Composable programs are easier to reason about. We may (confidentally) determine program behaviour by determining the behaviour of sub-programs -> fewer bugs.
Composable programs scale indefinitely, by composing more and more sub-programs.
There is no distinction between a "small" and a "large" application; only "smaller than" or "greater than". [2]
[2] This is why we cannot (refusal) answer the ever common question, "Are there any large programs written using Functional Programming Language X?" It would be dishonest to answer it!