If AI writes the code, who tests it?

AI is writing code faster than ever. The question nobody's answering well: who makes sure it actually works?