> How do you avoid the spaghetti code that (in my experience) most AI models spit out?

That's the neat thing—you don't.

Frontier models, especially Opus 5.5, have vastly stepped up their game in terms of code quality. Furthermore, the sheer volume and speed of code they produce has meant that humans just don't have time to read and review the code even if it were the most well-written code in the world. The frontier is now structuring your loops, graphs, and evals in such a way that agents with a fresh context review the agent-produced code looking for bugs and errors. If humans review anything, it's an agent-produced summary of the code changes.

Business people now expect speed of delivery that humans can't match. Humans are now the bottleneck; agents will step up to replace them.

I'm sorry, as someone that uses agents daily, I just don't buy it. When the whole thing collapses into a pile of junk, someone is on the hook, and customers are unhappy. I don't find this approach to engineering to be a sane way of going about things (and if I'm proven wrong, then I would expect the whole software engineering process to become fully automated and repeatable, which has not yet happened).