This experience may be relevant: Investigators at other schools have
sometimes written inexperienced faculty members at our institution
into their applications with the understanding, "You don't have to
write anything.  When the grant is made, we'll just expect you to
execute the work described in the application."  Without exception these
projects have floundered because the PI who does not writing lacks
emotional commitment to the project.  For example, it's not true
he or she doesn't have to write anything.  She must supply Other
Support information, prepare a biographical sketch, and shepherd the
application through our grant processing, contracting, and regulatory]
procedures; later she must actually do the work and write up the results.
A faculty member who writes nothing almost alwasy underestimates the
amount of time the experimentation and associated duties will consume.

People who advocate grant writers sometimes seem to believe the
scientist's job is to conceive "The Idea," and that words and numbers
are a cosmetic that make the idea look pretty, and can be applied by
mere functionaries like grant writers.  But the successful scientists
I know without exception regard writing as a SCIENTIFIC skill in which
they take pride (even if they find it distasteful); and that The Idea
has no existence independent of the words and numbers that express it;
thus the only person who can write a grant for endocrinologist A is
endocrinologist B, and then it ceases to be A's application.

Development offices sometimes hire grant writers, and such persons
can indeed write applications for research "infrastructure" or "development"
or even training; but anyone who's hired to conduct experiments ought
to be able to write her own applications.

EDITING is another story. Many applications could be improved by skilled
editing.  But most of the requests fro grant writers I've encountered come
from people whose idea is no crisper than "I want to look at enzyme
X in patients with prostatic hyperplasia."  They need help, as you
suggest, from another scientist, not from a wordsmith.