At any given moment, the programmer cares about different aspects of code. One moment, they're trying to correctly structure an abstraction layer; an hour later, they may be debugging an issue that needs them to walk up and down the abstraction ladder several times. One moment, they need to understand the data flow on success path; the other, modify the overall parallelism characteristics of the program.

Cramming all these conflicting requirements and cross-cutting concerns into the same files, to be read and written directly, is what's bottlenecking us. That's why we have never-ending holy wars on what is "clean code", why our editors work on such low conceptual level, why we get languages like Haskell or Rust, packed with cutting-edge math research output. We're hitting a Pareto frontier - trying to make some things more readable and explicit makes other things harder.

The solution is to stop working on code files directly, and let software synthesize views amd editing interfaces - textual and graphical - on demand. Many perspectives on the same underlying artifact, that itself will not need to be optimized for human consumption.