The junior developer path just got much harder

Three years ago I wrote a guide for people trying to enter IT. The core of that guide was to identify the right field you want to enter, and then learn the skills that field actually needs. It made sense at the time. It's how I did it, and it's how most of the people I work with did it.

I'd write a very different guide today. The path I described is narrower now, in some places closed, and I don't think the industry is being honest about why.

What actually changed

AI came into play, and junior roles got erased.

The numbers back it up. Between 2023 and 2025, US employment for "programmers" dropped by 27.5%. In the same period, employment for "software developers" — a category that skews more towards design and judgment work — fell by just 0.3%. That's from BLS data, reported by IEEE Spectrum. Entry-level software postings overall dropped roughly 60% between 2022 and 2024. And 84% of developers now use AI tools in their day-to-day work, up 14 points in two years.

What the split tells you is that the category that collapsed wasn't "writing software". It was writing software at the level juniors used to be hired to write. The straightforward tickets, the well-described bugs, the boilerplate that used to fill a junior's first year. That work hasn't disappeared. It's just being done by a senior dev with an AI assistant in a fraction of the time.

The AI paradox for juniors

AI is, at the same time, the best thing and the worst thing that ever happened to people trying to enter this field.

The best thing is obvious. Any concept explained on demand, infinite patience, a tutor available at 2am when you're stuck. When I was starting out I'd burn hours on Stack Overflow trying to understand why something didn't work. A junior today can get a personalized explanation in thirty seconds. That's genuinely a gift.

The worst thing is more subtle, and I don't think it gets talked about enough.

AI is bad at knowing what a specific person needs. It answers the question you ask, at whatever complexity that question implies, without noticing whether the answer fits your actual situation.

A recent example from my own work. I was designing a simple email sender service for one of my side projects. I asked AI for an implementation plan, and it came back with an enterprise-grade solution: a cron job running every five minutes, a proper queue, complex rate limiting, retry strategies. All technically correct. All completely wrong for what I needed, which was a cron job once a day and no rate limiting at all, because I have a handful of users.

I could spot that because I have years of experience to recognise when a solution is oversized. A junior can't. They will read that response, nod, and start building an enterprise email pipeline for three users. And the AI will never tell them "forget all of that, just send a few emails from a script once a day."

And then there's the bigger problem. AI ate the exact tasks juniors used to learn on. The simple CRUD endpoints. The boilerplate components. The well-described ticket a team lead gave you because they knew you could handle it and learn something in the process. That work was never really about the output. It was about the reps. Hundreds of small decisions that, over months, build instinct.

AI removed the easy tasks and left the hard ones. For a senior dev that's a productivity gift. For a junior it means there's no on-ramp left.

The gap you can't AI your way across

I mentored a junior developer recently. Smart person, motivated, good fundamentals from school. And I realised something that I think is the most important thing I want to say in this post:

The gap between someone who has done professional software development for years and someone who hasn't is huge. And it is not a gap AI can close.

It's not a gap in syntax or concepts, those things AI actually helps with. It's a gap in judgment. Knowing when a problem is worth solving. Knowing when a ticket is ambiguous and you need to stop coding and go find the PM. Knowing when a library is going to bite you in eighteen months. Knowing how much to refactor and how much to leave alone. Knowing what "good enough" looks like on this team, on this codebase, for this feature.

You don't read that in documentation. You don't prompt your way to it. You get it by shipping things, watching them break, fixing them, watching the fix break something else, and slowly building an internal model of how real software behaves under real conditions.

Experience can't be filled with quick learning. You have to gain it. There are no shortcuts.

And here's the uncomfortable turn. This is exactly the skill AI can't transfer, because it's the skill of judging AI output. AI is a multiplier on judgment. A senior dev with AI is formidable, because there's real judgment there to multiply. A junior with AI is fragile, because multiplying no judgment leads nowhere. They can produce code, absolutely, but they don't yet have the instinct to notice when the code is wrong in ways that matter.

The five-year question

So what happens in five years?

I don't think we'll have a shortage of people who can produce code. AI will keep getting better at that. What we'll have is a shortage of people with real judgment. People who can hold a product vision in their head, translate it into work AI can do, and tell whether the output meets the goal. Sometimes even figuring out what the goal actually is, because the person asking doesn't fully know.

That's a senior skill. It gets built through years of making small decisions on real systems with real consequences. And we're currently not training anyone to become senior, because we stopped hiring the people who would have become senior.

I don't know how this resolves. Maybe the industry course-corrects and builds some AI-native version of apprenticeship. Maybe we end up with a bifurcated field: a small number of very well paid senior engineers, and a long tail of people who couldn't climb into that tier because the ladder was pulled up behind the current generation.

I'm not optimistic about option one happening fast. Companies optimise for the next quarter, not for where their talent pipeline will be in 2030.

If you're trying to enter the field now

I'll write a separate post about what I specifically look for when I'm hiring a junior in 2026 — from the other side of the table, as a tech lead who still does hire them, when the situation allows. For now, three things that weren't true when I wrote my 2022 guide and are true now.

The first-job bar is higher. The path I described — get hired, learn on the job — still exists, but at far fewer companies, and they're harder to identify from the outside.

AI is a learning amplifier, but it is not a substitute for shipping. A junior who uses AI to understand concepts while building real things will grow. A junior who uses AI to produce things they don't understand will plateau fast and not realise it.

And the places that still properly mentor juniors — the ones that give you real tickets, real code review, real senior engineers who'll spend time on you — are now the rarest and most valuable places to work. If you find one, the salary is almost secondary.

More on all of that in the next article.