I made a similar argument several years ago, after being part of Ward Cunningham's Fit project (a precursor to tools like Cucumber): "The Problems With Acceptance Testing." http://www.jamesshore.com/Blog/The-Problems-With-Acceptance-...