Entry-level software hiring is down in the United States, the United Kingdom, Canada, Germany, the Netherlands and Sweden, while senior employment has held up far better. That is the headline of a Financial Times analysis published this week, and it matches what most junior engineers already feel: the first job has become the hardest one to get.
The easy story is that AI took those jobs. The data does not fully support it, and the real story is more useful. This post is for two readers: junior engineers trying to get through the door, and the hiring managers deciding who gets through it. The short version for both is the same. AI did not remove the need for engineers. It removed the need for engineers who can only type code, and it made engineering judgement the thing that gets you hired.
What the numbers actually say
The FT analysis makes four points, and they are worth reading together rather than picking the scariest one.
- Juniors are hit, seniors are not. Entry-level software hiring fell across all six countries. Senior-level employment stayed comparatively stable.
- The AI link is weak outside the US. In the UK, Canada and Australia, the connection between AI adoption and falling employment is largely absent, even though their hiring trends look like the American one.
- Productivity went up, output did not. In US sectors adopting AI, productivity gains have not turned into faster growth in output or revenue. What they coincided with is flat or falling headcount.
- The economy explains a lot of it. The hiring slowdown and the layoffs may reflect inflation, higher interest rates and general uncertainty more than direct replacement by AI.
Independent research points the same way on the first point. Stanford's Digital Economy Lab found that employment of 22 to 25 year olds in the most AI-exposed occupations now sits 19% below where it would be had it kept pace with less exposed peers, with no comparable gap for experienced workers (Brynjolfsson, Chandar and Chen, "Canaries in the Coal Mine?").
Put those together and you get a more precise picture than "AI is eating junior jobs". Companies under cost pressure stopped hiring the people they would have had to train. AI gave them a reason to believe they could get away with it. Whether they actually can is a different question.
Why companies stopped hiring juniors, and why that will not last
A junior hire has always been a bet: you pay for six to twelve months of learning and get an engineer on the other side. When money is expensive, that bet is the first thing a budget cuts. When a team lead can ask an AI assistant to write the boilerplate a junior used to write, the bet looks even worse on a spreadsheet.
The flaw in that logic is in point three above. If AI made teams dramatically more productive, output and revenue would show it. Mostly they do not. What AI removed is the cheapest, most visible part of junior work: writing a function from a clear spec. It did not remove the parts that actually cost senior time: understanding why a service is down, reading a system nobody documented, deciding whether a generated change is safe to ship.
And seniors do not appear from nowhere. Every team that stops hiring juniors today is choosing to have no mid-level engineers in three years. We made the long version of this argument in why hiring junior engineers is still worth it.
For junior engineers: AI is the fastest way in, if you use it as an engineer
Here is the part that is genuinely good news. AI has made it faster than ever to learn the engineering skills that companies still pay for, as long as you use it to understand systems rather than to skip understanding them.
Two juniors can use the same AI assistant very differently. One pastes the error, accepts the fix and moves on. The other asks why the error happened, reads the logs the fix depends on, and checks what else the change touches. After six months the first has a portfolio of code they cannot explain. The second has the judgement of someone with two years of experience. Only the second one gets through a good interview.
What that looks like in practice:
- Learn the system around the code. Use AI to explain what you are looking at: a
Dockerfile, a failing CI job, a Kubernetes manifest, a systemd unit. Then run it, break it and fix it yourself. Linux, networking, containers and CI are where AI helps a learner most, because the explanation is instant and the feedback from a real machine is honest.
- Treat every generated change as a code review. Read it line by line, run it, and be able to say why it is correct. Reviewing AI output is now a core skill, not a side one. We wrote a checklist for it in how to review AI-generated code.
- Debug before you prompt. When something fails, form a guess first, then ask the model. Comparing your guess to its answer is how intuition gets built. Skipping the guess is how it never does.
- Build one real thing end to end. A small service with a database, a deploy pipeline and monitoring teaches more than ten tutorial apps. AI makes it realistic for one person to build all of it, which was not true five years ago.
- Be able to narrate your process. In an interview you will be judged on how you think. Practise explaining what you checked, what you ruled out and why you trusted the answer.
The gap the market opened is not "people who can use AI" versus "people who cannot". Everyone can use AI. The gap is between people who can tell when AI is wrong and people who cannot, and that is an engineering skill you can learn faster now than ever before.
For hiring managers: stop screening juniors for the part AI already does
If entry-level hiring is down, every junior you do hire matters more, and the old screens are badly suited to finding the right ones. An algorithm puzzle measures exactly the skill AI made cheap. A junior who memorised fifty LeetCode patterns and a junior who can actually debug look identical on that test.
What separates them shows up in a real environment:
- Give them a broken system, not a blank file. A service that will not start, a failing test suite, a container that exits on boot. Watch how they find the cause.
- Let them use AI, and watch how. Do they read what the model produced? Do they verify it? Do they notice when it is confidently wrong? This is the single best predictor of how a junior will work on your team in 2026. We covered how to score it in how to test whether an engineer knows how to use AI.
- Score the process, not only the result. A junior who gets 80% of the way with a sound method is a better hire than one who got lucky with a pasted fix. Session recordings make this visible after the fact, so the reviewer does not have to catch it live.
- Keep a human in the decision. AI can summarise a session and point at the interesting moments. The call on whether someone can grow into your team belongs to an engineer who knows the team.
This is how EasyEnv runs junior assessments: a real Linux environment built from your stack, AI tools allowed, the session recorded, and a person making the decision. The approach for AI-native candidates specifically is in assessing AI-native junior engineers.
Where this argument has limits
To be fair to the other side: the data is early, and it is mixed. The weak AI link outside the US may simply mean adoption there is a year or two behind. Some routine programming work is genuinely being absorbed by AI tools, and a junior whose only skill is writing simple CRUD code from a ticket is in a worse position than they were in 2021. Learning engineering fundamentals does not guarantee a job in a market with fewer openings. It does put you in the group that companies still hire from when they hire at all.
The takeaway
The first job got harder for reasons that are mostly economic, with AI as the accelerant and the excuse. For junior engineers, the response is not to compete with AI on typing code but to use it to learn the engineering around the code faster than anyone could before. For hiring managers, the response is to test juniors in the conditions they will actually work in, with AI on the desk, and hire the ones who can tell when it is wrong.
If you hire juniors: what does your first-round assessment measure today, and is it something an AI assistant could already pass?