I'm a sort a "start at the top and bottom and meet in the middle" guy. I think about the top level algorithm first, coding it in understandable chunks, calling stubs for most of the functionality.

Then I look at code reuse, and figure out what bottom chunks are already written for me, and start writing glue in between the top level and bottom level. If I'm missing low-level technology, I code that, trying to be general enough to solve this problem as well as the next couple of predicted revisions or reuses. I've written a lot of code in my life, so predicting reuse is pretty second nature. {grin}

I run my program after adding about every 10 lines of code or so, stopping to insert print statements if it doesn't run right. I never have to look at more than 5 or 10 lines to debug then. The program becomes its own test harness.

-- Randal L. Schwartz, Perl hacker