However beyond that using subroutines gets you into the following paradox. Suppose that working with heavily factored and organized code is 5 times harder per page of code. Suppose that it does the work of 20 pages of unfactored and unorganized code. If you assume that the issue of tracking down the same bugs in unfactored code is similar to the issue of changes made for one reason affecting things all over the place, then that is a gain in productivity of a factor of 4. However whenever picking up that code I can guarantee you that you will feel that factor of 5 issue.

As for the assumption, that is where experience comes in. It is easy to break things into a million subroutines. What you learn over time is how to break things into a million subroutines that organize into clumps which present loosely coupled interfaces to the rest of the world. After you do that you just do not modify existing public interfaces lightly. Add a parameter, sure. But don't lighly assume that nothing depends on old (even old and stupid) behaviour.

Also another good idea - the test suite - makes this easier by giving you a good idea what behaviour someone else is depending on somewhere which you didn't remember.