AI took more than junior developer jobs and the bill comes later

A ticket comes in for a small bug fix. Hand it to the junior on your team and you wait a day, review something that half works, and sit down to explain what went wrong. Describe it to Claude and the patch merges before lunch.

The second path wins on every number your team reports this quarter. It also removes the only training that junior was going to get, and it removes it at every company making the same call on the same morning.

Researchers at Seoul National University interviewed 14 software engineers in South Korea about that decision, half of them seniors deciding where the work goes and the rest students about to enter a job market that has stopped hiring them. Korea makes a useful test case. Work-related AI usage there reaches 51.8 percent, close to double the U.S. rate, and major technology firms have shut down open recruitment of junior developers.

One founder with more than twelve years as a developer described what happened after he cut his junior headcount.

“We let a lot of junior engineers go in the second half of last year. But there was no impact whatsoever.”

Asked what a junior could contribute that the tools could not, he answered: “What can a junior engineer do better than a Claude subscription? I don’t think there’s anything.”

The same founder, earlier in the same interview, had walked through his own early career: “When I was an undergrad, I started as a part-time developer at [Large Enterprise A], and the first thing I did was what we call the menial work of development. Writing code as told, debugging every day, checking and testing whatever came back… I did a lot of so-called development grunt work. […] that’s how I think I got to where I am now.”

The thing being cut is the feedback

Software has eaten its own bottom rung before. Assembly gave way to C, memory management got handled for you, and the ladder moved up a level each time. Something different is happening here.

Minor bug fixes, routine implementation, and debugging never got assigned to juniors because they were valuable. They got assigned because doing them badly, and then having a senior explain the badness, is the mechanism that produces judgment. That mechanism is what the tools absorbed.

A senior at a mid-sized game company put the operational consequence in one line: “The cost of code generation is so cheap by using GenAI, but the cost of code verification is so expensive.”

Verification is the scarce input now. It is produced almost entirely by having been wrong in ways that cost something, and the pipeline producing it runs through work nobody assigns anymore.

Juniors can name what they lost

The students in the study describe the gap in the same terms their seniors do. One said he does not know what not to do, having seen only the good examples. Another said: “I don’t know what I don’t know.”

A senior at a mid-sized re-commerce company arrived at the same place from the hiring side: “The reason someone is called a senior is that they know how work should not be done, but now juniors can’t learn from the bad examples.”

Universities stopped enforcing the struggle

The classroom half of this runs on grading curves. One student described giving up on doing assignments unaided: “Everyone except me was using GPT, so everyone except me was getting nearly perfect scores. I no longer had the freedom to work through assignments on my own, making mistakes as I went, so in the end, I had no choice but to use it too.”

The same student took two courses in the same field, one where the assignments could be completed with a chatbot and one where they could not: “My grades came out the same. I definitely studied much more in the machine learning course. The deep learning course just felt like I got better at using GPT.”

The transcript looks identical either way. That is the part that should worry anyone reading resumes.

Korean IT job postings dropped 43 percent between 2023 and 2024, and the graduates competing for what remains are the ones this system produced.

In the United States, programmer employment declined 27.5 percent over roughly the same stretch.

Two industries already wrote the rules

Airlines hit this problem when cockpit automation matured. FAA Safety Alert for Operators 13002 tells carriers to preserve opportunities for manual flying so pilots retain the skills the autopilot performs for them. Nuclear plant operators face a parallel requirement under 10 CFR 55.59, which mandates periodic simulator training on emergency response that routine automated operation never exercises.

Both rules exist because a regulator decided skill maintenance was too important to leave to whoever was running the schedule that week. Software has no equivalent and has not asked whether it wants one.

One team’s answer costs almost nothing

The same re-commerce engineer described how her team onboards newcomers, starting with a task engineered to be trivial: “Finding something like a missing period in one of our production projects, putting the period in, committing it, opening a pull request, getting reviewed, and pushing it all the way to production”

Later assignments come with less scaffolding until the newcomer is asked to find a problem and solve it without being told how. She calls the design “a kind of learning by doing.”

The study found one senior, at a large enterprise with formal onboarding pipelines, who reported that junior work had changed very little. Absorption ran furthest where nobody had written down how juniors were supposed to develop.