What should I be using AI for?

Personally if I wanted needlessly repeating unmaintainable hallucinated code in my projects while also frying my neurons and turning my brain into mush I'd eat some shrooms, visit my own little 3072 dimensional embedding space, and do it myself - which would also save me from having to sell a kidney to cover the hallucination token costs. I suppose that means I'm not the target audience for your question?

Joking aside, the only practical use I see for it is rapid prototyping, GUI mockups, documentation/example search engine, stuff like that. As long as it doesn't touch production code. And even in that case I'd use Qwen3.6 over paying for the garbage tiers of frontier model providers where you can't even be sure which specific models prompts are routed to, or in the case of Claude accidentally triggering it into "thinking" you're training a model on its output, causing it to intentionally output wrong or misleading data (have fun trying to debug that one if/when it happens!). I suppose security audits is an area where it can actually be useful (if you manually validate its findings and do the patching yourself), but that involves sending Scam Altman & co all your data/code which given the nature of these companies and their CEOs seems like a pretty bad idea.

TL;DR: don't rely on AI too much, run local inference where possible, don't vibe code production code.

i created my app completely with AI's help in android studio, i literally didnt code a line myself.

Jesse said:

Documentation and cleaning up formatting is a couple of decent ones actually. I make a lot of bs notes that mean nothing to even myself sometimes.

They actually weren't even that great at security audits (Gemini and Kiro). They bring up a lot of stuff as high risk that actually isn't even an issue.

But I still see no real use for people outside of coding.

Click to expand...

Oh yeah documentation is definitely a valid use case. But I would disagree with cleaning up/formatting code because it's functionally the same as vibe coding in the sense that you're rolling the dice on hallucinations and potentially introducing bugs or other issues that aren't immediately obvious. But I understand that some people find it's worth the potential time savings. I guess it also depends on the type of project, but yeah.

I'm not surprised re security audits. These things don't actually think/reason. They turn tokens into embedding vectors that get stacked into a 2D matrix which in turn gets rotated and multiplied by the weight matrices, the result of which is then turned back into vectors and output tokens. Super oversimplification of course, but the point is matmul operations as they are in the LLM architecture will never "understand" the code it's looking at, it only gets increasingly better at making it look like it does. It probably takes the top models and a massive context window to even have it be useful for security audits, or any other task isn't highly focused.

candyman said:

i created my app completely with AI's help in android studio, i literally didnt code a line myself.

Click to expand...

Pure vibe coding is fun until the project becomes something you depend upon and sufficiently complex to the point where simple prompts don't fix that one bug it has or make that one change you want it to make exactly how you want it. At that point you hit a hard wall and realize that you needed to know how to actually code yourself. Not saying this will happen for your app, but it's worth pointing out there are hard limits to what you can do when you tell the AI to make you a black box and you don't understand what's actually going on inside that black box or how it works.

Hyperz said:

Pure vibe coding is fun until the project becomes something you depend upon and sufficiently complex to the point where simple prompts don't fix that one bug it has or make that one change you want it to make exactly how you want it. At that point you hit a hard wall and realize that you needed to know how to actually code yourself. Not saying this will happen for your app, but it's worth pointing out there are hard limits to what you can do when you tell the AI to make you a black box and you don't understand what's actually going on inside that black box or how it works.

Click to expand...

yeah i figured that out early on and made sure before every change i did i had a backup of the previous working copy. its easy to roll back in android studio to a working version and try again with AI

Why not learn how to code? It's easier than ever given all the tutorials out there and it'll give you complete control over your app in a way that you'll never have by relying 100% on AI. It's well worth the time investment, even if you don't think so right now because simple prompts are still working for you.

Hyperz said:

Why not learn how to code? It's easier than ever given all the tutorials out there and it'll give you complete control over your app in a way that you'll never have by relying 100% on AI. It's well worth the time investment, even if you don't think so right now because simple prompts are still working for you.

Click to expand...

i know a little now thanks to AI telling me, i got more HTML coding knowledge than android app knowledge, but as i watch what AI is doing in the app i learn certain things from it.