First, programs and databases will grow together continuously. The missing piece of the puzzle here is that currently most people view databases as a single correct representation of the world, and assume that whether you have an acid database or a nosql solution, all transactions must eventually be reconciled, and a single truth must emerge. This is patently false. In actuality, database systems will begin to resemble distributed version control systems very closely over the next five to eight years, until they are almost indistinguishable. The two will meet in the middle as all distributed revision control systems end up using off the shelf graph database systems, and these graph database systems incorporate additional functionality to support one them. At some point most editors will be capable of speaking directly to the database systems without file intermediates, and if we need to transmit program data from one place to another, it will be via cloning/replication, or a textual dump and restore in some (rare) circumstances. Make no mistake about it, the key-value and column based architecture of current nosql solutions is going to transition to graph databases in the not to distant future - the algorithms to make efficient use of graph databases aren't quite there yet but it is getting very close. The idea of a single consistent world view is going to shift to various contexts and sub-contexts where consistency need not hold throughout the hierarchy (exactly in the manner of forks and branches). We will share data by establishing contextual nesting and hierarchies of information primacy.
Files are an artifact of early computing architecture, and holding on to them beyond their useful lifetime will only anchor us.