It’s fairly common for developers that run one time jobs. These one time jobs usually need to persist outputs, e.g. mapreduce writing intermediate outputs.
An interesting way to deal with this issue is to offer a datastore or a remote fs that is compatible with the language built in file operations, like fs.appendFile or fs.readFile.
With this you get a way to persist files from any machine, but also the ability to use easy file api primitives.