I agree with you, basically, but that is the substance of my complaint. Git leaves all of its messy internal bits exposed and does not provide a clean abstraction layer in its interface. Many people have written such abstraction layers, but they tend to be incomplete, and none of them is a standard. Everyone who wants to use git - which includes everyone who wants to merely collaborate with other people who use git - therefore has to learn git's internal implementation details.
It's one thing to learn the internal implementation details when you are a power user who really cares about a particular tool and wants to master it: it's another thing to require everyone to gain low-level familiarity.
The whole point of a computer - indeed, the whole point of computing, if you look at it through a lambda-calculus lens - is to provide abstractions over lower-level details, thereby allowing human brains to spend less time worrying about those details and more time manipulating higher-level abstract concepts.
In failing to offer a high-level interface, git wastes human brainpower which could otherwise be spent solving actual problems.
You don't need to know what an AST or a symbol table or an intermediate representation are to use gcc; you don't need to understand static single-assignment form, linker relocation entries, or ELF segments. You can use the tool without any knowledge of its internal mechanisms. This is not true of Git, and that is why I hope it can be improved or replaced with something that has more respect for human brain power.