This is so applicable to our industry. I heard Chuck Moore say once that most microcoded division routines are bad, because it's the first and only time that engineer writes one. He said his tenth division routine was far better than his first.
It concerns me that in the rush to build abstraction upon abstraction, we're neglecting the tools and practices that make it possible to work on and improve the fundamental layers.
Considering that one of the current trends is the development of unikernels (like MirageOS), network stacks in userspace, etc, I'd say we're still thinking about the lower layers.
It concerns me that in the rush to build abstraction upon abstraction, we're neglecting the tools and practices that make it possible to work on and improve the fundamental layers.