Maybe with LLM/Code Assistance this effort reduces? Since we're mostly talking mathematics here, you have well defined algorithms that don't need to be "vibed". The codegen, hopefully, is consistent.