Nostr Web Client

What do you call it when you use ai to assist in coding but don't just "vibe" with whatever it says until you get the result you want? What if you're the kind that explains a problem, asks for potential issues and then actually read the response and follow it with "But if I add this part isn't that just redundant since I already do that in this section of the code?" And then it goes, "you're right, adding that would be redundant." But it did also show you why and how the original bug was a problem and then you fix it yourself instead of using the half slop artifact?

I'm that kind of coder I guess. half vibes, half PoW. Perfectly balanced right?

Reply to this note

Please Login to reply.

Discussion

Eric FJ 🪬⚡️ 7mo ago

This is the way.

ภ๏รtг๏ภคยt 7mo ago

Depends if your paying for the Ai or not.

If not, you have an assistant.

If paying for the Ai, you have a worker.

w3ird_ 7mo ago

I treat AI like a really new junior dev. it rarely does what I want the first time. however I've started making it give me plans as documents and then have it implement from the document. like, design a database schema with uml diagrams and then write the migrations and such. create the openapi spec before creating the API. it works decently well while also making sure there is good documentation for future me.

OceanSlim 7mo ago

I use it similarly. I often use it to probe ideas and make outlines before I begin working. Like I have an app idea. I'll give it my requirements and ask like what kind of data structures COULD I use and what should I actually use for this and why. And tell me the reason I would and wouldn't use the other types. then I make an informed decision on how to proceed myself. Without AI I would be doing the same thing except I'd have to read a lot more documentation.

w3ird_ 7mo ago

this is pretty much what I do too. glad to hear I'm not crazy! what about languages? do you think it does better writing JavaScript vs golang, for example?

OceanSlim 7mo ago

It's pretty good at most languages I find. the smaller you can make the task, the better results you'll get. I find myself mostly saying, "no, don't add that dependency" or "this code is redundant" when it's wrong. It's barely ever a syntactic failure. Generally speaking though, I find the more adoption the language has, the better the ai is with that language. So it's probably a lot better with javascript and python than it is with golang. But with good direction, it's just fine with go. But you can do lower level things with go, so you have to actually understand data structures and algorithms before you implement something that just doesn't work well for the intended purpose.