The entry-level software roles are disappearing. Looking at the big tech companies, SignalFire's 2026 State of Talent report says that the hiring of new grads is down by almost 65% compared to 2019, and even 76% for early-stage startups. That's the market any newcomer faces. It's rough.
But it's not as simple as "AI took these jobs, case closed" — it's more like the work has been relocated. Instead of writing much of the code, people are now pointing the AI to write it, and then verifying if it works. Which is why, we think, what helps a junior get hired has changed from collecting credentials to having a portfolio of things that visibly work.
Why the junior roles are so hard in 2026
The case is simple: the things juniors were usually hired for got automated first.
Like SignalFire say, almost 65% less for big tech companies compared to 2019, and 76% for early-stage startups; even top-20 CS programmes' graduates are much less likely to land a role at a big tech company now than they were a few years ago. But, they still post these roles, which are often being filled by more experienced people.
Why? Because when it comes to the junior positions, the first 1–1.5 years used to be about doing a lot of boilerplate and running unit tests, or performing low-hanging fruit bugfixes; that's what the big companies handed to AI. It was like an entry ramp — you do simple things for little money while learning how to do the hard ones; the ramp has been removed because the AI has eaten the simple things.
It's not only about the AI, though. The market situation plays a role too — budgets, interest rates, the post-2021–2022 over-hiring correction etc; but we reckon the AI's share was the biggest at the lowest rung.
Where did the work go? Into overseeing the AI
Ofc, the same reports that show juniors struggle also show the rise of a role of steering the model rather than competing with it. For example:
AI Engineer is the fastest-growing job on LinkedIn's 2026 Jobs on the Rise list; four of the top five are AI-related
Stanford's AI Index 2026, based on Lightcast data collected from billions of postings, says that 2.5% of US postings now require AI skills (which is an increase by 55% in a single year), and the postings referring to "agentic AI" (building and operating autonomous agents) went from 0.06% to 0.23% of the market, which is a rise of more than 280%
LangChain's 2025 State of Agent Engineering, based on the survey among ~1300 practitioners, shows that majority of them already ran agents in production (which was barely a thing 2 years ago)
So, it's not that people don't write code anymore — it's that they write it themselves less often. The focus has shifted to verifying that it works, instead of implementing it. It's about choosing what to build, telling the model to build it, and checking if it works.
Because typing became cheap, but judgement didn't.
What actually gets a junior a job in 2026?
A portfolio of things that actually work — not an impressive title or another year being "mid-level" on a CV; showing evidence that you can take an actual problem to a working solution. Because once the machine writes the code, the tenure doesn't tell the recruiter much anymore. It used to be a proxy for "I wrote a lot of code, I learned from it"; but it's not that simple anymore.
What tells more is the artefact itself — the fact that you built sth real and working, the fact that it's publicly accessible so anyone can just see it. What matters here is that it works, not that the code is impressive.
Nowadays AI keeps getting better at everyday tasks, everyone reaches for it to save time, and patience for anything slower than it has to be runs thin. So what can a recruiter assess in 5 minutes?
Does it work?
Would I use it?
Can I see this person directs AI towards things that actually hold up?
If you can show a "yes" to all of these with a portfolio of your things, the title becomes far less relevant, and nobody asks about your experience as long as the artefact is just visibly working.
What does such a portfolio look like?
Quality over quantity. It doesn't need to be big — one thing that works and is actually used by a few real people does far more than ten started-but-unfinished projects or half-done tutorials with changed labels. One project that:
Is fully functional and finished
Works from the start. You share a link, and it loads; you share a repo, and it runs from a clean checkout; you show a demo, and there's no need for apologies
Solves an actual problem. It doesn't need to cure cancer, but its purpose is obvious to anyone — "I needed sth, so I built it" beats "I wanted to learn this framework, so I built sth with it"
Shows it was built with a vision and quality control at every step — you steering the AI, not just vibecoding it from a few prompts. You point at what needs to be done, the model writes the code, and then you check that it works; building a project this way and openly describing it as such is the best proof of the very skill people are hiring for
That's partly from our experience — we've been keeping a public open-source portfolio that we treat as tangible evidence of our ability to do the work. We think anybody starting now should aim for that too.
Because if the entry ramp is gone, don't wait for it to be built again — make your achievements speak louder than your title.
