today's systems want nothing. They have no ambition to replace anyone. They do what they are asked, and what they are asked most often is the junior work, because that is the work that is easiest to describe. Nobody knows what the self-improving systems now being built will be able to do, and I am not going to pretend otherwise. But more capable is never more human. A system that drafts better than a first-year associate does not become an associate, and it does not become a partner either. It becomes a faster way to skip the years in which partners are made.

So the machine is not the problem. The problem is an economy that will, quite rationally, stop paying for apprenticeship the moment the apprentice's output can be had for free.

The cliff that nobody is charting

Think about the shape of this over twenty years, because that is the timescale on which it bites.

The senior people in every field today were trained the old way. They are fine. They are, in fact, more valuable than ever, because they can check the machine's work, and checking is where the value has moved. For the next decade, organizations will run on their judgment and the machine's output, and the numbers will look wonderful.

Then those people retire. And the cohort behind them will have spent a career supervising output they never learned to produce. They will know what a good memo looks like without ever having written a bad one. They will know the answer without the years of being wrong that make an answer trustworthy. Some of them will be brilliant anyway; brilliance finds a way. Most will be something new: fluent, fast, and hollow at exactly the point where the hard call gets made.

This is not a prediction about machines. It is arithmetic about people.

Apprenticeship has to become deliberate

For most of history, apprenticeship was free. It came bundled with the work, because the work had to be done and someone had to learn to do it. That bundle is coming apart, and the only response I can see is to start paying for the learning on purpose, as a line item, knowing it will look inefficient on every dashboard you own.

Some of what that means is concrete. Let junior people produce the first draft before they see the machine's, and compare. Keep at least some of the reconciliations manual, not because the machine cannot do them but because a person needs to have done them. Put the junior in the room for the hard call, and make them say what they would do before the senior speaks. Treat being wrong, cheaply and often, as a thing the organization has to manufacture now, because the world has stopped supplying it for free.

Some of it is cultural, and harder. A firm that bills by the hour has every incentive to let the machine do the hours. A startup that has to ship this quarter cannot easily justify a two-year investment in someone's judgment. The pressure runs one way, and it runs toward a future in which experts are a legacy asset, slowly depleting, that nobody is replacing.

The question worth asking now

I have spent a working life in industries that were repeatedly remade by tools, and the pattern is always the same. The tool arrives, the output gets cheap, and everyone looks at the output. Almost nobody looks at what the old way of producing it was quietly producing on the side.

This time, the side product was people who know what they are doing. It is worth asking, in your own organization, a simple question that no efficiency metric will ask for you. Ten years from now, who here will be able to tell when the machine is wrong? And what, exactly, are we doing today to make sure that person exists?