You end up capturing the same information - only capturing it in an Object-Relational Meta Model gets you one more level of detail and allows you to automate a bunch of code such as schema definition language and object mapping code. In my extension to GLORP library, I automatically reshape the existing classes everytime a meta model (and database connection) is opened if it detects changes.

The difference is that using the visio toy, you just end up with pictures and files. You'll still be stuck writing all the software.