Who do you promote in 2030?
The entry-level squeeze is what happens when AI takes over the work juniors used to learn on. Nobody is fired. The role simply stops being refilled, and the bottom of the ladder quietly disappears.
You stopped hiring at the bottom because AI does that work now, and the numbers this year look better for it. The bill arrives in about four years, at the moment you need somebody who has been here long enough to be trusted with something.
Before you read on
The decision is made this year and the consequence lands in a different one, which is the whole reason this is easy to keep postponing. Pick a year and look at the same company.
Two junior roles went unfilled after people left. The work got absorbed, partly by AI and partly by people who were already good at it. Your cost per output went down and nothing broke. Nobody in the company experiences this as a decision, which is exactly why it never gets discussed.
Checking output, cleaning up, handling the routine cases that used to be somebody's training ground. Two of them have said something about it. One has left, and the exit conversation mentioned growth. When a team lead role opens, the internal shortlist has two names on it and both are already stretched.
The people who would have been ready by now are working somewhere that kept training them. So you hire from outside at a premium, and they need a year to understand your business, and in the meantime the person holding the knowledge is the one you have been meaning to give a sabbatical since 2027.
All three panels are readable without clicking. The years only change which one is on top.
How it shows up
This challenge never arrives as a plan to stop hiring juniors. It arrives as three separate sensible choices, each one made by a different person.
The measured part
Stanford's Digital Economy Lab works with ADP payroll data covering millions of workers across more than 730 occupations at tens of thousands of American employers, updated every month. They call the project Canaries in the Coal Mine, which tells you what they think they are looking at.1
That last point is the one worth repeating in your own leadership meeting, because it changes what kind of problem this is. There is no wave of dismissals in the data and there is no economy-wide displacement.1 What there is, is a door that stopped opening. Software engineering, marketing and customer service are the clearest cases so far.
Two other readings point the same way. Entry-level job postings in the United States are down by roughly a third since early 2023, and at the companies people most want to work for the share of entry-level hires has fallen while the median employee has got several years more experienced.23
Nobody decided to stop training people. Twelve separate reasonable calls added up to it.Paul Musters
Six short questions. Three minutes. Nobody sees who answered. You get the level, the cost of it, and the step after.
Where it comes from
Summarising, first drafts, the routine cases, the tidying. It looked like low-value work on a cost sheet and it was doing two jobs at once. It got the work done and it taught somebody the difference between output that is fine and output that is wrong. Automate it and you keep the first job and lose the second, and only one of those shows up this quarter.
Two junior salaries this year against a search fee, a premium and a year of onboarding four years out. Every incentive in a normal budget cycle points one way, and the person who benefits from the saving is rarely the person who pays the bill. That is not cynicism, it is just how planning horizons work.
The cheapest senior you will ever hire is the junior you kept.Paul Musters
Hiring belongs to whoever has the vacancy. Training belongs to the manager. What your company will look like in five years belongs to a strategy document. So the decision to stop refilling the bottom rung gets taken twelve times by twelve people, each of them locally right, and it never appears on an agenda as itself.
Do this first
This takes an hour with whoever keeps your headcount numbers, and it is the kind of thing that changes a conversation because nobody can argue with their own data.
Put these four things on one page.
Line four is the one that does the work. If the first number is falling while the second is rising and the fourth is shorter than the third, the arithmetic does not close, and it will not close by itself.
There is a version of this that goes wrong, so a warning. It is tempting to write down names of people who could probably do it. Write down only the ones you would actually appoint tomorrow if you had to. The gap between those two lists is the real finding, and it is usually larger than anybody expects.
Level by level
There is a version of this where it works out. The junior work is gone and something else takes its place: reviewing output, spotting where the machine is wrong, judging quality. That is real work and it can teach somebody a great deal, faster than the old route did.
It requires that somebody notices and designs for it. In the five AI Culture Levels we use with clients, that is Level 3 behaviour: the way of working is written down, so a new person can be pointed at it and can be measured against it. At Level 2 there is nothing to hand over, because everyone works their own way and none of it is documented. A new junior at Level 2 has nothing to be junior at.
An Operating Profile in use. Personality type and AI level in one profile, with the agents that fit it.
A headcount table tells you how many people you have and how long they have been here. It does not tell you who is actually growing, and those are different questions. Somebody two years in who is running whole workflows and judging output is further along than somebody six years in who has been doing the same thing carefully since 2020.
The measurement is per person and it holds two things: the way somebody works, and their AI level. That gives you a bench you can see rather than a list sorted by start date. The people who could be ready in three years are usually on it, and they are usually not the names that came up in the meeting.
emaho measures one Operating Profile per person: personality type and AI level in a single profile. On that we build a personal set of AI agents that fit how that person works, inside the tools they already use. Fifteen minutes to complete, first profile free, built for companies between 20 and 500 people.
Fifteen minutes per person. No credit card, no strings.
What works
Put the pipeline on somebody's plate by name. One person who owns the answer to what this company looks like in five years, with the headcount numbers in front of them once a quarter. Right now that question belongs to a document, and documents do not notice things.
Then redesign the junior role instead of deleting it. The work is different now: checking output, catching what the model gets wrong, handling the cases it cannot. Somebody who does that for a year learns quality faster than the old route taught it, provided there is a person next to them who can say why something is wrong.
Protect the teaching time explicitly. Your seniors are doing the checking because they are fast at it, which is efficient today and takes the teaching out of the week. Two hours a week with somebody less experienced, in the calendar, treated as work rather than as goodwill.
And keep hiring at the bottom, in smaller numbers, on purpose. Two people who are early and who you actually intend to grow. The companies that come out of this decade with a bench are the ones who kept a thin line open when everyone else closed it.
If the names you could actually appoint in three years fit on one hand, I would like to hear who is on it and what they are missing. It is a conversation I have often and I usually have something useful to say about it.
The price of leaving it
Start with the one you can put a number on. A senior hire from the market costs a search fee, a premium over an internal promotion and roughly a year before they know your business well enough to be trusted with the difficult things. Multiply that by the number of roles you were quietly planning to fill from inside.
Then the retention effect, which arrives sooner. Your good people can see the shape of the ladder above them. When the rungs disappear, the ones with options leave for a place where they can see a next step, and those are the same people you were counting on for line four of your test.
And the slow one. Judgement in a company lives in people who have seen enough versions of something to know when it is off. That takes years and it is built on exactly the work that got automated. Skip a cohort and you do not notice for a while, because the seniors you have are still carrying it. You notice when they retire, which is the worst possible moment to find out.
What people search for
Where this goes next
These three are the challenges that show up in the same conversation about what the company will look like.
Fewer juniors means wider spans and more checking for whoever is left. The manager is where that lands, and manager engagement fell nine points in three years.
Read this one In the full set 15If the bottom layer is gone and half the work runs through something that is not employed, the chart on your wall describes a company you used to have.
Read this one In the full set 19The other half of this one. Even the juniors you do have stop building the judgement the job used to teach, because the first answer arrives before they have thought about it.
Read this oneTwenty-five things that break inside a company once people start using AI, each with the research behind it and the level where it starts to bite. This one starts to bite at Campfire, and 2 of the others start there too.
Your next step
The one-page test tells you how short the list is. A profile per person tells you who is on it, and that is rarely the same as the list sorted by start date.
First profile free · no credit card · built for companies of 20 to 500 · you decide what your team gets to see
Rather start at company level? The Culture Level scan is six questions, three minutes, and asks nothing of you.
Fifteen years of leadership and team development in Dutch scale-ups. That practice now sits in software: one Operating Profile per person, with agents that actually fit. He writes these pages from what he runs into with clients, not from a research summary.
Numbers are quoted as published. The Stanford figures were checked at the source. The three moments in the recognition block are composites from conversations rather than accounts of named companies.
