Capacity Planning
Deciding in advance how much capacity will be needed, given growth, seasonality and failure scenarios, and ensuring it can be there in time.
Autoscaling has not removed the need for this, it has changed its shape. What still requires planning: the things that do not autoscale (database primaries, licences, third-party quotas, IP address space), the things that scale too slowly for a spike, and the account-level quotas that cap the autoscaler.
A workable method: establish the current headroom by load testing to saturation, project demand including seasonality and known events, size for peak with a failure rather than for average, and identify the lead-time items — a quota increase, a reserved-capacity purchase or a schema migration cannot be done on the day.
The output is a date: at the current growth rate, the first constraint is reached in month N. That turns capacity from an anxiety into a scheduled piece of work.