Developers aged 22 to 25 are down 19% from their late-2022 peak. Entry-level software postings have dropped 28%. Computer science graduates now face a 6.1% unemployment rate — higher than liberal arts majors. And the worst of it didn’t hit when ChatGPT launched. It hit in 2024 and 2025, when AI coding assistants stopped autocompleting lines and started completing tickets.
I read Laurie Voss’s breakdown on Seldo.com yesterday, and honestly? It rattled me. Not because the data was surprising — I’ve felt this shift in my own hiring pipeline — but because he connected dots I’d only sensed individually. The numbers are brutal, but the real story is what happens next. And that’s where most of the hot takes get it wrong.
The Numbers Nobody Wants to Talk About
Let me just lay out the data so we’re all looking at the same picture. Voss pulled from Stanford’s Digital Economy Lab, the BLS, Indeed, GitHub’s Octoverse, and Appfigures. The convergence is hard to ignore.
The headline is simple: the entry-level software engineering job is disappearing. Not evolving — disappearing. Stanford found that companies adopting generative AI cut junior developer hiring by 9-10% within six quarters. The BLS data shows “Computer Programmers” declined 16% year-over-year through May 2025, and “Web Developers” dropped 11%. Those aren’t categories you recover from in a quarter or two.
But here’s the paradox the hot-take articles miss: total developer employment is actually up 4.4% since October 2022. Seniors and mid-level engineers are doing fine. The 41-to-49-year-old cohort grew 14%. The bloodbath is concentrated in one specific demographic — young people trying to get that first foot on the ladder.
And that ladder? It’s broken.
The Old On-Ramp Doesn’t Exist Anymore
This is the part that hits closest to home for me as someone who both writes code and manages a team. The traditional path into software engineering was straightforward: you got hired to write mediocre code, a senior engineer reviewed it, you slowly absorbed judgment and context, and a decade later you were the senior engineer reviewing someone else’s code.
That chain is now severed. AI writes the mediocre code — faster, cheaper, and with zero feedback loops. So companies stop hiring the junior developer who would write that code, which means nobody is in the queue to become the senior who reviews things in five years. The math is brutally simple.
I’ve seen this play out in my own corner of the industry. When we post a junior developer position now, the quality and volume of applicants is different than it was in 2022 — not because the candidates are worse, but because there are so many more of them chasing far fewer slots. Meanwhile, I’m using AI coding tools daily to handle the boilerplate tasks I would have handed to a junior a few years ago. It’s efficient. It’s also quietly dismantling our talent pipeline.
The Long Tail Nobody’s Counting
Now for the twist. The reason Voss’s analysis stands out is that he doesn’t stop at the bad news. He also shows that the “long tail” of new developers he predicted did materialize — it just doesn’t call itself programmers anymore.
GitHub added a record 36 million new accounts in the last Octoverse year — the fastest growth ever. 80% of them used Copilot within their first week. The iOS App Store, which had been flat for eight years, grew 24% in 2025 and a staggering 80% year-over-year in Q1 2026. The category mix shifted toward productivity and utility apps — first-timers solving their own problems rather than studios chasing revenue.
Vercel reports 63% of its users are non-developers. Lovable hits 100,000 projects per day with 60% non-developer users. Replit has 50 million users. The ability to make software is spreading through every job title, just like typing and spreadsheet skills spread through the 90s.
That’s genuinely good news for the world. More people building things means more problems get solved. But it’s cold comfort if you’re a CS grad who can’t land that first job.
The Fork in the Road
Voss highlights two companies representing opposite futures for the industry, and I think every developer should pay attention to which one wins.
IBM is tripling entry-level hiring. Their theory? AI-equipped juniors can do senior-level work if you redesign the role around customer contact and specification — using AI for the typing, human judgment for the strategy. They’re betting that junior engineers who grow up with AI tools will develop better judgment faster, not slower.
Salesforce hired zero engineers last fiscal year. Zero. Their bet is that AI eliminates the need for entry-level technical roles entirely.
Which one wins determines whether this profession has senior developers in 2036. If every company follows Salesforce, the senior shortage becomes a crisis. If enough follow IBM, we rebuild the ladder — but it looks different than the old one.
What This Means for Filipino Developers
I think about this from a Philippine context because that’s where I build things and work with people. The BPO and offshore development industry here has historically been a massive employer of junior engineers. Western companies outsourced their entry-level coding — maintenance, QA, CRUD apps — to Filipino developers who then grew into senior roles.
If AI eats those entry-level tasks first, what happens to the pipeline here? The Philippines produces thousands of CS graduates every year. The university curriculum is catching up to industry needs, but it’s not keeping pace with how fast AI is reshaping those needs.
On the flip side, the “programming as a capability” shift could actually benefit developers in emerging markets — more people building software means more demand for the kind of judgment, context, and local knowledge that AI doesn’t have. The Filipino developer who understands both the code and the business domain becomes more valuable, not less.
But that only works if we actually invest in building that judgment. Which brings me to the real question: who’s going to mentor the next generation?
Rebuilding the Ladder
There’s a pattern I’m seeing in the teams that handle this transition well, and I wrote about it in my guide to reviewing AI-generated code like a senior engineer. The teams that succeed aren’t the ones that replace juniors with AI. They’re the ones that use AI to free up senior time for actual mentorship — code review, architecture decisions, and the kind of judgment transfer that happens in real conversations, not PR comments.
The DuneSlide wake-up call proved that AI-generated code comes with real security risks — 45% fails basic OWASP tests. We need humans who understand security, architecture, and the difference between code that compiles and code that’s correct. Those humans have to come from somewhere.
If you’re a junior developer feeling the squeeze — and I know several who are — my advice is genuinely different from what I would have said two years ago. Stop optimizing for “writing more code.” Start optimizing for understanding why code works the way it does. Review AI output critically. Learn to read and refactor, not just generate. Build judgment, not just output.
I explored this idea more in my piece on the AI hype cycle’s creaking joints — the tools are real, the capabilities are impressive, but the human layer of judgment is what separates useful software from dangerous software.
Where We Go From Here
Voss ends his piece with a line I keep coming back to: “We are not watching the death of programming. We are watching programming stop being a job title and become a capability, the same way ‘typist’ stopped being a job title when it became a thing everyone was expected to know.”
I think that’s right. But transitions are painful for the people caught in the middle. The developer who starts their career in 2026 faces a fundamentally different landscape than the one I navigated in the 2010s. The onus isn’t just on them to adapt — it’s on the industry to rebuild the ladder we’ve dismantled.
IBM is trying. Salesforce isn’t. The rest of us need to decide which camp we’re in, because the choice we make today determines whether there’s a generation of senior engineers in 2036 — or just a lot of AI-generated code with nobody qualified to review it.