[pathdependent]

Ignoring what other users have said (bloat, speed, noise, etc), I find that my code in an editor like ST2/TextMate/VIM ends up being substantially better and more maintainable than code written in a full-blow IDE. The full-blow IDE may mask code smell by making it easy to push on when things get ugly.

For a trivial example, if you can't remember the name of some function you wrote in another module, it may be the case that the function was poorly named or the code you are currently writing is trying to do too many things. In an full-blown IDE, it doesn't feel wrong; in a simple text editor, it actually hurts. (Yes, ST2 has autocomplete, but IIRC, it's based on a tokenizer, not a parser. In either case, I operate with it disabled.)

IMHO, it's a trade-off. It requires marginally more mental effort to compose code without the features of a full-blown IDE. This marginal increase in effort may slow me down considerably in the beginning to middle parts of a project compared to someone using a full-blown IDE; however, by the time I "catch up," my code will probably be more conducive to safe expansion and maintenance, and for all but trivial projects, that's the majority of development time.

[Flow]

I find that with a good IDE I can regard every name as a temporary name till I feel I clearly know what the best name would be.

A bit like a sketcher could draw a few lines and see where that leads.

[msbarnett]

That's less a property of IDEs than a side-effect of working with extremely static languages that are amenable to that kind of global-rename analysis.