The entry-level job market broke. Here's the opening it left behind

Recent graduates are unemployed at 5.7 percent, against roughly 4.2 to 4.3 percent for the workforce overall, and computer science graduates specifically are worse off at 6.1 percent, nearly double the rate of most other majors, in a market where entry-level postings overall have fallen by single digits year over year and software-specific listings by considerably more, on every tracker's count. A London School of Economics and University of Warwick study found remote work predicts the entry-level decline better than AI does, mostly because remote hiring makes on-the-job supervision more expensive and supervision is exactly what junior roles used to run on. But AI is not nothing: early-career workers in roles most exposed to it, customer service and software development among them, have seen employment fall 16 percent relative to their more experienced peers over the same period. The market is genuinely tighter than it was three years ago, for reasons that are only partly about AI. Whatever the cause, the junior role graduates were counting on is harder to get than it used to be.

• 5.7%

  recent graduate unemployment, against 4.2–4.3% for the workforce overall

• 6.1%

  computer science graduate unemployment — nearly double most other majors

• -30%

  entry-level software engineering postings, year over year

• -16%

  relative employment decline for early-career workers in the most AI-exposed roles

The apprenticeship model in most companies never paid for output alone. It paid a junior to absorb judgement, slowly, by doing the tickets nobody senior wanted while someone more experienced corrected their mistakes, and the salary was the wrapper around that training. AI has taken over a large share of that ticket queue. What it has not taken over is the judgement the apprenticeship was quietly training in: deciding which problem is worth solving, reading a room, owning a decision when it turns out wrong. None of that got automated, it just lost its usual training ground. That is the loss, and it is real, not imagined.

Jensen Huang told Carnegie Mellon's graduating class this year to run, don't walk, toward AI, and said no generation has entered the workforce with more powerful tools or greater opportunity than this one. The advantage a 45-year-old normally has over a 22-year-old is a decade of pattern-matching on how the tools work, and that advantage does not exist yet in AI-native building, because the tools are too new for anyone to have banked it. The gap right now runs between fluent and not fluent, and fluency in these tools is measured in weeks, not years.

40 2021

29 2024

Median age of the founders behind AI unicorns, per Antler's tracking.

Signal	What it shows
The Leonis AI 100	Median AI startup founder age of 29 at the point of founding
Zach Yadegari, Cal AI	Built a calorie-tracking app to $30M+ ARR by age 19, then sold it to MyFitnessPal

Research firm Antler tracked the founders behind AI unicorns and found the median founder age fell from 40 in 2021 to 29 in 2024, and a separate industry study, the Leonis AI 100, put the median age of AI startup founders at 29 at the point of founding, most coming straight out of university or a research lab rather than a decade of corporate seasoning. Zach Yadegari is the sharpest version of this: he built Cal AI, a calorie-tracking app, to more than thirty million dollars in annual revenue by the time he was nineteen, and sold it to MyFitnessPal. He did not out-experience anyone. He shipped faster than people twice his age who were still waiting for permission.

Everyone should start a company is bad advice dressed up as inspiration. Most people should not, at any age, and a founder's path is harder and lonelier than the pitch decks make it look. What is being said is narrower: if a graduate is facing a market where the traditional door is genuinely stuck, building something real while still job-hunting is no longer the side project it used to be. The tools now let one person do in a few weeks what used to take a small team two months. Ship something narrow, put it in front of real users, and a candidate walks into their next interview knowing more about how these tools behave in practice than most of the people on the other side of the table. Do it twice, and the years-of-experience line on the job posting stops being the obstacle it looks like.

The market really is squeezed for graduates trying to get hired into someone else's system, and it would be dishonest to say otherwise. It is comparatively open for graduates willing to build their own, not because the barrier is low, but because for the first time in a long time, nobody older has much of a head start. The tools are new enough that no one has seniority in them yet. That includes the graduate deciding what to do next.