Ask HN: Best practices for rapidly changing database schema?

Recent discussion has focused on using JSON blobs as the rows of "schema-less" databases (http://news.ycombinator.com/item?id=496946). Commenters note that such approaches are appropriate when the database is really just a glorified hash table, and advanced database operations like joins are not necessary.

I am developing such an application. In it, each record is a JSON object. The problem is is that as I prototype, my schema is changing several times on a daily basis. To compound the issue, the structure of the JSON object does not always correspond with the semantics I need expressed. I am considering using the adaptor pattern to abstract the JSON object using a class. I am also considering some sort of versioning system, to keep track of schema changes, but I don't know what.

Can people propose (or link to) best practices for agile development and prototyping when there is a rapidly changing database schema?