Raw LLM Responses

I use AI in my work for an e-commerce company and I have been able to solve problems using AI, primarily in search, tagging and content generation, moderation etc., good use cases for AI, and oh Chatbot. But that's not without hallucinations, inconsistent results and lot of dev intervention. It sometimes fails to tag a bright pink dress as pink and instead will tag it as red etc., It also fails to truly do any kins of intelligent analysis etc , on large scale. And we are using Gemini and Open ai enterprises grade apis, so not like free solution - even then, I see LLM use to be quite limiting. Talking about Loveable and Replit, I have now tried to create 3 websites, one with User acess and 2 just catalog websites (didn't even nee any payment integration or buy functions) and ot failed so miserably at them! It's so frustrating to have it create a website that argg! I'd rather pay someone from Fivr or hire someone from Behance and get it done to my exact liking! After having spent days on trying to get it to spit out. Website of my liking and my specifications , which btw remember was not too fancy just a modern minimalist catalog websites, I realised it lacks the basic tools to create a website - it loses context and when you edit one sextion, it forgets the previous prompts and you end up with an either or situation. Yoi can either have a apink butron or gage it in upper right corner for example (simplifying the example Obviously). The fact that it's a LLM and not a magic knowing tool becomes clear when you ask it to edit a picture - remove just white background from an image of a bottle - before adding it to a section..it just doesn't understand what you're asking it to do. All of the above pain points can be solved for...but then I wonder if Loveable and Replit will just become a better version of a WiX or a Shopify? Regarding AGI, unless the govt is hiding their latest models from us - which is very possible - I don't see how we can get there wotj what we have