That depends. If you know that your data contains a highly irregular, non-exhaustive list of extra attributes that have no relations to other parts of the data model, a JSON column might be the right choice. It's rarely the best choice, but I can totally see applications where it is.

I use JSON for data that needs structure, but doesn't need to be relational. That data is then returned with the rest of the row in a SELECT. I will use JSONB for the same reason, and I also need to query that data once in awhile.

JSONB is also very useful for HTTP JSON APIs because you can have the database construct the JSON response and the API server just returns it. I found it to be faster than constructing the JSON response in the API server most of the time. Of course, it depends on the use case whether you can do that.