Raw LLM Responses

In order to build complex projects with a lot of files and a lot of code with AI, you have to break it down into small sections/tasks, then spoon feed the AI. Claude from my experience is the only one that works best because they have a project knowledge section where you can store files. But you really need to have an understanding of the project and cannot rely on the AI knowing everything and building everything. it also has limited memory and will make minor mistakes. At times there can be a small issue that's causing the code not to work and it won't catch it. If you present a problem with your code that has 20 possible solutions but only 1 right solution, it may only identify 5 possible solutions and continually recommend them in a loop. SO the videos of people saying the built a full stack project using nothing but AI in 30 minutes, just know they're doing to for views, likes and subscribers. And another limitation with AI is the chat limitations. Even if you buy a paid subscription, you're limited in how much you can chat on a daily/monthly basis. At a minimum, you will at least want to understand some parts of the language you're using, so you won't waste tokens trying to ask the AI to do something and depend on it to figure out what you're trying to do. Even if it does build it and it works, it may not be scalable or even customizable because of how the code was written.