Most of the conversation about AI and jobs is pointed at the wrong horizon. The news cycle is full of transformation forecasts for 2028 and beyond. The thing that should be worrying senior leaders is happening right now, in the first year of the career ladder, and almost nobody is pricing it in.
Gartner reported in March 2026 that AI will significantly transform 32 million jobs per year in the near term, with workflow-heavy roles hit first — service desks, business analysts, project managers, junior engineers (CIO, 10/03/2026). A Resume.org survey of around 1,000 US business leaders, cited in the same piece, found that 21% of companies have already stopped hiring entry-level employees because of AI, and half expect to do the same by 2027. One in three expect entry-level roles to be eliminated at their organisation by the end of this year.
Strip the framing and what you have is an experiment being run at scale without a plan for what happens next. Cut year one from the career, and you are not just removing a cost line. You are removing the apprenticeship system that produces the people you will need in 2030.
The number that should worry you
BCG’s April 2026 analysis estimates that 50–55% of US jobs will be reshaped by AI over the next two to three years. Most stay. Most change. 14% fall into what BCG calls “rebalanced roles” — same headcount, redesigned work, higher skill floor (BCG, 03/04/2026). The honest read on the data is that AI is changing jobs faster than it is cutting them.
But the BCG paper also flags the uncomfortable second-order effect: “as AI absorbs much of the routine work that has historically justified large entry-level hiring cohorts, fewer execution-focused positions will be required.” Stanford-adjacent labour research has already detected it. Employees aged 22–25 in AI-exposed jobs saw employment decline by 16% compared to similar workers in less-exposed roles (AI Agent Store workforce brief, 14/04/2026).
16% in a single age cohort, inside a single economic cycle, with AI adoption still in early innings. That is not a gentle rebalancing. That is the bottom rung disappearing while the ladder is still in use.
Why the apprenticeship model matters
I have spent thirty years watching platforms and work patterns shift — mainframe to client-server, on-prem to cloud, waterfall to agile, manual ops to SRE. Every one of those transitions had casualties, but every one also had a mechanism that made the next wave of seniors possible. The mechanism is always the same, even when the technology changes. Juniors do low-stakes work, in proximity to seniors, under the cover of forgiveness that comes from being new. They pattern-match. They inherit judgement by osmosis. They make the mistakes that turn into instinct.
The thing that creates a good senior engineer is not formal training. It is 1,000 conversations about tradeoffs, most of them unplanned, most of them under pressure, held with someone two or three rungs further up the ladder. You cannot replicate that by reading a runbook. You cannot simulate it in a course.
When a team’s routine work is absorbed by an agent, the conversations go with it. The senior engineer still gets the exception. The junior never gets the routine case that would have built their instincts. The pipeline that produces the next generation of seniors is quietly cut, and nobody has to make the call explicitly. It just happens through a hundred small procurement decisions and a few headcount freezes.
The lag nobody is modelling
Senior capability has a development lag measured in years, not quarters. Most organisations plan in quarters. That mismatch is how you end up with a capability crisis in 2029 caused by a hiring decision made in 2026. By the time the gap shows up as missed delivery or shallow architecture work, the people who would have filled it have already moved industries or given up entering ours.
What the optimistic case misses
BCG argues, reasonably, that entry-level roles will not disappear — they will be redefined for candidates who can “take on higher-order tasks, such as supervising AI outputs, managing exceptions, and contributing to more complex problem solving earlier in their careers.” The theory is elegant. Juniors start higher up the value chain because AI handles the drudge work.
The problem with that theory is that it assumes the ability to supervise AI outputs is something you can hire for in a 22-year-old. It is not. Supervising AI output is a senior skill dressed in junior clothing. You need to know what the correct answer looks like before you can judge whether the model produced it. That knowledge comes from the work the model is now doing.
Put differently: the optimistic case describes a workforce in which every junior is already halfway to senior on day one. That workforce does not exist. It cannot exist, because the very steps that would have produced it are the steps being automated.
A framework for protecting the pipeline
If you run an engineering, product, or delivery organisation, this is a problem you can act on without waiting for a policy response. Four practical moves, in priority order:
- Protect at least one “apprenticeship lane” per team. One project, one workstream, or one type of ticket that is deliberately kept as human-first work even when an agent could do it. Cost it as a training line, not a delivery line. The ROI is a functioning senior in three to five years.
- Redesign the junior role around judgement, not throughput. If AI can do the routine, the junior’s job is no longer “do the routine.” It becomes “review and correct the routine, flag the edge cases, write up the patterns.” This only works if there is a senior close enough to correct them. Matrix the reporting lines accordingly.
- Treat the first 18 months as paid training, not paid work. This is the mental shift most organisations will refuse. But the honest accounting already looks like this — juniors cost more than they produce in year one. The lie was that they paid for themselves via routine output. That lie is gone. Price the seat at what it actually is.
- Measure flow of seniors, not just flow of tickets. Add one KPI to the delivery dashboard: the count of engineers, analysts, or PMs who moved up a band this quarter, and the count on track to move up in the next four. If the number is zero, you are consuming seniority without replenishing it.
Where this breaks
Strong opinions, loosely held — there are three places this argument is weakest, and I would rather name them than pretend they are not there.
First, the apprenticeship model has always worked better in theory than in practice. Plenty of organisations ran juniors through five years of low-stakes work and produced mediocre seniors. Removing the broken version of the model may force something better. Possible. Not guaranteed.
Second, the training case looks different outside knowledge work. In trades — plumbing, electrics, skilled manufacturing — apprenticeship is explicit and paid for as such. That model works. Knowledge work borrowed the apprenticeship idea without the honesty. Maybe the right response is to make it explicit, the way the trades already have.
Third, there is a genuine possibility that AI tutoring — personalised, patient, always available — closes part of the gap that used to require proximity to a senior. I am sceptical that it closes all of it, because the thing a good senior gives a junior is not information. It is taste. But I would not bet against better AI tutoring moving the needle on the simpler end of the problem.
What to do on Monday
Three concrete actions for anyone running a team this week.
- Audit your current work pipeline for routine tasks that are about to be automated. For each one, ask: which junior was learning from this? Where does their learning move to?
- Pull out your last two years of promotion data. Plot the count of people who moved from band 1 to band 2, and from band 2 to band 3. If the trendline is flat or falling, you already have the problem — AI is just about to accelerate it.
- Have one explicit conversation with your CFO about pricing year-one seats as training, not delivery. You do not have to win the argument this quarter. You have to start it.
The organisations that come out of this decade with strong senior benches will be the ones that made a deliberate decision to protect the pipeline when the incentives were pointing the other way. The ones that drift through it will be fine until about 2029, at which point they will not be.
Sources: CIO on Gartner forecast (10/03/2026) · BCG, AI Will Reshape More Jobs Than It Replaces (03/04/2026) · Workforce Impact weekly brief (14/04/2026).
Leave a Reply