Ask HN: Patterns for vibe coding you found useful?
When coding entire projects with LLMs, have you found any coding techniques that improved the quality of LLM output?
I've coded a dozen of micro projects using Claude models, actually mind-blown how fast you can test your ideas, in much less time. While remaining very usable, the quality seems to fall off after ~10 KLOC (much earlier if the entire project is a single file JS app). Then I need to somewhat start paying attention.
So it got me thinking, should I code the same way as on any codebase where I'd be working with a team, or is there something that helps LLMs scale, while the codebase remains grokkable? Vibing can be fine for demos, exploring ideas and such. Absolutely not for production code in larger projects you expect to scale. You (or someone who can) actually do need to pay attention to the code there.