The Problem with “Vibe Coding”
37 points by jmmv
37 points by jmmv
I think he really hits it on the head, as is often the case with any of his works, all of his talks have not only aged incredibly well, but are works of art in their own regard.
Are people claiming that vibe coding is the same as producing a product? Almost everyone I see using the term has precisely the same feelings about the code produced as the author has, i.e. that it’s a hacky prototype, not a product. Even in Andrej Karpathy’s tweet which appears to have coined the term he shares that view
It’s not too bad for throwaway weekend projects, but still quite amusing
According to Gary Tan, a quarter of YC’s current batch is products that are mostly vibe-coded. Also, vibe coding
is the dominant way to code. And if you are not doing it, you might just be left behind.
I take this to mean that investors think vibe-coding is the future because it will help them get to market faster, and they don’t really care about technical debt. I can’t imagine it’s much fun to maintain one of those products programs, though.
Startups are selling to YC/Gary Tan, who are in there turn selling the story that GenAI is the future to their investors. The actual market accepting the eventual product is a distant third in this chain.
I take this to mean that investors think vibe-coding is the future because it will help them get to market faster, and they don’t really care about technical debt. I can’t imagine it’s much fun to maintain one of those products programs, though.
I agree with this 100%, but I think this is how startups have always worked. At my day job, I work at an ex-startup that made it to the big leagues, and the product is riddled with these sort of “worse is better” design decisions that were made just to get to market faster. And my job is primarily whittling down this tech debt accumulated over the years.
I think that the future is (mostly) vibe-coding for startups and (mostly) hand-coding for enterprise companies.
this is how startups have always worked.
Oh, for sure. In the hands of capitalists, AI isn’t fundamentally changing the dynamics, it’s just accelerating the processes that are already there. From their perspective, it hasn’t ever made sense to care about technical debt.
I think this post is right in the wrong way. This is exactly the end game of vibe coding - we want to build products, not programs, and “vibe coding” promises to remove the need to program so we can focus on what we’re actually trying to do at the business layer.
It’s nowhere near that point yet, but that’s why people are excited about it.
Do we? For a lot of programmers, especially outside of webdev, product is a mean to have food on the table while they create programs. Programming is fun!
It depends. Sometimes I like programming, but for some tasks I just want to get the script done so I can move to the fun bits.
And of course, for non-programmere, often programming isn’t fun at all, or they care about it less than the thing they’re building. Look at the popularity of GameMaker or other stuff.
Similarly, I’m sure a lot of people love drawing. But sometimes I just want a drawing of an OC or something, and for that purpose just prompting an image generator would be much easier than spending months learning how to draw.
The only thing that LLM/Vibe coding has demonstrated is that it can work at the scale of small programs that do one thing and don’t need to be polished. Small tools, prototypes, experiments. But as the post says that is a very different kind of task than developing a product. Generating code is not the hard part of product development. It never has been. Proving that LLM’s can do that is a much harder ask and anyone saying it’s obvious that they’ll get there is making a claim based on faith not evidence. They might be right but I’ll reserve the right to be extremely skeptical about it.