Single threaded, NumPy offers basically the same syntax and speed as Fortran when it comes to vector operations. Therefore, the slides boil down to "CoArrays being the Fortran killer application".

I don't feel this is the whole truth. You can (and many people do) write a modern Fortran MPI-CoArray application for HPC and it can be slim, compile and run fast. No big ecosystem, like a small C application. The same is most likely not true if you include Python in your toolchain and then use some FFI to call C or Fortran in performance critical sections. As the author, you might be able to hide the ugliness (such as Numpy does it), but it can be a show stopper for maintainability and future generations working with the code. That's probably why still so many new HPC codes are pure Fortran/C codebases and Julia or Python remains the language for postprocessing data on workstations.

Author here. I spoke about Python and Fortran to an association of chemical engineers a few nights ago. Some folks here may find the slides interesting.

Maybe at one time fortran had the most mature libraries for simulations but is that still true?