I currently approach it from the "cyclical" time-management model. I'll know that things are never finished, but what I can do is finish a sketch of a broad system, or a detailing of a smaller portion, and return to it later. Repeating that a few times gets very powerful results.

What never works is when I start building an eternal software edifice. My name is Ozymandias, king of kings...