Of course we won't 1-shot a new CAD kernel but the human work will likely shift more towards identifying failure modes, specifying test cases and sometimes solving really hard details. Sometimes even working out details that have no precedent in literature or free code.