The reason you are seeing vibe code projects getting abandoned everywhere is because vibe coding is _grueling, never-ending labour_, like all software development. It is not fun. It is work. And the less foundational knowledge and experience you have about software development, testing, systems architecture, symbolic logic, and DevOps, the faster it becomes work, and the faster the project becomes abandoned, or melts down into a buggy mess.

It seems fun.

At the beginning.

Reply to this note

Please Login to reply.

Discussion

It's constant labor, that also costs you money to actually produce something that at least half way working.

Like going to work, only you pay your employer.

😬

Experienced software engineer, when they hear someone is getting into vibe-coding

👏 there’s a big difference between bringing an idea to life, and the load that takes to review, maintain, take design decisions, .. all this requires experience and knowledge and it’s only fun when you have these skills/background.

I don't find it fun, even then. I have been in software development for nearly thirty years and have already enjoyed it less than cleaning the bathrooms.

I just tolerate superhuman levels of suffering. 😂

Around the same here, and despite all the pain and patience it brings to me some kind of comfort to see how i learn to do things more solid over time

I agree: vibe doesn't replace the labor. The key is what you commit to after the initial excitement fades. If you approach vibe coding with a dual commitment -- to ship and to learn -- it absolutely works out.

Bitcoin Calendar is vibe coded, but it wouldn't be useful to anyone without that persistent, non-glamorous work. You can't produce value without putting the work in.

AI is not simplifying things (its their false promise)

Software development has become more complex.

So when you add these things together, it’s just a very abstract layer on top of a lot of complexitity. And only people can handle this as they are able to build a mental modal around it. And that mental load is what counts and makes you competent or not.

Or something like that.

Yeah, I now need a model of the code base that also includes the proper way to interact with the LLM. I think gen-coders don't notice this, until the project reaches a certain size and they have to actively babysit the AI and explain seemingly obvious stuff to it, for hours.

At the beginning, the issues are more generic and the AI does it all automagically.

By coding day 55, the dev is typing in all caps. 😂

And they go in de mode

JUST ONE MORE PROMPT TO FIX IT

Reviewing AI generated code - it's a travesty that ignorance of the underlying technology on the part of the "viber" is both hidden from themself and also so obvious in a PR.

Reviews increasingly become about social skills in deconstructing ego instead of debating technical implementations on their merits. Find myself lacking the former.

Learning anything new is painful and reading documentation is necessary in know what questions to ask; to others, to yourself or to an AI.

nostr:nevent1qvzqqqqqqypzplfq3m5v3u5r0q9f255fdeyz8nyac6lagssx8zy4wugxjs8ajf7pqqs0fx489zv7kcdfekjupm4lxz5kgenu9cln72d9c3u40cjgzlul3zqy5z6xh

It's all down hill after printing Hello World!