the code reviews can't be effective because the LLM does not have the tribal knowledge and product context of the change. it's just reading the code at face value