Microsegmentierung kann als Strategie helfen. Das heißt Probleme so sehr in kleine Pakete zu teilen, sodass der Code klar abgegrenzt ist. Das hilft die Gewalt drüber zu behalten. CLI Systeme wie Claude Code zum Beispiel, übertreibt sehr gerne in der Implementierung und man verliert sehr schnell den Überblick. Man braucht ein wenig Zeit, um sich eine Roadmap bereitzulegen, damit die negativen Aspekte kompensiert werden.

Schneller lernen durch gezieltere Erklärungen ist auch mein Fazit. Und Refactoring nur bei kleinen, abgegrenzten Bereichen. 

Why I enjoy being a programmer....on top of fixing cusomters problems and inventing how to implement their business needs is that....I like to write the code ;). Of course sometimes I do use AI to generate some boilerplate starting point, but then I just adjust and detail it with my own hands for my needs. Also it works for me as a more handy Stack Overflow :). So maybe I would need to change into being a plumber at some point, but I will never become a vibe coder, I just like to write the code :). English ain't a programming language yet ;). But not every use case is good for AI, at least not now, since it can't handle huge code bases and stuff that AI companies could not feed into LLM from the internet. So in the project I am currently in, that is based on completly custom framework that has no information on the internet, it is very much useless. It is not that I am very much against AI, but I am not an evangelist or hurra optimist in this case, I will not doom and gloom that programming will end in 1 year because AI will replace us soon. There is still plenty of cases where AI is useless, even in our field and believe me, when investors finally asks for the profits, 95% of those AI vibe companies will just die down.