For most of computing history, a programming language had one job before it could do anything else: make a machine's instructions understandable to a person. Syntax, naming conventions, clean code, and familiar abstractions are all remarkable inventions when the person is doing the writing.
The reader is changing
Large models do not approach a codebase as a human does. They do not need a neatly named variable to remember what a value means. They resolve patterns, constraints, and relationships across a graph of information. Human-first syntax adds a translation layer that a machine author may not need.
This does not make readability irrelevant. It moves readability up a level. We should be able to inspect intent, guarantees, and consequences without requiring every implementation detail to be authored by hand.
A smaller surface for a bigger system
LinkedSpec starts from that premise. It keeps the parts that communicate purpose and makes the representation negotiable. The output can change with the target environment while the intent stays stable.