Live data from Hacker News

Vibe Coding Is Overrated

transitions.substack.com

11–20 of 93 posts

Re: Vibe Coding Is Overrated

#11
post #3

Vibe coding is definitely problematic for someone who isn't a strong coder, aka me. Recently for fun, I used ChatGPT to code a python script to scrape another news site and render it in the style of HackerNews. It took 21 iterations of code updates to get a working version. It isn't bad but doesn't pull the latest articles (I don't know why) and hell if I could debug it.

With experience and ADHD I learned to think in units that can be debugged separately. The experience tells me where those lines should be drawn, and the ADHD taught me to avoid creating units that are too big to fit in my head.

Re: Vibe Coding Is Overrated

#12
post #7
post #6

You must be doing it wrong since Anthropic's CEO has claimed that in 12 months, nearly all code will be written by AI.

Wasn't this 12 months ago?

It was two months ago, and the prediction was 90% of code written by AI in 3-6 months, and ~100% in 12 months. Maybe he will be right but in a cursed monkey's paw sense where the amount of high quality code stays about the same, but that becomes a rounding error compared to the amount of barely functional spaghetti code being churned out by agents.

Similarly ~100% of written word will probably be penned by LLMs soon if not already, but that doesn't necessarily mean the writing is any good, only that LLMs can type really fast for pico-pennies on the dollar.

Re: Vibe Coding Is Overrated

#16
post #10

I totally agree, most of the time the LLM is printing part of the training data. And what we tend to see are cherry-picked samples.

The other day I managed to dork the clouformation template that my llm was basing its code on from github(similar class structures) and got a better result working from that code. Also the way it will edit version files versus using the cli like with cargo/package.json is concerning, especially because its examples take no account of depreciation

Re: Vibe Coding Is Overrated

#17
I would guess that for every app that was vibe coded and made $80k in revenue, there are thousands of attempts that went off a cliff or into a wall, generating $0 revenue and wasting a human’s time. So it’s just survivorship bias. Sure, it’s possible to make something useful, but you will probably fail.

I’m using AI to increase my productivity, but whenever I’ve vibe coded (not intentionally, just by getting caught up in the dumb vibes) I’ve regretted it. I’ve ended up with a tangled mess.

Re: Vibe Coding Is Overrated

#18
There are parallels to the “rapid application development” push of the 90’s. Visual Basic, Tcl/Tk, Python with Tkinter, HyperCard; all of them promised shortening the development cycle and democratizing computing. Code was interpreted rather than compiled, dynamically typed rather than statically, and a lot of batteries were included.

It sorta worked, and sorta didn’t. I’m seeing no evidence that this round is different. LLMs allow coding via natural language and assuming a lot of context that is typical to human conversation, but a lot of coding is delving into nuance, which is going to be work, no matter the tool.

Re: Vibe Coding Is Overrated

#19
I have already vibe-coded countless number of scripts and two side projects that were on my "list" for years, these projects would have never seen the light of day if not for Cursor. That's the whole point.

Re: Vibe Coding Is Overrated

#20
post #19

I have already vibe-coded countless number of scripts and two side projects that were on my "list" for years, these projects would have never seen the light of day if not for Cursor. That's the whole point.

Is it. For small scripts like these you can get far with basic search and sources like docs/github/SO. Maybe cursor is more “fun”, “engaging”, and the like, but it’s not like it was a difficult job before.
Post reply on HN