Don't fund software that doesn't exist

This is an interesting piece by Andreas Mueller (one of the main developers of the popular scikit-learn project):

He goes over some results from surveys of the success rate for grant-funded software. Seems particularly relevant since we’ve just received some funding for the maintenance of GMT from NSF. Basically, they found that most software that starts as a result of a grant don’t get used. And even projects funded to transition from a small group to community-developed don’t achieve their goals. Some things for us to think about and lessons to learn since this is the exact goal we have in mind for this new grant.