This is a specific application of "meta", but a valuable one for many people here and is probably generalizable to other areas:

Pretty much the best programmer I know (an author of a number of Boost libraries) once reduced his "secret to success" to DRY Principle.

The way I understood it that he looked for higher level of abstraction even if it was locally suboptimal (in small programs cut n' paste works fine); it might have taken longer in some instances, but made him a way better at creating abstractions.