The tasks most amenable to automation are routine, well-defined and verifiable. Those are also the tasks organisations traditionally gave to new hires, because they are how someone learns the work.
The training problem
Expertise is built by doing large amounts of unremarkable work until the patterns become intuitive. A junior lawyer reading hundreds of contracts, a junior analyst building dozens of models, a junior developer writing routine code.
Automate that and the output is produced faster. The learning that used to be a by-product does not happen.
Nobody planned this and the consequence is a pipeline problem that appears years later, when the senior people who would have been trained do not exist.
What organisations are trying
Some are keeping junior roles and changing the work – juniors review and correct machine output rather than producing it from scratch. That builds judgement, which is arguably the more valuable skill, and it skips the fluency that judgement usually rests on.
Others have cut junior hiring and are discovering the consequence has a delay of several years, which is longer than most planning horizons.
What this means for someone starting out
The path in is narrower and the skills that matter are different. Being able to produce competent routine output is worth less. Being able to evaluate output, spot what is wrong, and understand why is worth more.
That is harder to learn without the foundation, which is the whole difficulty.
The honest position
This is an unsolved problem being handled badly by most organisations, because the cost lands on a future manager rather than a current budget.
Anyone starting a career should be deliberate about seeking work that builds judgement, since it will not arrive automatically.