This sarcastic remark is entirely misguided. Broken crap does not generate revenue; working code does. But, oftentimes, no code is as effective as broken code, and takes longer to fix. This is when you should use your judgment. “Break things” is not the same as committing faulty code, it just means that validating some of your hypotheses in a controlled environment may have a higher cost.

Just to make things clearer, I am a programmer myself, and have never been in a management position before. My overall impression is that I learn faster when I make mistakes and need to fix them. Expecting otherwise would contradict most of the experts on human learning.

Actually, exactly. I'm in the business of writing software that's useful to people. The way they tell me it's useful to them is by paying me. At any point you may determine it isn't useful to you. Just don't pay me.