He probably means foreground garbage collection, which only occurs upon a write request from the host and there are no free erase blocks to remap the write to. This is sometimes referred to as Lazy GC. The original Intel controller and corresponding firmware were among this category. As long as the SSD had empty erase blocks, all was fine. But performance degradation could be potentially severe upon exhausting the free erase block pool. The drive would then revert to foreground GC and the writes will have high latencies.
To avoid this, simply use overprovisioning on this type of SSD. With enough overprovisioning, drives with foreground GC will remain fast while having extremely low write amplification, like 1.02 without tricks.
For SSDs with background garbage collection - also referred to as aggressive garbage collection - overprovisioning and TRIM are less important to maintain performance, but rather will decrease write amplification. SSDs of this type burn quicker through their write cycles, because of the higher write amplification. This type of SSDs may show temporary drops in performance after many (random) writes, but recover after x minutes to almost perfect performance.
However, such types of SSDs are slightly less reliable than SSDs with foreground GC, due to the higher window of opportunity the SSD is vulnerable to corruption if power would be lost at that moment.
Due to the higher write amplification, as well as the lesser (theoretical) reliability, I would prefer foreground GC over background GC. Simply apply enough overprovisioning for the job it is going to be used for.
By "apply overprovisioning", do you just mean buy a drive whose size is larger than what you need, i.e. don't use all of a drive's available space?