I'm not a programmer but I have some stuff I want to do with python. So what I've found useful is taking the output of one LLM, Claude for instance, and giving it to another, Gemini for example, and writing "this is what claude thinks. Evaluate it..." Then if I have to I will take that result back to the first LLM and ask it to evaluate it.

It might just be in my head but it seems like they are a lot more critical that way than if I suggest something. ...Or maybe all of my suggestions really are great...