There's an argument to be made that replicating existing software is extremely useful. Most of the time when you're writing a compiler for a new language, you'

There's an argument to be made that replicating existing software is extremely useful. Most of the time when you're writing a compiler for a new language, you'll be doing things that have been done before. Because most of the concepts in your language are brought along from somewhere else. That said: I'd always want a compiler and language designs to be well considered. Ideally, the authors have some proofs of soundness in their heads. Perhaps LLM will make formal verification more feasible (from a cost perspective) and then our mind about what reliable software is might change.