I'm a scientific programmer (not a ton of experience but understand history). It's changed a lot over time but it feels like nothing was much better than writin

I'm a scientific programmer (not a ton of experience but understand history). It's changed a lot over time but it feels like nothing was much better than writing Fortran / bindings to Fortran until very recently (of course, take my recency bias into account). Yet, still, significant amount of "critical code" are written in Fortran. It's a slow-moving world in part because there is a lot of corner case complexity / significant logic. Parallelism in the form of CUDA / transparent GPU via Python is changing that a bit, but some algorithms are still not parallelizable.