vibe coding is too binary of a term and is causing confusion. The way I’m starting to look at it is agentic coding with different levels of human-in-the-loop. Sometimes you want to be in the loop (pair/symbiotic programming) sometimes you don’t (vibe coding) vibe coding isn’t necessarily bad, its just bad for certain problems: ones where you should be in the loop. And yes i still think humans should still be in the loop. For now.

Replies (19)

I don’t think he’s tried the latest models. Things are changing rapidly. They are good enough to pair program with now and have just become a really good tool. Maybe if he used them he could finish his game in less than ten years.
Yeah, I remained highly skeptical until I tried Opus 4.5 and Sonnet 4.5 It still struggles with complex logic, but it’s incredible what it is capable of now
no, it would still be him programming. He would just be able to do it faster, but agents on lots of simple tasks like having an army of junior devs. He can stay on the hard stuff (this is how I work now) He just needs to open his mind on how this can accelerate dev instead of strawman attacking the idea that ai is coding everything, which is just naive.
the axiom's avatar
the axiom 1 week ago
I don't think he is completely against LLMs, mostly against the hype if you see this take you'll see that there are many types of approaches to LLM coding but in the current context if you say anything good about AI you'll immediately be lumped together with the most unhinged vibecoders like steve yegge and if you say anything against it you'll be seen as a luddite from your own comments I get the impression sometimes that you are a luddite, others that you're a booster
I think my skepticism and hype have been equally balanced on here. When i am disappointed i say so, when i am excited i share that. the latest models paired with claude code have tilted me toward hype after playing with them recently
the axiom's avatar
the axiom 1 week ago
looks good, I've done similar things, better than trying to guide a retarded junior you did read it afterwards, right? that's the kind of usage that I think makes sense, but it offers very marginal gains compared to all the hype boosting
no its no longer “this tool barely works” Its now: “How do i scale my workflows to apply agents to solve my 1000s of github issues while i work on the hard stuff”
yes i review it over course pair this with the fact it can be doing this in a loop over a large list of simple issues, it committing along the way. i think it could be a huge boost to real software maintenance burdens
the axiom's avatar
the axiom 1 week ago
except you'll get git conflicts all over the place people should try to vcs with pijul instead
Justin Moon's avatar
Justin Moon 1 week ago
Yea but vibecoding is such a charismatic term it's hard to use any other lol
> because they're not like alphazero, they just spit words This seems to be changing right now: here's a paper on recent, roughly speaking, AlphaZero-like research that specifically uses coding problems. They make it learning on experience rather than on traditional datasets.
the axiom's avatar
the axiom 1 week ago
I don't know how you found this but it's definitely interesting, maybe this time human programmers will become unnecessary