>How do you avoid the spaghetti code

You don't. This is demonstrated by

>Claude Code CLI ... is there anyone in the world that knows how that spaghetti machine works

Having a very active, and happy user base.

Now this a simplistic view, and I think even heavily AI first dev teams still glance at the diff to make sure things aren't going too awry or ask questions. But the concept "spaghetti code" and "optimized for human understanding" are assumptions being revisited entirely in 2026.

I am not trying to get into a flame war, but I have seen a lot of complaints about Claude Code, not to mention their huge backlog of bug reports (https://github.com/anthropics/claude-code/issues) on GitHub. Just because they have lowered the bar for software quality and people have accepted it, does not mean that they are delivering quality software. I know at the very least we agree that these ideas are being re-explored in 2026, but I don't see evidence that it's a positive improvement for software development/quality. How many sites or services does one use on a daily basis that are just broken? Is that really the best our industry can aspire to?

EDIT: also, from the software engineer perspective, it is an order of magnitude more tedious to maintain software projects that lack quality, consistency, or standards. If I could measure the effect of AI vibe-coding in terms of personal pain, I would say it makes software development more painful when code is not well thought-out. There is a way to use AI responsibly, but I see more often people taking shortcuts and throwing "engineering" out the window. It's possible to ship quality with AI, but that's not the path of least resistance.