depends on what you are trying to do. (Score 1)

If you are writing the inner loop of a graphics engine that is going to run a few hundred million times a minute, then you need to optimize from the get go. If it's only going to run once a frame, then go ahead and be sloppy at first, because it's more important to get the behaviour correct than the code running fast. After you have the behaviour correct then start thinking about how to tune your algorithm. People also tend to forget that how your data structures look also impacts performance, how large your data is and how it impacts cache thrashing makes a big difference on machines with small caches. I always combine top down with bottom up design practices to get a good balance of performance and usability. I have a saying that I use for the game industry, no one playing the game cares how good your code looks, all they care about is is the game fun and does it crash. Of course that doesn't mean that you can write unreadable code, but it puts things in perspective. Things that need to be used for a long time have different requirements than a product that has a shelf life of a couple of years. As languages change so do our coding styles, how often to we go back and look at things we wrote years before and think I can't believe I wrote it that way. We are working in an evolving industry, nothing lasts forever.