The Rung We Removed

I have a clear memory of my first real engineering task.

It was not glamorous. It was not strategically important. It was, by any honest measure, the kind of work that existed precisely because someone more senior did not want to do it. I was given a specific problem, a limited scope, and just enough context to get started. I made mistakes. I asked questions that probably seemed obvious to the people around me. I learned things that I could not have learned any other way, because they only exist in the gap between understanding something in theory and making it actually work.

I think about that period often when I read the current data on entry-level hiring in tech.

The fifteen largest technology companies cut entry-level hiring by 25% between 2023 and 2024, according to research from SignalFire. That trend has continued into 2026. LinkedIn data shows new software engineering job postings down another 15% in the first two months of the year compared to the same period in 2025. The employers’ assessment of the job market for new graduates, tracked by the National Association of Colleges and Employers, is now at its most pessimistic since 2020.

The explanation being offered, consistently and confidently, is that AI tools can now handle the work that junior engineers used to do. Code generation, documentation, testing, basic debugging: tasks that once served as the training ground for early careers are increasingly being automated or absorbed into the workflows of more senior engineers working with AI assistance.

The short-term logic is not wrong. If a senior engineer with AI tools can produce the output that previously required two or three junior team members, the math of headcount reduction is straightforward. Boards understand it. Investors reward it. The quarterly numbers support it.

What the quarterly numbers do not capture is what gets quietly dismantled in the process.

There is a tendency, in the current conversation, to describe entry-level positions primarily in terms of the tasks they contained. And from that perspective, automating those tasks looks like pure efficiency.

But junior roles were never really about the tasks. They were about the formation of judgment.

The work itself, the repetitive, constrained, supervised work, was the mechanism through which something harder to measure was being built. Pattern recognition. The ability to read a codebase and understand not just what it does but why it was built that way. The instinct for when something is about to break before it breaks. The social intelligence of a team: how decisions actually get made, how disagreements get resolved, how trust gets established between people who depend on each other’s work.

None of that is in the task description. All of it is in the experience of doing the work, making the mistakes, and having enough proximity to more experienced colleagues to absorb what cannot be written down.

IBM’s chief human resources officer made this point recently in terms that are worth sitting with: the company would triple its hiring of junior talent specifically because without that pipeline today, there would be no middle managers or senior engineers in ten years. The formation process cannot be skipped. It can only be deferred, and deferring it does not make it cheaper. It makes it a crisis that arrives later, at a moment when you are least prepared for it.

This is where the conversation usually stays technical, and where I think it misses the most important part.

The question of who enters a profession and how they are formed is not just a talent pipeline question. It is a leadership question. The senior engineers of 2035 are the junior engineers of today. The people who will eventually run technology organizations, lead transformation programs, make consequential decisions about how AI gets deployed and governed, are right now trying to find their first role in a market that has decided their starting point is no longer economically justified.

The IEEE’s analysis of the 2026 job market put it directly: with AI tools performing more of the foundational work, the expectation is that new graduates must enter at a higher level almost from day one. But proficiency at a higher level is built on the foundation of having worked at the lower one. Expecting people to slot in mid-level without having been junior is not raising the bar. It is removing the ladder and being surprised when fewer people reach the top.

There is also something worth naming about what this does to the diversity of the people who eventually make it through. The candidates best positioned to demonstrate mid-level competence from the start are those with the most resources, the best networks, the most access to hands-on experience outside of formal employment. Removing entry-level roles does not create a meritocracy. It creates a higher barrier that filters for privilege as much as ability.

Leave a comment

I am not arguing that AI tools should not change how engineering teams are structured. They will, and in many cases they already should. The productivity gains at the individual level are real, and organizations that ignore them will pay a different kind of cost.

What I am arguing is that efficiency at the team level and investment in the pipeline are not the same decision, and treating them as interchangeable is a mistake that compounds quietly over time. The cost of removing the bottom rung does not show up in this quarter’s numbers. It shows up when you need to promote someone into a senior role and realize the people who would normally be ready have not had the formation that makes them ready. It shows up when the institutional knowledge that used to be transmitted through proximity and mentorship has nowhere left to travel.

IBM understood this. Most of its peers are currently making the opposite bet.

Which of those positions will look correct in ten years is not a difficult question, even if it is an inconvenient one right now.