Nobody Warned These Workers—Now Their Whole Career Faith Is Gone
In this article
> **Bottom line:** A 2025 Stanford Digital Economy Lab study found employment for workers aged 22-25 in the most AI-exposed occupations — software development chief among them — has fallen roughly 13% relative to older workers in the same roles since late 2022, even as overall employment kept growing.
This isn't a hiring slowdown that reverses next quarter.
It's a structural break in how companies build engineering teams, and the junior developers caught in it are describing something closer to grief than career anxiety.
If you manage a team, mentor a bootcamp grad, or are one yourself, the math you were taught about paying your dues no longer works the way it used to.
I spent four hours on a video call in March with a 24-year-old named Priya who'd done everything right.
Computer science degree, two internships, a GitHub full of side projects, 340 applications sent between January and March 2026. Six interviews. Zero offers.
"I keep thinking I did something wrong," she told me. She hadn't.
The Setup: A Junior Engineer Pipeline That Stopped Working
I've spent the last eleven years building infrastructure at companies ranging from a Series B fintech startup to a 40,000-person enterprise.
Part of that job, at every stop, was interviewing and mentoring junior engineers. It's the part of the job I liked best.
Sometime around mid-2024, the funnel changed. Not the applicant pool — that stayed just as strong, arguably stronger, because more people were learning to code with AI tutors than ever before.
What changed was the demand side. Teams that used to budget for two juniors per senior started budgeting for zero.
I asked a VP of engineering I'd worked with at the fintech company why. Her answer stuck with me: "Claude and Cursor do the work I used to hire a junior to do.
Why would I take on the management overhead?" She wasn't being cruel. She was doing arithmetic that, from where she sits, is completely rational.
That arithmetic is now showing up in national data, not just anecdotes.
The Core Insight: The Career Ladder Lost Its Bottom Rungs
Here's what the Stanford research actually measured, and it's more specific than the doom-scroll version you've probably seen on X.
Economists Erik Brynjolfsson, Bharat Chandar, and Ruyu Chen analyzed payroll data across millions of workers and found the employment decline wasn't evenly spread across tech.
It was concentrated in early-career workers doing the *most automatable* parts of a job — the exact tasks that models like GPT-4 and Claude 3.5 started excelling at in 2024, a trend that ChatGPT 5 and Claude 4.6 have since pushed to an even scarier level: writing boilerplate, debugging routine errors, scaffolding CRUD endpoints, writing first-draft unit tests.
Meanwhile, employment for workers over 30 in those same job titles kept climbing. **The jobs didn't disappear. The entry point did.**
That distinction matters more than the headline number, because it changes what the crisis actually is.
This isn't "AI replaced programmers." It's "AI replaced the specific, repetitive work that used to be how programmers became good enough to stop needing supervision." Priya isn't competing against a chatbot.
She's competing against the fact that the tasks companies used to assign to someone like her to build judgment are now assigned to a model instead.
What This Looks Like Inside a Team
I watched this happen firsthand on a platform team I advised in late 2025.
The senior engineers loved their new Cursor-and-Claude workflow — genuinely, no hype needed, their velocity on infrastructure tickets roughly doubled.
But the team lead told me something I didn't expect: "I used to hand our new hire the annoying migration scripts to learn the codebase.
Now I just have Claude do the migration and I never get to see if the new hire can think."
The learning-by-doing loop that turned confused 23-year-olds into competent engineers ran on tasks that were tedious enough to delegate but instructive enough to teach.
AI ate exactly that category of task first, because it's the easiest category to automate. It left the hard judgment calls at the top and the entry-level tasks at zero, with nothing connecting them.
It's Not Just Software
I want to be careful here, because Hacker News readers will (correctly) push back if I make this sound like a tech-only story. It isn't.
The same Stanford data shows early signs in customer service, accounting, and paralegal work — anywhere a junior person's core value proposition was "does the routine work fast so the senior person can focus on judgment calls."
Law firms are already restructuring how they staff first-year associates around AI-assisted document review.
Junior accountants are watching AI tools handle reconciliation work that used to be their entire first two years.
The pattern is the same everywhere it appears: **the bottom of the ladder gets automated before the top does, because the bottom is where the work is most repeatable.**
That's the part nobody warned this cohort about. Career advice for the last thirty years assumed the ladder would still be there when you showed up. Nobody updated the advice.
The Reality Check: Where the Panic Overshoots
I don't think this means junior developers are obsolete, and I want to push back on the doomers as hard as I push back on the "just learn to prompt" crowd, because both are lying to you in different directions.
First: companies that stop hiring juniors entirely are making a decision that will bite them in five years.
Someone has to become the senior engineer who can catch what Claude 4.6 gets subtly wrong in a distributed systems edge case.
You don't get that person by skipping the step where they learn to be wrong in smaller, cheaper ways first.
A few engineering leaders I've talked to already see this — one told me flatly, "we're eating our seed corn."
Second: the juniors who *are* getting hired right now look different than the ones who got hired in 2019. They're not being hired to write code nobody wants to write.
They're being hired because they've already demonstrated they can direct an AI tool, catch its mistakes, and understand systems well enough to know when the output is subtly wrong.
That's a real, valuable, learnable skill — it's just not the skill bootcamp curricula were built to teach as recently as 2023.
Third, and this is the part that actually worries me more than the hiring numbers: the psychological toll is real and it's not something a market correction fixes.
Priya isn't losing sleep over unemployment insurance math.
She's losing sleep over a story she was told — study hard, build projects, pay your dues, the ladder is there — that turned out not to be true for her specific timing in history.
That's a different kind of damage than a bad job market. It's the loss of a framework for how effort translates into outcome, and you don't rebuild that with a stronger resume.
The Practical Takeaway: What to Actually Do With This
If you're early-career right now, stop optimizing for the 2019 version of "getting good." The skill that's scarce isn't writing code — it's the judgment to know when AI-generated code is wrong, and you build that judgment the same way you always did: by reading a lot of other people's code, breaking things in low-stakes environments, and asking senior engineers *why* something works, not just whether it does.
Use Claude or Cursor to get past the boilerplate faster, then spend the time you saved on the part that used to take a decade — understanding failure modes.
If you manage a team, the VP of engineering I mentioned earlier was right about the short-term math and wrong about the long game.
**Budget for junior hires the way you'd budget for R&D, not headcount efficiency.** Give them the tasks that teach judgment even when a model could technically do them faster, because you're not optimizing this sprint — you're building the senior engineer you'll need in 2031.
And if you're the one 340 applications deep with zero offers: the ladder didn't disappear, it moved. It's harder to see, it starts in a different place, and nobody sent out a memo.
That's not the same as it not existing.
Have you watched someone junior on your team lose faith in the path you took to get where you are — and did you know what to tell them?
I'd genuinely like to know what's actually working, because most of us are improvising this one in real time.
---
