Most of the writing about AI and jobs is either a forecast or a threat. Neither is much use if you run an organisation and have to decide what to put in a job description next quarter. The practical changes are smaller and stranger than the forecasts, and they show up in role design long before they show up in headcount.
The shape of a junior role is changing first
Junior work has traditionally been a mix of two things: the tasks nobody senior wants to do, and the repetition that builds judgement. AI is very good at the first category. That is a problem, because the first category is how people used to earn their way into the second.
Teams that handle this well are deliberate about it. They let the system do the volume and then put juniors on the review seat, where they see a hundred cases a week instead of ten and have to decide which ones are wrong. That builds judgement faster than doing the work by hand ever did, but it only happens if someone designs for it. Left alone, the junior role just gets hollowed out.
Review becomes a named job, not a favour
When a system produces work, somebody has to be accountable for it. In most organisations that accountability is initially unassigned, which means it lands on whoever is most conscientious. That person quietly absorbs a second job and burns out in about two quarters.
If an AI system produces output that reaches a customer, name the person who owns its quality, put it in their objectives, and give them the time. Review work is real work. Treating it as something people do between other things is the single most common staffing mistake we see.
The scarce skill is specification
The people who get the most out of these systems are not the ones with the best prompts. They are the ones who can state precisely what a good outcome looks like, including the edge cases, before anything gets built. That skill used to be distributed thinly across product managers, analysts and senior operators. It is now the bottleneck.
This is good news for hiring, because specification is teachable and it is visible in a work sample. Ask a candidate to write the acceptance criteria for something ambiguous. You will learn more in twenty minutes than from any number of questions about tools.
Spans of control get wider, and that has a cost
When a team ships more with the same people, the natural move is to widen the manager span. It works for a while. What breaks is coaching, because coaching is the first thing a stretched manager drops, and it is the thing that was quietly producing your next generation of seniors.
If you widen spans, fund coaching somewhere else explicitly. Otherwise you are trading a visible efficiency gain for an invisible succession problem that arrives in about three years.
Four questions worth asking this quarter
- Which of our junior roles are now mostly review, and have we said so out loud?
- Who owns the quality of each AI-produced output, by name?
- Where is specification being done well, and who is doing it?
- If spans have widened, what replaced the coaching that used to happen?
None of these require a restructure. They require deciding what the work actually is now, and writing it down before the organisation decides by default.