[javajosh]

>The article doesn't provide any good example of misuse.

It did describe the misuse pretty well, though. The idea is that out-of-band schema modifications are a process/workflow issue that needs to be directly addressed. As stated by OP, this is an easy way for anomalies to creep in - what if the already present table has different columns than the one in the migration? IF EXISTS lets a migration succeed but leaves the schema in a bad state. This is an example of where you would prefer a migration to "fail fast".