Do you let LLMs write both your tests and your code?

I'm curious what stance others have taken? Or what they've experienced with this? I'm toying with adding a rule to our values.md that the LLM can only significantly generate one or the other.