No, I think there is always something you can improve or add. The point, I think, is to do it carefully, though, without breaking things. Python is an example of how not to approach software evolution. Emacs and Clojure are examples of how software can be maintained over the long term without breaking backwards compatibility.

When I was young (and, ahem, stupid), I didn't care — I was all gung-ho, shoot from the hip, and if something breaks, tough. As I matured, I learned that backwards compatibility and long-term maintenance should be valued and cherished. Not just because if you're not careful, you have to fix the things that break, but also because as you start doing more and more things in your life, you have less time for each thing and you'd rather move forward rather than spend time fixing things that break.

In a field as young as software, three decades is too long for almost any software tool to remain frozen. Many new ideas might have built up — especially as the world around you (and needs) changes. You might keep writing extensions for your tool, or release periodic updates, or freeze your tool and call it “done”, and write a new tool every few years… they are all different ways to handle the same underlying mechanics.

Not to jump down your throat, but "eMacs" is hurting my eyes.

I blame sTeve jObs.

"[a program] is never finished, only abandoned"

TeX and LaTeX are pretty close to 'done'