AI Is Killing the Entry-Level Job — and the Ladder Your Career Was Built On

Somewhere in an office right now, a manager is looking at a task that used to go to the newest person on the team — a first draft of some code, a chunk of background research, a standard contract or report — and quietly routing it to a machine instead. That single decision, multiplied across millions of teams, is why the conversation about entry level jobs and AI has turned so anxious so fast. The tasks we once handed to juniors are precisely the tasks generative AI does first and cheapest. The bottom rung of the career ladder is not being upgraded. It is being sawn off.

I want to be careful here, because there is a lot of noise. But the core worry is real and it deserves a clear-eyed look: if the entry-level job is what disappears first, then the whole structure we use to turn beginners into experts is in trouble. This is not just a story about who loses a job this year. It is a story about where the next generation of senior people is supposed to come from.

Why the bottom rung is the first to go

Think about what an entry-level knowledge worker actually does in their first two years. A junior developer writes boilerplate, fixes small bugs, and produces first-pass code that a senior reviews. A junior analyst pulls data, summarises documents, and builds the slide nobody senior wants to build. A first-year associate at a law firm reads through discovery, drafts routine clauses, and formats memos. A new marketer writes the fifth version of ad copy and cleans up the brief.

Now list what today’s AI tools are genuinely good at: first-draft code, summarising and searching large piles of text, generating standard documents, and churning out variations of copy. The overlap is almost total. The work we designed as training wheels — low-stakes, well-defined, heavily supervised — is the same work that is easiest to automate. That is not a coincidence. It is the same property that made a task suitable for a beginner (clear rules, low risk, lots of examples to learn from) that makes it suitable for a model.

I have written before about which office jobs AI replaces first, and the pattern holds here with a cruel twist: it is not that entry-level people are worse than seniors, it is that their tasks are more legible. A senior’s value is bound up in judgement, relationships, and knowing which of the ten plausible answers is the right one. A junior’s value, for now, is mostly in output — and output is exactly what these tools produce.

The work we designed as training wheels is the same work that is easiest to automate. That is not a coincidence.

The broken ladder

Here is the problem that keeps me up at night, and it is bigger than any single layoff round. Careers are ladders. You do the grunt work not only because it needs doing but because doing it is how you learn. The junior developer who fixes a hundred small bugs is building an intuition for how systems break. The associate who reads ten thousand pages of discovery learns what actually matters in a case. You cannot skip these rungs. Expertise is compressed experience, and the experience mostly comes from the boring tasks at the bottom.

So ask the uncomfortable question. If AI does all the junior work, how does anyone become senior? If there are no first-draft tasks left for a human to draft, where does the person who reviews and improves the AI’s draft acquire the taste to know a good draft from a bad one? A firm can run for a few years on the seniors it already has. But seniors retire, leave, burn out. The pipeline that replaces them runs straight through the entry-level roles we are now automating away. Break the bottom of the ladder and you do not just hurt this year’s graduates — you starve the top of the ladder a decade from now.

This is the part the quarterly-earnings logic misses. Cutting junior headcount looks like pure efficiency on a spreadsheet: same output, lower cost. The cost is deferred and invisible. It shows up years later as a missing cohort of mid-level people who never got trained, a thinning of institutional knowledge, and a strange new scarcity — plenty of AI to do junior tasks, and almost nobody who knows enough to supervise it well. If you have wondered will AI take my job, the honest answer for early-career workers is that the immediate risk is real, but the structural risk — a career with no on-ramp — is the one we are barely talking about.

The India campus-placement warning

Nowhere is this more concrete than in India, where an entire economic dream is built on the entry-level tech job. For two decades the deal was simple and astonishing in its scale: study engineering, get placed on campus by one of the big IT services firms, and step onto a ladder that could lift a whole family into the middle class. Hundreds of thousands of graduates a year walked through that door. The door is now narrowing.

I want to hedge carefully, because precise numbers here are slippery and often overstated. But reports suggest the big Indian IT services companies have slowed fresher hiring markedly compared with the frenzy of a few years ago, and some surveys of engineering graduates find a widening gap between how many are qualified on paper and how many are actually getting placed. Executives themselves have talked openly about needing fewer people to do the same work as automation improves. I have looked at the shape of the Indian IT layoffs in 2026 elsewhere; the campus-placement squeeze is the same story caught one stage earlier, at the moment of entry rather than exit.

The reason this matters beyond India is that the Indian IT model was, in a sense, the world’s largest entry-level training machine. It took raw graduates and turned them into working professionals at industrial scale, largely by giving them exactly the kind of routine, well-specified tasks — testing, maintenance, basic development, documentation — that AI now targets. When your national growth story depends on that on-ramp, a broken ladder is not an abstraction. It is millions of young people with degrees and no first job to grow from.

The same move, a new machine

None of this is new in the way it feels new. Step back far enough and you see a pattern I keep coming back to: a technology arrives, and the story we are told is that it will free us — and the reality is that it captures something and hands the gains upward. It is the same move, a new machine, every time. This is the heart of how technology gets captured: the question is never simply what the tool can do, but who takes the upside, who pays the cost, and who is in a position to fight back.

Look at the ladder-breaking specifically and history is almost repetitive about it. The power loom did not just make cloth faster; it dissolved the weaver’s apprenticeship, the years-long path by which a boy became a skilled master with bargaining power. The craft ladder was broken, and what replaced it — at first — was tending a machine for a wage that no apprenticeship protected. Mechanised printing hollowed out the compositor’s trade. Bank-branch computerisation and then the ATM reshaped what a junior teller learned and how far they could rise. In each case the new technology arrived first as a solvent for the entry-level rung, and only much later, after painful decades, did new kinds of ladders get built.

Every revolution breaks a ladder before it builds a new one. The question is who has to survive the gap in between.

That is the pattern in a sentence: every revolution breaks a ladder before it builds a new one. The optimists are not wrong that new roles eventually appear — someone had to build and maintain the looms, the presses, the computers. But “eventually” does a lot of quiet work in that sentence. There is a gap between the old ladder breaking and the new one being built, and real people have to live inside that gap. A graduate today does not get to wait out a thirty-year adjustment. The gains from the machine are captured now, at the top; the cost of the missing rung is paid now, at the bottom, by the youngest and least powerful workers. Who takes, who pays, who fights back — the questions do not change even when the machine does.

What could keep the ladder intact

I am not a fatalist about this. A broken ladder is a choice, not a law of physics, and there are things that would genuinely help — if we decide the pipeline is worth protecting.

The first is a deliberate, unfashionable investment in apprenticeship. If AI can do the junior task, then the junior role has to be redesigned around something AI cannot yet do: learning to supervise, to judge, to catch the machine’s confident mistakes. That means firms treating early-career hires as an investment to be trained rather than a cost to be cut — pairing juniors with AI tools deliberately, giving them real responsibility for reviewing and correcting AI output, and accepting that this is slower and less efficient in the short run precisely because it is how you manufacture seniors. The companies that do this will have a pipeline in ten years. The ones that optimised it away will be bidding for scarce talent they chose not to grow.

The second is policy, because individual firms face a collective-action trap: each one is tempted to freeload on everyone else’s training. This is where apprenticeship subsidies, training levies, and education that teaches judgement and supervision rather than rote output can shift the incentives. Historically, the ladders that got rebuilt after a technological rupture were rebuilt on purpose — through unions, guilds, public education, labour law — not by the market spontaneously deciding to be kind. Someone has to fight back for the rung, or it stays broken.

And for individuals — because I do not want to leave you only with structural gloom — the move is to climb faster than the ladder is being cut. Concretely, that means a few things:

• Become the person who directs the tool, not the person who competes with it. The durable skill is judgement: knowing what to ask for, spotting when the output is subtly wrong, and owning the result. Use AI relentlessly, but position yourself one level above the task it does.
• Get to real responsibility early. Seek out work where the stakes are high enough that a human must own the decision — client relationships, ambiguous problems, anything where the answer is not in the training data. That is where the rungs still exist.
• Build the tacit knowledge deliberately, since you can no longer absorb it passively by doing a thousand routine tasks. Ask why, watch how the seniors decide, treat every AI output as a case to critique rather than accept.
• Prize the domains where trust and accountability are the product — the ones where someone has to be answerable to a human on the other side. Those on-ramps close last.

The bottom rung is vanishing; I do not think there is much use pretending otherwise. But the ladder is not gone yet, and whether it gets rebuilt is genuinely undecided. It will not rebuild itself out of market kindness — it never has. It gets rebuilt when firms choose to train, when policy makes training pay, and when the people standing at the bottom of the ladder refuse to accept that there is no way up. That refusal is the oldest part of this whole story, older than any machine. It is the one thing the machine has never managed to automate.