You've Always Been Told to Learn to Code and the Job Would Follow. This Year, That Quietly Stopped Being True

I've had some version of this conversation three times in the last month. A neighbor whose son just finished a computer science degree. A friend asking whether her daughter should still bother with the coding bootcamp she'd already paid for. Someone in a WhatsApp group asking, half-joking, "is it too late to learn to code?"

For as long as most of us have been giving career advice, "learn to code" has been the safe answer. Stable pay, steady demand, a clear ladder: junior, then mid-level, then senior. I've given that advice myself, more than once, without thinking too hard about it. So I finally sat down and looked at what's actually happening to that first rung of the ladder. It's not what I expected, and it's not as simple as the scary headlines make it sound either.

#What the numbers actually say

Stanford's Digital Economy Lab, in its 2026 AI Index, found employment for developers aged 22 to 25 down about 20% since 2022. Separate hiring data from SignalFire shows the top tech companies cut entry-level hiring by roughly a quarter over the same stretch. Put those together with recent postings data and you get the number that's been going around: entry-level developer job postings down somewhere around two-thirds from where they were a few years ago. New graduates now make up about 7% of hiring at the big tech firms, down sharply from before. Unemployment for computer science graduates is sitting around 6%, higher than you'd expect for a degree that used to be treated as a golden ticket.

The reason isn't mysterious once you say it plainly. A lot of what a junior developer used to do, writing routine, repetitive code from a clear spec, is exactly the kind of work today's AI coding tools are good at. One experienced developer with an AI assistant can now get through the kind of workload that used to need that senior developer plus a junior one learning the ropes beside them. From a spreadsheet, the junior position looks optional in a way it didn't three years ago.

#The part that isn't simple

Here's where the story gets more interesting than "AI is taking the jobs," and worth sitting with instead of scrolling past.

Overall software jobs haven't disappeared. They actually grew, by around 11% over the same period, and senior developers are about as employed as they've ever been. Roles tied to AI infrastructure and security are growing fast, up 70 to 90% in some measures. What's happening isn't a shrinking field. It's a field splitting into two.

One half is the routine work, plain data entry screens, standard company websites, the kind of code that follows a well-worn pattern. That half is genuinely shrinking, because it's the easiest thing for AI to do well. The other half, work that needs judgment, that touches money or health records or security, work where a mistake is expensive and someone has to be accountable for it, is still hiring, sometimes aggressively. A few enterprise software companies, banks, and healthcare tech firms have actually grown their junior intake this year, just for a different kind of junior: one who reviews what the AI produces instead of typing it all out line by line.

#Why I think this matters beyond the headline

You don't have to be a developer, or the parent of one, for this to be worth your attention. The pattern underneath it shows up everywhere technology touches entry-level work, not just software.

The safest advice used to be "get good at the tool." The better advice now is "get good at judging the tool's output." Anyone who's early in a technical career, or advising someone who is, is really choosing between two paths that used to look identical: chasing the kind of skill that's about to be commodity, or building the kind of judgment that stays scarce even after the tool gets very good. That's not a software story. It shows up in accounting, in design, in customer support, in half the office jobs I know. Software just got there first because it's where the tools improved fastest.

#The actual aspiration

If you're early in a technical path, or helping someone who is, the habit worth building isn't "learn to code faster than the AI." That race is already lost, and it was never really the point.

Learn to catch what the AI gets wrong. That's the skill everyone above is quietly paying for right now: someone who can look at a chunk of AI-written code, or an AI-written report, or an AI-drawn plan, and say specifically where it's fooling itself. You build that by reading real code, not just tutorials, and by asking "why did it do it this way" more often than "did it work."

Go narrow before you go wide. A general "I can build a website" skill is common and getting cheaper by the month. "I understand how a bank's payment system has to behave when something goes wrong" is rare and stays valuable, because it takes real judgment, not just typing speed. Pick one area, even a small one, and go deeper into it than feels efficient.

And build something real, once, rather than another tutorial project that lives and dies in a folder. A small tool that an actual person uses, even one person, teaches you the parts AI can't hand you: what happens when the data is messy, when the user does something you didn't expect, when it breaks at 11pm and someone's waiting on you to fix it. I wrote about a related idea a while back, that curiosity, not raw coding speed, is the more durable skill of this decade, and this junior hiring shift is exactly why: the people getting hired now aren't the fastest typists, they're the ones who stayed curious enough to ask good questions about what they were building.

The door into tech isn't closed. It's just a narrower door than it used to be, and it opens for a slightly different kind of person than it did five years ago. Worth knowing which kind, before you, or someone you're advising, spends a year preparing for the wrong one.