I think (and this isn't just personal experience) most of the appeal of Jupyter is the fractioned persistent state that allows you to write incrementally but no

I think (and this isn't just personal experience) most of the appeal of Jupyter is the fractioned persistent state that allows you to write incrementally but not re-run increasingly long scripts. Inline plots are just gravy. (Yes yes write documents export LaTeX. Not sure people care about this at all) I've been developing a style where development "cells" are cached with joblib.Memory so re-running a script from command line doesn't take idiotic amounts of time. Further, I have litte context managers such that with Figure('filename.png'): plotting_logic() exports a file that immediately shows in the folder navigation bar of Atom. It's overall a better experience. What I would really like is being able to use joblib.Memory not as a decorator (which requires bunching up your "cells" in functions and calling them separately) but as a context manager. --- To the tune of the "Portlandia" song: the dream of MUMPS state is alive in joblib...