For most of the history of our industry, there was a simple deal. A company hired young people who knew little, gave them the small, boring work, and in return they learned. Five or ten years later, some of them were the senior engineers, architects and leaders the company could not run without.
That deal is breaking, and almost nobody decided to break it.
AI tools now write the small functions, fix the simple bugs, draft the tests and summarise the documents. That is exactly the work we used to give to beginners. So the logic looks obvious: why hire a junior when a senior with a good assistant can do the same work faster?
Each company that makes this choice is acting sensibly. Together, they are removing the first step of a ladder that the whole industry climbs.
The Work Was Never Just Work
The small tasks we gave to juniors were never only about getting the task done. They were how people learned.
Fixing a simple bug teaches you how the system really behaves. Writing a boring report teaches you what the business cares about. Having your code sent back three times by a patient reviewer teaches you what "good" means in a way no course can.
Seniority is not knowledge you can download. It is judgment, built slowly from thousands of small decisions, many of them wrong, made under the eye of someone more experienced. Take away the small decisions and you take away the way judgment is built.
This is not only a software problem. The same pattern is showing up in law, accounting, design, journalism and support. Anywhere the entry-level work can be automated, the entry-level job is at risk, and with it the training that job used to provide.
A Problem No Single Company Will Solve
Here is why this needs rethinking, not just a policy update.
Every company wants experienced people. Very few want to pay to create them. In the past, that tension was hidden, because junior people did enough useful work to pay for their own training. AI removes that cover. Training a junior now looks like a pure cost, and the benefit arrives years later, often at another company.
So each organisation waits for someone else to do it. That is a classic shared problem: what is rational for each company is harmful for all of them. In ten years, the senior people we have today will retire or move on, and the people who should replace them will not have had the years of practice that made them senior.
The result will not be a dramatic crisis. It will be slower and harder to see: systems that nobody fully understands, AI-generated code that nobody can properly review, and teams that are fast until something unusual happens.
What Gets Lost Between the Generations
It helps to be specific about what disappears when the junior path closes.
Context The reasons behind old decisions live in people's heads. Juniors used to absorb them by sitting close to the people who made them. Without that contact, the reasons leave when the people leave.
Review skill AI makes writing code cheap and reviewing it essential. But the ability to review comes from years of writing and being reviewed. A generation that skips the writing may never learn to judge it.
Healthy doubt Experienced people know when an answer looks right but is wrong. That instinct comes from having been wrong many times. People who learn only by asking a model learn to trust confident answers.
Leadership Today's engineering managers were yesterday's junior engineers. A company that stops growing juniors will one day have to buy all its leaders from outside, if they can be found at all.
Rethinking the Apprenticeship
I do not think the answer is to refuse AI or to hire juniors out of charity. The answer is to redesign how people learn, so that learning survives in a world where the old learning tasks are automated.
In practice, that means a few deliberate choices:
- Keep hiring at the entry level, even in small numbers, and treat it as an investment in future leadership, not as cheap capacity.
- Give juniors the review work, not only the writing work. Let them check AI output against the system, explain what is wrong, and defend their view.
- Pair every junior with a named senior who is measured, in part, on how that person grows.
- Make juniors explain AI-generated changes in their own words before they are merged. If they cannot explain it, they have not learned it.
- Rotate juniors through the unusual work: incidents, migrations, customer escalations. That is where judgment grows fastest.
- Write down the reasons behind decisions, so that context does not depend on who happens to sit near whom.
None of this is expensive compared to what it protects. It does require leaders to accept that a junior who learns slowly today is worth more than a model that answers instantly, because in ten years only one of them will be able to lead.
The Question for Leaders
Every leadership team using AI should be able to answer one question: who will be our senior people in ten years, and how are they learning today?
If the honest answer is "we will hire them from somewhere," it is worth asking where that somewhere is, and who is training them. If everyone gives the same answer, nobody is.
AI can do the junior work. It cannot become the senior. Someone still has to.
