The thinning bottom rung: AI and the vanishing entry-level job

A few years ago, the path into a software or data team had a reliable first step. You joined as a junior, and for the first stretch you did the unglamorous work: fixing the failing test, wiring up the boilerplate, cleaning the messy CSV, writing the function someone senior had already sketched on a whiteboard. It was tedious. It was also how you learned. Every small ticket taught you how the system actually fit together, and by the time you had closed a few hundred of them, you had quietly become someone who could be trusted with bigger things.

That first step is the rung being sawn thin right now. The tasks that used to fill a junior’s first year are the exact tasks an AI assistant does in seconds, for roughly the cost of a coffee. And when the cheapest version of the entry-level job is a subscription rather than a salary, the entry-level job itself starts to disappear.

This is the part of the AI story that matters most to anyone reading this site as a student or in their first few years of a data or engineering career. The headline numbers about AI and jobs are often vague and apocalyptic. The reality is narrower, more specific, and more useful to understand: the damage is concentrated at the bottom of the ladder, and that has direct consequences for how you should spend your time.

The number that should get your attention

The clearest signal comes from a Stanford analysis of payroll data, reported in late 2025, which found that employment among software developers aged 22 to 25 fell by roughly 20 percent between 2022 and 2025 — the same window in which AI coding tools went from novelty to default. That is not a rounding error. It is one in five of the youngest developers, gone from the headcount, in three years.

What makes the finding sharp is the contrast. Overall developer employment held up over the same period. Demand for experienced and senior engineers stayed strong; several large employers, including Google, have said publicly that they intend to keep hiring engineers even as AI raises per-person output. So this is not a story of software collapsing as a field. It is a story of a field that still wants people, but increasingly does not want them at the very start of their careers, doing the very work that used to be the on-ramp.

Why the bottom rung specifically

To see why the youngest workers take the hit, look at what AI assistants are actually good at. They are extraordinarily strong at small, well-specified, self-contained tasks where the answer is one of many similar answers seen before: generating boilerplate, translating a clear spec into code, writing a unit test for an existing function, fixing a syntax error, drafting a query, cleaning a dataset into a tidy shape. They are far weaker at the things that take taste — deciding what to build, noticing that the requirement itself is wrong, weighing two designs against each other, sensing when an output is subtly off in a way the tests would not catch.

Now line that up against how careers used to work. The well-specified, self-contained tasks were the junior’s job. That was the rung. You did the small stuff under supervision until you had absorbed enough context to take on the ambiguous stuff. AI does not threaten the whole ladder evenly; it dissolves precisely the bottom rung, because the bottom rung is made of exactly the work AI does most cheaply.

Mercer’s Global Talent Trends research, which surveys thousands of workers and leaders, lands on the same conclusion from the demand side: it flags workers aged roughly 22 to 27 as facing the highest displacement risk, on the explicit logic that the simple tasks which used to train juniors are the first to be automated. The same survey found that nearly all chief executives expect AI to drive at least some reduction in headcount within two years. You do not have to believe every CEO will follow through to see the direction of intent.

There is a quieter, second-order problem hiding inside this. Senior engineers do not appear from nowhere; they are juniors who survived a few years of doing the small stuff. If organisations stop hiring and growing juniors because AI covers the entry-level tasks, they are quietly eating the seed corn of their own future senior bench. The squeeze is rational for any single company in any single quarter, and collectively it may be a slow-motion mistake. That tension is worth naming, because it means the door is not closing as cleanly as the headcount numbers suggest — companies still need to grow people, even if many have not yet noticed the bill coming due.

The cost lands on the people least able to absorb it

The strain shows up in how early-career workers feel, not just in payroll spreadsheets. Across recent surveys, Gen Z reports the highest burnout of any generation — comfortably above 50 percent in several measures — and the highest financial insecurity, with a large majority saying money stress affects their mental health. Layer onto that the specific dread of training for a profession while reading that the entry door is narrowing, and you have a cohort carrying an unusually heavy psychological load at the exact moment they are supposed to be finding their feet.

Mercer’s data captures the mood shift bluntly: the share of workers who say they “feel good at work” fell from about two-thirds to under half in two years, while the share who fear losing their job to AI climbed from roughly a quarter to about 40 percent. Those two lines moving in opposite directions are, in a sense, the whole story of this moment — confidence down, anxiety up — and the youngest workers sit where the two trends cross hardest.

A fair caveat is in order, because these are surveys with different definitions and the labour market has more than one thing going on at once. The youngest developers were also hit by a broad post-2022 tech hiring correction, higher interest rates, and the unwinding of pandemic-era over-hiring — AI is not the only force in the room, and honest analysts say so. But the age-specific shape of the decline is hard to explain by a general slowdown alone: a recession trims hiring across the board, whereas this pattern bites the 22-to-25 bracket while leaving senior demand intact. When the cut is that precisely targeted at the people doing automatable work, the simplest explanation is the one the data keeps pointing back to.

How to stay hireable when the rung is thin

None of this means the field is closing to you. It means the cheapest, most automatable version of an early-career worker is what is vanishing — and the move is to not be that worker. Concretely, that comes down to being worth more than the work an AI does for free.

Go beyond what AI does cheaply. If your value proposition is “I can write a function from a clear spec,” you are competing with a tool that does it instantly and never sleeps. The durable value is upstream and downstream of that: figuring out what should be built, noticing when the spec is wrong, judging whether the AI’s output is actually correct, and integrating a dozen pieces into something coherent. Treat AI as a power tool you wield, not a competitor you race. The person who can prompt, review, correct, and take responsibility for AI-assisted output is worth far more than the person who could only have produced the easy first draft themselves.

Build judgment on purpose, since you will not absorb it by osmosis anymore. Older engineers got their judgment by grinding through hundreds of small tickets. If AI does those tickets, you have to acquire the same instincts deliberately: read real production codebases, study why systems are designed the way they are, do post-mortems on your own bugs, and force yourself to understand the code the AI wrote rather than shipping it on faith. When AI hands you an answer, the high-value habit is to ask why it is right and where it would break. Judgment is the thing that does not come in the subscription.

Ship real things, end to end. A portfolio of small projects you actually finished — designed, built, deployed, and kept running, with the rough edges and trade-offs you can talk about — is now worth more than a transcript, because it demonstrates exactly the integrative, end-to-end capability that AI does not replace and that employers can no longer take for granted in a junior. Owning the whole arc of one real thing teaches you more, and proves more, than closing a hundred isolated tickets ever did.

Own outcomes, not tasks. The frame that protects you is to stop thinking of yourself as a person who completes assigned tasks and start thinking of yourself as a person who is accountable for a result. Tasks are what get automated. Outcomes — “the checkout flow works and converts,” “the data pipeline is reliable and the numbers are trusted” — require someone to care, decide, and answer for them. That someone is hard to replace with a subscription, at any age.

The honest bottom line

The bottom rung of the ladder is genuinely thinner than it was, and pretending otherwise would not help you. The youngest workers in software and data are absorbing a real, measurable squeeze, and the anxiety they report is a rational response to a changing market, not a fragility to be coached away.

But the ladder still exists, and companies still need people who will grow into seniors — they have just made the first step higher and narrower. The way up is no longer to be the cheapest pair of hands for the simple work, because that work now costs a few cents. It is to become, as fast as you can, the kind of person who decides what the work should be, judges whether it was done right, and owns the result. AI did not raise the bar for the easy stuff. It removed the floor. Aim above it.