Distributed TensorFlow on Any Infrastructure with Mesosphere DC/OS

This is a reasonable example of building application understanding into your cluster management system. At what point do we get departmental access to 'AI' resources? Let's say you're BigCorp and people periodically want to create a model to filter data or look for trends or something. When does employee Bob log into his machine and say "new tensor instance; import foo.xlx, bar.csv, baz.sq3 into instance; generate model;'?