Sorry, I'll try to elaborate. What you described can work _sometimes_. If a particular feature is isolated from everything else, and if the optimization step is

Sorry, I'll try to elaborate. What you described can work _sometimes_. If a particular feature is isolated from everything else, and if the optimization step is a couple of tweaks that can be done later on, sure. However, the dangers of applying this tactic indiscriminately are: 1. Version 2 is postponed, and a lot of dependencies on Version 1 appear. After a while, moving from V1 to V2 becomes a herculean effort, you have to do heart surgery while the heart is beating. 2. There may be nothing in common between V1 and V2. Implementing bubble sort gives you no insight into quick sort; you literally have to start from scratch. This may be acceptable if there's urgency in delivering a solution as fast as possible, but in my experience most companies have the time to do it right but choose not to (or don't know any better). Also from experience, it would be nice if people understood that it may not be so simple to "just optimize it later", and that it would be better to spend a little bit more time in the beginning of a project to consider how to foundation of the project will scale in the future.