The operations associate used to build the board pack. She pulled numbers from five systems that disagreed with each other, chased down three department heads for commentary, reconciled the CRM, and handed the COO a first draft.
The work was dull and low-status. It also taught her how the company actually ran: where decisions got stuck, which numbers lied, and who signed things off.
The board pack was the output and the individual training was buried underneath it.
That education was never designed. You were not training her. You were buying cheap labour and letting the learning happen by default.
AI is going after the visible part of that job. The first draft of the board pack can now be largely automated. The meeting summary writes itself. The CRM discrepancies are flagged sooner.
The conclusion looks obvious. If the software produces the work, why keep the junior?
There’s also a tempting financial reason behind it. A manager can see the saving straight away in this quarter’s budget.
That is why the first response is often to cut or pause hiring.
But the task list was never the whole job. Around every draft sat a layer of chasing, checking, sense-making and preparing decisions that the software has not made disappear.
In fact, AI is moving the bottleneck.
The COO used to wait two days for that first draft. Now he gets five plausible versions within minutes and has to decide which narrative is true and which number will not withstand a board member’s questioning.
The constraint has moved from producing the work to judging it, and judgement does not scale the way generation does.
It is the same shift everywhere the apprenticeship used to hide under production. The senior lawyer no longer waits for a junior’s draft; she gets several and has to find the one clause that turns the deal.
The audit partner no longer waits for a graduate to trace every transaction. The anomalies arrive pre-flagged, and the work becomes deciding which are fraudulent and which are strange but legitimate.
The machine produces faster than ever, and every output lands on the desk of the most experienced and most time-poor person in the room. Now the scarce resource is the senior attention needed to judge the work.
Speak to any executive using AI, and they’ll confirm that context-switching fatigue is already real.
This is where the junior role has to be rebuilt, and not as an “AI reviewer” bolted to the end of the process, as if judgement were something you delegate downwards and quality-check at the finish.
The useful junior reduces the load on senior judgement rather than feeding it more raw material.
For the operations associate, that means she still uses AI to assemble the board pack, then does the human work the machine cannot. She finds the number that does not reconcile and chases down why. She notices the decision that has been open for three meetings. She walks into the COO’s office with the two questions that actually need answering rather than the forty pages that do not.
None of this is glamorous, but it teaches her how decisions actually get made.
That learning no longer happens on its own. The old role generated it for free through sheer repetition, and AI has removed the repetition. So the loop has to be rebuilt on purpose, and rebuilding it is the manager’s job.
That means reviewing the thinking, not just the output: what did you ask the tool to do, what context did you give it, which answer did you reject, and what did you verify yourself?
Those four questions, asked every week, are now the entire apprenticeship. A manager who only marks the output is training nobody.
If firms do not redesign roles, juniors get AI tools and produce more work than they fully understand. Managers get more work than they can properly review, and when mistakes happen, the firm blames the graduates for relying too much on AI.
Some of that criticism may be fair. Using AI well is not the same as having good judgement. A junior who produces confident but wrong work quickly is a real problem.
But the firm helped create that problem. It kept an old job, added new tools, failed to teach how to think with them, and then called it a talent issue.
This is why I do not think junior jobs will disappear. And the AI-enabled junior job will look very different from what many firms are building now: the same old role with one more tool.
The work is shifting from producing the first draft to making it useful. From preparing documents to preparing decisions. From doing the task to reducing the load on the person who has to judge it.
That is the real argument. Junior jobs are safe only if companies redesign them around the work AI has made more important.
The fair objection is that this still does not prove firms need as many juniors as before.
You probably do not. If the role was built on producing and formatting, it shrinks, and in places headcount shrinks with it.
Fewer juniors does not mean no juniors. It means juniors need to move from doing the basic work to making the faster work useful.
That is the choice. Firms can use AI to eliminate the old junior role, or to design a better one.
The first option saves money quickly. The second builds a role that is useful now: closer to decisions, closer to customers, closer to the messy work that machines cannot resolve.
Junior jobs will not disappear because of AI. But the work nobody misses is gone for good, and what remains has to be designed properly.