A customer runs 50 notebooks by hand every morning and wants it automated. IT will not approve any new cloud service.

Two traps, and good engineers fall into the second one more often than the first. The first is proposing a managed orchestrator anyway, which ignores the only stated constraint. The second is converting the notebooks into a clean Python package, which is better code and takes the work away from the analysts who wrote it, so every future change now routes through an engineer who is not there. Candidates who ask what scheduling infrastructure the customer ALREADY runs, before proposing anything, are the ones who have done this at a real client, because riding approved infrastructure is faster than winning an approval.