The value lies in the design space exploration, which is what an LLM can easily do.
[1] https://browser.geekbench.com/processors/snapdragon-x2-elite...
[2] https://browser.geekbench.com/macs/macbook-pro-14-inch-2026-...
Except the problem is not restricted to the actual ISA or its HDL implementation, etc.
It's even just getting space / time in a fab at that advanced of a process node.