Recently I have been working on improving the results of coding agents in my projects. Fighting with AI slop and various other issues and concerns. I've reached a method that works pretty well for me so far. Would love to hear your thoughts on the matter, what works for you? what didn't? do you agree with my approach?