Every senior professional I respect learned their craft doing work that no longer exists. The all-nighter rebuilding a financial model because the partner changed one assumption. The fourth rewrite of a research memo nobody would ever read. The unglamorous reps. That work is now the first thing an agent does — faster, cheaper, and without complaint.
Which leaves a question that sounds rhetorical but isn't: how do you become a senior if you can no longer be a junior?
The numbers say this is no longer hypothetical. Entry-level job postings in the US have fallen roughly 35% year over year, according to Revelio Labs. Employment for 22-to-25-year-olds in AI-exposed roles has dropped 13% since late 2022 — for young software developers, closer to 20%. Analysts watching the shape of the knowledge workforce describe an emerging hourglass: capacity concentrated at the agent layer and the senior layer, with the middle — the traditional training ground — quietly thinning out.
Consulting feels this first because consulting institutionalized the apprenticeship more than any other profession. The leverage pyramid was never just an economic model; it was a pedagogical one. Analysts built the model, associates pressure-tested it, managers shaped the storyline, partners carried the judgment into the room. Every layer learned by doing the layer below. Pull the bottom out of that pyramid and you haven't just changed the cost structure — you've severed the supply chain of judgment.
Here is the part most automation conversations miss: the grunt work was never only labor. It was the curriculum. The economist Kenneth Arrow formalized this in 1962 as learning-by-doing — competence compounds through repetition at the point of production, not through instruction. You didn't learn to spot a broken analysis by being told what broken looks like. You learned by building three hundred analyses, getting corrected on ninety of them, and slowly internalizing the difference. Thousands of small, low-stakes decisions, each with feedback. That was the tuition. Agents have now made the tuition free — by making the course unavailable.
The uncomfortable irony is that juniors equipped with AI look more productive than ever. A junior with an agent can produce a volume of code, analysis, and prose that would have taken a team a decade ago. But production without the ability to verify is fragility wearing a productivity costume. They need experience to manage the AI, and they can't get experience because the AI does the work they would have learned on. Left unaddressed, we are five to ten years from a generation of architects who have never laid a brick — leaders fluent in directing systems they cannot inspect.
So what does a serious response look like? Not nostalgia. The old apprenticeship had plenty of waste in it, and no client will pay for artisanal grunt work as a training subsidy. The response is to admit that development now has to be designed, where it used to be a byproduct.
Three shifts follow from that. First, move juniors from production to interrogation: their formative reps should come from reviewing, stress-testing, and breaking agent output — verification is the new first rung, and it teaches the same pattern recognition the old drafting work did, at higher intensity. Second, treat a portion of human-done work as deliberate tuition: some analyses should be built by hand not because it's efficient but because the firm is purchasing judgment for its future bench, the way airlines pay for simulator hours nobody bills to a passenger. Third, make seniors' reasoning visible: when the partner overrides the agent's recommendation, that override — and the why behind it — is now the scarcest teaching material in the firm. It used to transmit automatically through shared late nights. Now it has to be transmitted on purpose.
None of this is expensive compared to the alternative. A firm that automates away its training ground saves money every quarter until the day its bench is empty, and then no amount of money fixes it quickly — judgment has a long lead time, and you cannot rush an apprenticeship you never started.
The career ladder didn't disappear because anyone decided to remove it. It disappeared because every individual rung stopped making economic sense on its own. That's what makes this moment deceptive: no single actor is doing anything wrong, and the system is still failing. Rebuilding the ladder won't happen by default, because it was never built by default in the first place — we just had the luxury of forgetting that. Someone in every firm now has to own the question the pyramid used to answer for free: where, exactly, will your next senior come from?
✌🏻
Sources: