Probably a significant number of HN readers have done something like this at one point. IIRC there is a MIT course where you do a modern Unix clone in one semester.
There is an ocean between that and being an expert in Linux internals.
Oh, and the original UNIX was done in one week by one guy.
I have not contributed much to Linux -- my claim to fame is submitting a one-line patch which generated 50+ emails of arguments from LKML before Linus merged it -- but I have read a ton of the Linux source code and familiarized myself with many of its internals many times, as well as doing extensive low-level systems work in userspace against the Linux API/ABI. I often use it as a reference in my own osdev work, or working on Hare, etc. Have read a lot of the syscall API surface, DRM internals in depth, dcache and several filesystem implementations, io_uring, etc. Not ignorant to what would be involved in making a Linux-compatible kernel.
What this brings to light is the ability to compare Helios (microkernel) and Unix clone development, or how it was said in the post: reimplementing already existing design vs designing something new.
Because of this, I think his statements carry some weight.
There is an ocean between that and being an expert in Linux internals.
Oh, and the original UNIX was done in one week by one guy.