This is what I have seen as well, enormously bloated codebases, with perhaps 5-10x as much code as there should be. Also really whacky implementations, things

This is what I have seen as well, enormously bloated codebases, with perhaps 5-10x as much code as there should be. Also really whacky implementations, things like every page load, sucking up a bunch of MD files, inserting them into an SQL database, and then querying that database to render the page, so every single page load is a total reconstruction of the database. That’s just one example of 10 I could give right now. Vibe coding might get you a janky zero to one, but the problem is that the price that you pay for that speed, is something so tangled, so messy, that the LLMs themselves cannot even get you to “two” I actually like LLM coding and I find it helpful in many cases, but vibe coding (meaning where people don’t really understand the implementation and just hit accept) generates huge amounts of very low quality code, which isn’t just a matter of aesthetics, but has real product quality and security implications