We Stopped Making Seniors.

Last month, a student handed in the best assignment in the class.

Clean structure. Sharp analysis. A market breakdown I'd have been proud to put in front of a client ten years ago.

In office hours, I asked him one question. "Walk me through why you ranked the second option above the third."

He couldn't.

His mind had simply never made that decision. The model made it. He read the output, decided it sounded right, and handed it in. He couldn't tell me why option two beat option three, because he never really thought about it. His mind was never in the loop when the choice got made.

He wasn't lazy. He was the top of the class.

That's the part that kept me up.

 

Everyone is arguing about whether AI will take entry-level jobs.

Wrong argument. There's no clear evidence this is happening everywhere, right now. And even where it is, it's unclear whether it's structural or just a transition phase as companies relearn that jobs aren't just bundles of tasks.

But the problem isn't only "AI taking the junior's tasks."

The real problem is that AI is dismantling the only path we've ever had for turning a junior into a senior.

 

The Vanishing Bottom Rung

Let me give you some numbers.

The Stanford Digital Economy Lab tracked workers aged 22 to 25 in the jobs most exposed to AI. Their employment fell 13% between 2022 and 2026. It wasn't layoffs. The people already in those seats mostly kept them. The drop came almost entirely from companies that quietly stopped hiring new juniors.

The World Economic Forum has a name for it now. The vanishing entry-level.

Sit with the mechanism, because the mechanism is the whole story. Nobody is firing the young. Companies are just deciding, one budget meeting at a time, that they don't need the next batch. Why pay a 23-year-old to write the first draft, build the model, clean the data, and make the deck, when a machine and a senior can have 10x the output? No salary. No onboarding. No learning curve. No management time.

On paper, obvious. Cut the cost, keep the output.

Except the output was never the reason you hired a junior.

The Grunt Work Was the Gym

Let me remind you of simple facts every serious profession figured out centuries ago and is now forgetting in about thirty-six months.

The bottom rung of a career was never about the work. It was about the reps.

The junior lawyer who reads four hundred boring contracts isn't there because the firm needs four hundred contracts read. She's there because somewhere around contract two hundred, she starts to feel the one clause that's wrong before she can explain why. That feeling has a name. We call it judgment, and you cannot download it.

The junior analyst who builds the same model fifteen times isn't producing fifteen models. He's wiring the intuition that lets him glance at the sixteenth and say "that number is wrong" without redoing the math.

This is the apprenticeship. It is older than the corporation.

The medieval guilds had it: apprentice, journeyman, master. Medicine has it: the residency, years of unglamorous work because that's how the pattern recognition gets burned in. Aviation has it: thousands of hours in the right-hand seat before you are ever allowed to touch the left.

Every craft that matters built a ladder. From the outside, the first rung looks like grunt work.

But it was never "grunt work." It was the gym.

AI just handed everyone a forklift and told them to skip leg day.

You get the output today. You get the fragility in five to ten years.

 

Bainbridge, Again

I wrote about Lisanne Bainbridge a while back (Letter 73). In 1983 she named the central irony of automation: the more you automate a system, the less prepared the human is to step in when the automation fails, which is precisely the moment you need them most.

She was writing about cockpits. It scales all the way up.

Automate the junior's work and you don't just lose the junior's output. You lose the senior that junior was going to become. Then, on the day the machine produces something confident, polished, and catastrophically wrong, you reach for the person who's supposed to catch it.

And the bench is empty.

Because five years ago, you stopped building it to make the quarter look good. To "show" you were on the AI train. Congratulations.

 

The Uncomfortable Truth

AI is the most powerful teacher humanity has ever built. A patient, infinitely available tutor that will explain anything, at any level, at 2 am, without making you feel stupid. We could be raising the sharpest generation of professionals in history.

Instead, most people are using it to guarantee no learning ever has to happen at all.

Same tool. Opposite outcomes. AI is an amplifier, not an equalizer. Point it at someone building judgment, and it accelerates them. Point it at someone avoiding the work that builds judgment, and it beautifully accelerates the avoidance at scale, with perfect formatting.

The student in my office wasn't failed by AI. AI gave him a brilliant draft.

He was failed by a system, his teachers, future employers, and, honestly, me, that never made one thing clear:

the draft was supposed to be the start of his thinking, not the end of it.

Sometimes AI should be a brainstorming partner. Sometimes it should be a tutor.

But when it's time to decide what you believe, what you want to say, and what you stand behind, your thinking has to go first.

Almost nobody is teaching that (except cohorte.co :))

 

The Loan Nobody Knows They're Taking

The companies pulling the ladder up behind them are making a bet they don't even realize they're making.

The CFO who cuts junior hires this year "saves," say, four hundred thousand euros, and books a clean win. The quarter looks great on paper. The board nods.

(Never mind that many companies don't actually know how to turn AI into real value yet. For now, let's assume the savings are real.)

That same CFO, in 2030, will pay a fortune for senior people who can judge what the machine produces, own the outcome, and put their name on it.

Those seniors will be rare, because we spent five years not training any. Scarcity sets the price.

The "four hundred thousand saved" comes back as a multiple: paid to contractors, paid to the last person who still understands how the system works.

You didn't cut a cost. You took out a loan against your future, and the interest rate is brutal.

 

The New Apprenticeship

So what do you actually do? Three answers, depending on where you sit.

If you manage juniors: protect the friction.

Your instinct will be to hand them the AI and celebrate how fast they ship. Resist it. The struggle you're itching to optimize away is the exact thing that builds the person. Make them do it the slow way once before they earn the fast way. Not as punishment. As training. A junior who has built the thing by hand even a single time reads the machine's version completely differently. They know where the bodies are buried, because they once buried one.

If you are a junior: AI is your teacher first, your assistant second, your amplifier never. Not yet.

The most dangerous move available to you right now is using AI to look senior before you are. It works. The output is convincing. And it quietly guarantees you plateau the moment the problems get hard enough that judgment is the only thing that saves you. Use the tool to understand, relentlessly. Every time it hands you an answer, make it show its work, then argue with it. The goal is never the finished deliverable. The goal is the coordinates in your own head. Your expertise is GPS coordinates, and you only earn them by walking the territory yourself at least once.

If you hire: bring the apprentice back, and give the job its real name.

The entry-level role didn't die. It changed shape. The junior's old job was production. Production is now cheap. The new entry-level job is verification: checking the machine's work, catching the plausible-but-wrong, learning what "good" looks like by inspecting a thousand examples of almost-good. Put "AI Verification Analyst" on the org chart if you need a title. It is the same ancient function: the apprentice learns the craft by examining the master's work. Except now the master is a machine that's right most of the time and confidently wrong just often enough to end a career.

That might be the most important job of the next decade. And almost nobody is hiring for it, because they're too busy congratulating themselves on the headcount they cut.

Back to the Student

I emailed him after office hours. I told him the draft was genuinely excellent, and then I asked him to do something strange with the next one. Write it himself first. Badly. Struggle with it. Get it wrong. Then, and only then, open the AI and compare.

He pushed back. "Isn't that slower?"

Yes. That's the point.

He wrote back two weeks later. The second assignment was worse on paper. Messier. Thinner. But when I asked him to walk me through his choices, he talked for twenty minutes without stopping. He was in the room this time. He had made the decisions. He could defend every one of them.

That is a senior being born. Slowly, awkwardly, the only way it has ever happened.

We are about to find out what happens to a profession, an industry, a whole economy, that decided seniors were too expensive to make.

I would rather we figure it out before the bench is empty.

Build the ladder. Climb it the slow way. And make sure there's someone on the rung behind you.

AI is only as good as the human operating it. Right now, we're quietly choosing not to make the humans.

Have a great weekend.

— Charafeddine (CM)