Skip to content
Sign in
← All notes

What happens to junior jobs when AI does the junior work?

The routine tasks being automated were not busywork. They were the mechanism by which people became senior, and nothing has replaced the mechanism.

Careers8 min read

Drafting, first-pass analysis, routine research, summarising, tidying data, writing the boring tests. These were the junior workload, they were genuinely inefficient, and they are the tasks AI does most convincingly.

Automating them looks like pure gain until you ask what they were for. They were not only output. They were the years in which somebody built the pattern library that later gets called judgement.

Expertise was a by-product of doing the unglamorous work

Nobody learns to spot a broken argument by being told what a broken argument looks like. They learn it by writing several hundred of them, having each one corrected, and gradually noticing the shape of the mistake before they make it.

The same is true of code review instincts, of knowing which customer complaint is a symptom of something bigger, of sensing that a plan is a week of work rather than a day. All of it came from volume plus correction, applied to work that was, individually, not very important.

Remove the volume and the correction still has nothing to act on. The junior produces a generated draft, it is broadly fine, it gets approved, and nobody learned anything. The output improved and the person did not.

Expertise built the old way: many small pieces of unimportant work, each one corrected, until the shape of the mistake becomes visible before it is made. The volume was the teaching mechanism.
The unglamorous work was not only output. It was the apprenticeship.

The trap this creates for organisations

The immediate economics are clear and they point one way: fewer juniors, more tooling, faster delivery. Every quarter this looks correct.

The bill arrives later, and somewhere else. Senior people leave, retire or move, and the bench that should have replaced them was never built, because the work that builds it was automated for a saving that felt obvious at the time. You cannot hire your way out of that quickly, because everyone else made the same decision in the same quarter.

This is a slow failure with no alarm attached, which is the kind organisations are worst at avoiding.

An obvious saving taken this quarter, and the gap it leaves years later: the senior bench that should have been growing underneath was never built, and cannot be hired quickly because everyone made the same decision at the same time.
The saving is immediate. The bill is not, and it arrives somewhere else.

What actually works for junior development now

The answer is not to ban the tools and make people do it the slow way for their own good. That is unpopular, hard to enforce, and wastes real efficiency. It is to be deliberate about the thing that used to be automatic.

  • Give juniors the review seat, not only the production seat. Critiquing generated work builds the same pattern library that producing it used to, and it is a better use of the time.
  • Ask for the reasoning, not the artefact. If somebody cannot explain why the output is right, they have not done the work regardless of what they submitted.
  • Create small decisions with real consequences and let people own them. Judgement is built from outcomes, so it needs outcomes.
  • Protect some slow work on purpose in the area somebody is meant to become expert in. Not all of it. Enough to build the instinct.

What this means if you are the junior

You are in a harder position than the generation before you and it is worth naming rather than pretending otherwise. The path they took, doing volume until it turned into instinct, has been partly removed and nobody has published a replacement.

The practical response is to seek out the parts that build judgement rather than the parts that finish tasks. Ask to review. Ask why decisions were made. Make predictions and check them. Do some work by hand deliberately. It is slower in the short term and it is the entire difference between three years of experience and one year repeated three times.

Common questions

Is AI eliminating entry-level jobs?
It is eliminating a lot of entry-level tasks, which is not the same thing but affects the same people. The routine drafting and analysis that justified a large junior intake is now cheap, so the commercial case for hiring at volume has weakened. The roles that remain tend to demand judgement earlier, which is a harder starting position.
How do juniors gain experience if AI does the routine work?
By moving from producing to evaluating, deliberately rather than eventually. Reviewing and critiquing generated work builds the same pattern recognition that producing it used to. Beyond that: own small decisions with real consequences, make predictions and check them, and do some work by hand on purpose in the area you want to be expert in.
What is the experience gap?
The gap that appears when organisations automate the tasks that used to develop people, then find some years later that nobody has the judgement required for senior roles. It does not show up in any quarterly metric, which is exactly why it accumulates.
Should companies still hire juniors in the AI era?
Yes, but the job has to be designed differently. Hiring juniors to produce output that AI produces faster is genuinely hard to justify. Hiring them into a role built around reviewing, questioning and owning small decisions builds the senior bench you will otherwise be trying to buy in a market where everybody wants the same people.

PDP Quest exists because of the problem underneath all of these: when output stops indicating capability, you need another way to know who can actually do the work.

See how verification works →