Thread

After vibe coding for the last week, I realized I really like to code, and I hate having to watch YouTube and scroll my phone while the ai generates slop. Its a great technology and a time saver and I'm definitely thankful for it. But I need to figure out better flows with it, where I'm actually shipping quality products with it, while also not just wasting time scrolling nostr while the AI builds 10 features with 35 bugs. Its the best and the worst feelings I've had for a piece of tech ever 😂

Replies (4)

I have two days before I'm off on a vacation and I think I'll spend whatever free time i have doing some actual vibe coding: Vibing to awesome tracks while I write my own code 🎸
ABH3PO's avatar ABH3PO
After vibe coding for the last week, I realized I really like to code, and I hate having to watch YouTube and scroll my phone while the ai generates slop. Its a great technology and a time saver and I'm definitely thankful for it. But I need to figure out better flows with it, where I'm actually shipping quality products with it, while also not just wasting time scrolling nostr while the AI builds 10 features with 35 bugs. Its the best and the worst feelings I've had for a piece of tech ever 😂
View quoted note →
I like to build slowly with it, give it specific feature or even just careful scaffolding instructions. Let it do the work of a 23 year old mid-level developer who has bothered to memorize all the autistic shortcuts and framework/library syntax that will be obsolete in 5 years. Then I review the code, tell it to eliminate all the stupid exceptional slop code that is never going to get hit, and we iterate towards cleaner, modular, reusable solutions.
Yep! You'll figure it out. I burn through so many tokens just doing, "build this function like this..." more hands on than "implement this feature". Right now they still just don't seem to have the ability to process a "big picture" project. My main projects are nearly 1M lines of code, it's just not able to process anywhere near that much info when it can only read a few dozen lines of a file at a time in most agent tools right now. Right now, they are more suited for moving the pen while you watch them and tell them what to write. At least for me. Id rather spend tokens working really slowly in a positive direction, instead of having it over-confidently build entire features poorly, break things, then have to delete it all and start over myself. I just want to make me faster.