> But the bottleneck is somewhere else entirely. It's thinking in terms of models. Formalizing and abstracting what we know intuitively about the world and about the things we want to automate.

This is a bottleneck but one that exists even without programming.

In my opinion, the biggest bottlenecks are two fold. Software frameworks are coded to be coded against and the abstractions we use to communicate between sub-systems in programming languages require us to code (parameterized sub-routines).

Software frameworks that are composition centric (you hook up parts instead of calling parameterized sub-routines) make programming a lot easier. You don't even need to code against composite centric frameworks: you can just hook things up using general purpose VPLs.

Parameterized sub-routines lead to way too much specialization of interface making software frameworks both difficult to use and not composition centric. This is why we have to code against our software frameworks. I also feel that parameterized sub-routines are the reason why "there has been no progress since the 1970s".