Correct, and again, I’m someone who uses AI, I’m not one of the ones who are purely against it. The definition what what I’ve both done and seen:
Using LLMs to code - fine. It’s a tool, use it as you’re saying. This is not what I’m talking about.
Using LLMs to vomit out all of your code into one commit is vibecoding. It’s clearly vibecoding when there weren’t even multiple commits. No trial and error, no refinement, just one giant dump.
I do care how it’s coded because it shows how well it will be maintained into the future. If someone uses LLMs over months to build something it will probably be maintained. If someone vomited out all of that code in one vibecoded session, it’s very likely they’ll be bored and in a month it’ll be abandoned. That’s what I’m looking at, because tests can be automated now.
Maybe I’ll be wrong and in a month they’ll have committed hundreds of times and are trying to mature it. I have no evidence of that though, so there is no reason I’d risk trying it out.
Using LLMs to vomit out all of your code into one commit is vibecoding. It’s clearly vibecoding when there weren’t even multiple commits. No trial and error, no refinement, just one giant dump.
Yeah, I completely agree there. Just like with regular coding, you have to go slowly, implement functionality in small focused chunks, review, and iterate.
And I’d also wait and see whether the project continues to be maintained before considering it. I just posted it because I thought it was neat, and might be worth watching to see how it develops, not that I’d use it over a more mature option at the moment.
I think we’re actually in complete agreement here.
I don’t think you’re in the wrong to say that code nobody has reviewed has not proven safe, and a brand new project (of any kind) is not proven reliable. A brand new big project that’s purely vibe coded is every red flag of unsafe and unsustainable.
Correct, and again, I’m someone who uses AI, I’m not one of the ones who are purely against it. The definition what what I’ve both done and seen:
I do care how it’s coded because it shows how well it will be maintained into the future. If someone uses LLMs over months to build something it will probably be maintained. If someone vomited out all of that code in one vibecoded session, it’s very likely they’ll be bored and in a month it’ll be abandoned. That’s what I’m looking at, because tests can be automated now.
Maybe I’ll be wrong and in a month they’ll have committed hundreds of times and are trying to mature it. I have no evidence of that though, so there is no reason I’d risk trying it out.
Yeah, I completely agree there. Just like with regular coding, you have to go slowly, implement functionality in small focused chunks, review, and iterate.
And I’d also wait and see whether the project continues to be maintained before considering it. I just posted it because I thought it was neat, and might be worth watching to see how it develops, not that I’d use it over a more mature option at the moment.
I think we’re actually in complete agreement here.
I don’t think you’re in the wrong to say that code nobody has reviewed has not proven safe, and a brand new project (of any kind) is not proven reliable. A brand new big project that’s purely vibe coded is every red flag of unsafe and unsustainable.