Ant, a lightweight JavaScript runtime
5 points by loige
5 points by loige
I would normally find this very interesting, but what I saw when I looked into this project doesn't build confidence.
The project is described as "hand-built." I think the author means that the engine is a new one and not an existing one. But it is clearly vibe coded (along with the website) which does not match most people's definition of "hand-built" any more than my IKEA furniture does. It's vibe coded to an extent that the agent seemingly lifted large, recognizable amounts of code from another project which looks like a license washing attempt.
The author also shows a fair amount of bravado about how they "built a javascript runtime in one month." Reading this article, it does not come across at all that the author used an LLM rather than writing it. In fact I think they went to some length to hide this.
I don't know what the price of this was. Probably sleep. Probably health. Probably a lot of time I could have spent doing literally anything else.
Probably tokens?
Not to paint this project with a scarlet letter, but vibe coded projects are a dime a dozen and don't indicate quality without further evidence of actual effort, and the behavior of the author seems to exaggerate their effort already.
This bit from the issue you linked made me laugh loudly:
How did I find this out? I just asked LLM "what is the origin of this source code?" — and just like that, it told me "hey, that's pretty much verbatim https://github.com/cesanta/elk plus some additional functions".If you scan newer repositories on github in a similar way you will find lots of license violations.
This bit from the issue you linked made me laugh loudly
In a good way or a bad way? 🥺
The idea of asking the LLM to show what was plagiarized struck me funny. Not sure I'd call that "in a good way" ... it struck me funny because, if that worked, couldn't we just ask them to not plagiarize shit in the first place?
The idea of asking the LLM to show what was plagiarized struck me funny. Not sure I'd call that "in a good way" ... it struck me funny because, if that worked, couldn't we just ask them to not plagiarize shit in the first place?
True, that in a way actually proves that the llm knew the elk code well enough to recognize it, and it could prob reproduce memorized code, which was the very thing that happened there.
As acknowledged before, back in sept of 2025 an agent reproduced large portions of Elk in an early version. This was unacceptable, regardless of whether it was intentional.
The entire overlapping implementation was removed, alongside the design and rewrite around a bytecode VM, removed all pre-bytecode release binaries. The current JIT is based on an explicitly identified and licensed fork of MIR.
To distinguish that failure from an intentional "license-washing attempt.", I must say that I have no interest in concealing someone else’s work or relicensing it as mine.
If you believe material from Elk, or any other unattributed project. If anything remains in the current tree, please point me to the files or passages, and I will remove or attribute anything that should not be there.
The fact that it was unintentional plagiarism honestly isn't that reassuring, and doesn't really negate the overall point I was making, that there's no clear signal of the individual thought and effort you have personally put into this project.
I don't mean to make this personal but I think it's very interesting that when the agent does something you consider good, it's "I built a JS runtime in one month." But when it plagiarizes, "the agent reproduced large portions of Elk..." Is there a double standard?
Heres a better timeline for reference:
Have you reviewed the current codebase before concluding that it shows no individual thought or effort? Criticism of the early version is justified, but it does not necessarily describe the project as it exists today.
No, typically for projects being shared online, I'm not going in and extensively reviewing their code first. I read your blog and concluded that this wasn't likely a good place for me to spend any more effort. I hope you'll understand, there are a lot of slop projects being shared these days, including JavaScript runtimes, and your website is very similar to what they all look like.
If you want to earn trust I hope you'll take some of the feedback you're getting here.
I wonder if there are any other popular (maybe even universally known) tools already using the name ant? No way to know.
It's also a little weird to use furry-inspired art for something vibecoded. Furries are famously protective of human-made art and reject slop.
I wonder if there are any other popular (maybe even universally known) tools already using the name ant? No way to know.
not to argue but that tool is not in web development/javascript adjacent land?
It's also a little weird to use furry-inspired art for something vibecoded. Furries are famously protective of human-made art and reject slop.
I'm using coding agents to accelerate implementation, not to replace understanding or review. The architecture, technical decisions, testing, and final responsibility are still mine, so I don't think "vibecoded" is an accurate description here.
As for the artwork, it isn’t AI-generated, it was commissioned.
I'm using coding agents to accelerate implementation, not to replace understanding or review. The architecture, technical decisions, testing, and final responsibility are still mine, so I don't think "vibecoded" is an accurate description here.
When there was actually an instance of very clear, probably unintentional, plagiarism in your project, how can you really argue with a straight face that the decisions are all yours? Clearly those ones were not. The decisions are being made by the agent and you're, hopefully, looking at them to some extent and then approving them. This is very different from making decisions.