Haphazard schema is the quickest way to develop terrible performance, loss of referential integrity, and insane queries. Well, outside of sticking everything into a JSON column.
I have occasionally stuffed json into proprietary software data fields to get that data accessible/usable by another system.
(High performance not required)
And some closed proprietary software does things like adds a few additional fields for pragmatic end user extensibility like this.
The practice predates JSON, but sometimes your bespoke string or ID or whatever in field "Custom 1" is all compromise you need to make things work well.
It works fantastically well, and don't let anyone tell you that you MUST bloat the CLI interface of your program with every possible dial or lever it contains. We should all be cogent of the fact that, in this very young and rapidly evolving profession, textbook and real-world best practice often do not overlap, and are converging and diverging all the time.