Raw LLM Responses

I completely agree. I've had a lot of fun playing with AI and code generation since it came out, and I'm currently doing research on how we can best use AI to speed up our development time. It's hard to put a lot of details in a reddit thread, but where I've seen it provide the most time savings is in generating code for common classes of problems (almost like generating boilerplate code for your own specific situation) and in helping reduce the time spent on reading through documentation for a framework or library you may not be familiar with. The other application that works great, but outside of code generation is using it as a "rubber ducky" to help think through problems. AI often gets things wrong. Just taking what it generates often is not correct as you're applying it to your specific system. The other big thing is that engineers know how to prompt AI to generate code base in their knowledge of the codebase. A non technical person is not going to be able to do the same thing. I believe that properly utilized AI can help entry level and junior engineers get more up to speed on the code base faster and makes it easier to hand more tasks to them. In short, people get so excited about AI because it does a decent job at creating simple apps. Having used it a lot, limitations quickly become apparent when you want to build a large system. And AI continually improving doesn't overcome that problem for the same reason that code was invented. Code provides a very precise and unambiguous way to describe a system whereas natural language has a lot of ambiguity. Even AGI super intelligence doesn't overcome that hurdle because whoever is speaking to the AI (no matter how advanced the AI is) will be using ambiguous language. Code isn't going anywhere anytime soon, nor are any engineers (including entry level and juniors)