Yes, that point is true - it's easier to go from the stricter environment to the more open one - I guess that's basically just how entropy works. Note that I don't necessarily disagree with the article, I just wanted to offer some perspective on why I think the schemaless approach can often make sense, especially in how CouchDB handles it. If at some future point, the worlds of RDBMS and NoSQL fold into one DBMS handling all the use cases perfectly, I'm all for it - it's just that right now I'm still skeptical about going all schemaless on an RDBMS, since it's still a rather new feature, and the underlying system is very complex and has grown for decades.