The job does not disappear. It changes shape.

The World Economic Forum's Future of Jobs Report 2025 puts a number on the next five years: 92 million roles displaced, 170 million created, a net gain of 78 million. It is a genuinely optimistic figure, and it describes an experience nobody will actually have.

Net numbers hide the interesting part. The person whose role is in the 92 million does not receive a fraction of the 170 million as compensation. What they get is a job with a different centre of gravity, usually in the same building, often with the same title. The report is clearer about this than the coverage of it was: it estimates that 39% of the skills required to do existing jobs will be transformed or out of date by 2030, and that 22% of jobs will be structurally disrupted.

What actually leaves a role

It is rarely the job. It is a layer of the job, and it is almost always the same layer: the part that is repetitive, rule-shaped, and done because somebody has to do it rather than because anybody chose it. Chasing a status update. Re-keying a number from one system into another. Producing the weekly summary that three people skim.

What that layer has in common is that it is legible. It can be described precisely enough to be handed over, which is exactly what makes it automatable and also what made it tedious. The parts that resist description, deciding what matters, judging an exception, knowing which client will react badly to a standard email, tend to stay.

The residue problem

There is a failure mode worth naming. Automate the legible half of a job badly and you do not get half a job back, you get a job made entirely of exceptions. Every case that arrives at the person is now, by definition, one the system could not handle. The easy work has been removed and the difficult work has been concentrated.

That is a real risk and it is a design decision, not an inevitability. A system that hands back only what it cannot do, with no context and no draft, has made someone's week worse while reporting a productivity gain. A system that hands back a decision with the work already assembled around it has not.

Removing the boring half of a job is only a gift if you do not leave the person holding a role made entirely of exceptions.

What this means for how you introduce it

The WEF found that 85% of employers intend to prioritise upskilling their existing workforce over the period, which is a more interesting statistic than the displacement figure. It says most organisations expect the same people to be doing the changed work. That is a bet on transition, and transitions are managed or they are not.

• —Name the layer you are handing over, precisely. If it cannot be described well enough to write down, it cannot be handed over well enough to trust.
• —Decide what comes back and in what state. A returned exception with no context is a cost transfer, not an automation.
• —Protect the apprenticeship. If juniors learned the job by doing the legible half, replace that path deliberately rather than discovering its absence in two years.
• —Change the measure at the same time. A role assessed on volume will look worse after the volume moves, even when the outcomes improved.

The honest version of the argument is not that nobody loses anything. It is that the unit of change is a task, not a person, and organisations that plan at the level of tasks tend to keep their people. The ones that plan at the level of headcount tend to find out that the judgement they removed was load-bearing.