I disagree, their work cannot be important enough to justify them getting away with straight slander.
This is the second time I’ve seen them speak publicly and the second time they’ve caused a major drama.
The first time was with the alleged Ravencrest situation where the leaders (astral and another) were accused of abusing their custom magics to get ahead in the clan leaderboard.
Astral’s response boiled down to “what are you gonna do about it”
Yeah fair enough but it’s Vetex we’re talking about here that can do punishments, we can’t expect him to demote Astralworks. At best I’m just hoping that he loses something that he benefits from, such as a custom magic or losing tester access.
Never gonna happen. Vetex will never punish astral. I’d put money on it.
Oh I agree but in any professional environment astral would have been fired the first time they did this bullshit.
Like an accusation like that could be career ruining.
Agreed, especially when it’s from a developer of a popular game. It doesn’t matter if said developer’s reputation is bad considering majority doesn’t even know that and they won’t bother to do research aside from taking their words at face value because of their position.
yeah like, with everything that’s happened recently in the roblox space in general I am getting really really fucking sick of people making accusations and then acting like the resulting damage was an accident.
how is vibecoding bad again..? From what I know it’s a great tool to use for programmers which makes the entire coding process easier
Vibecoding is bad because it generates the code outright for you instead of assisting, AI should be a tool, not the creator.
If you rely on vibecoding, your own skills would gradually decrease until you become so reliant on it that you can’t do anything but vibecode.
AI code agents can generate code, but never fully the design. Vibecoding for rapid prototyping is fine if you ask me, but if you ship that without human oversight you’ll see that you’re secure database is actually open to the wild. It is still important that a human should design the underlying architecture and structure of an app
Dumb question but is astral actually a she i hear 50% of the community call them a he and the others call them a she. Which is it?
I’m genuinely baffled that Astral, despite acting like an asshole and causing drama, is still on the team.
Having asshole players is one thing, but when one of your devs says shit to the point that it sparks drama surrounding another dev, that’s a red flag they should fuck off from this game and never come back.
There are a lot of factors in terms of codebase architecture and system documentation that you need for vibecoding (though for the case I’m describing, it’s more like AI agent engineering) to actually be efficient and reliable.
It can very much be a significant boost in coding power (5-10x or more) but it takes over a week of proper onboarding of every pre-existing system for it to properly persist, as AI agents have only so much context they can be aware of at once. It’s a significant initial investment that requires planning and human expertise to properly cultivate.
Plus, AI won’t always account for things you don’t know, and it also isn’t fully updated on very new stuff as well for local coding environments as there it relies purely on the model’s knowledge and existing context it reads.
Source: I’ve used Claude Code and Codex for about a month and a half now and onboarded it almost fully with my codebase, having various skills and commands integrated to make things efficient and optimized for implenentation and consistent coding style.
In short, vibecoding is only really a bad thing if you do not properly on-board or know what you’re working with/trying to do from an architecture perspective
Does it matter? I opt we settle on a Voldemort style way of referring to them. I think it’s a good idea to start with the OGs and modify if needed before making more.
They Who Shall Not Be Named
You Know Who (OG works here)
overall, regardless of how good or bad it actually is, its AI so everyone hates it and its an extremely bad look.
nobody wants to hire a vibecoder, so accusing somebody of being one could be seriously damaging.
We’re talking about Roblox coding here, AIs have so little knowledge about it that they frequently get things wrong, and then come most of the developers; who have exactly 0.5% knowledge on the things you’ve stated.
Vibecoding is okay, but you have to be experienced to really use it effectively.
ChatGPT’s pretty informed about Roblox Luau. Claude kinda is but ChatGPT generally is a little better trained for it.
It just falls short for new APIs like QueryDescendants() unless you explicitly tell it how it works.
So yeah your point about experience is true because it gives you the ability to catch those little things it may not do optimally/in a way that’s modern.
ai doesn’t understand code
it thinks it does but the code it spews out outright doesnt work, and when it does theres usually a major oversight
for example:
“yo ai fix this line of code”
“done”
“you just made the entire code a comment, dont do that”
“done”
“now you just removed the code entirely, dont do that”
“done”
“okay so what did you do this time”
“nothing like you told me to”
“i told you to fix it not to do nothing”
“done”
client initiated a disconnect
that, along with the general consensus of ‘dont use ai to do creativity related works for you’ = ai is bad when used for coding
This tends to happen when the AI doesnt have the context necessary, or the prompt you gave leaves a lot of room for ambiguity.
The scenario prompting you gave is like the worst possible case of trying to get what you want, without any real guidance aside from just the prompt you gave it. “Fix this bug” → fix it how? Fix what? What is the intended behavior? You need the discipline to find out and communicate what you actually want, each and every time, or you get nonsense and waste time. You also need discipline to update context docs each time it gets something wrong so that it will be less likely to make the same mistake later.
Also, I prevent that by making sure that my prompt always includes how something should be done (if possible), and I tend to only use the advanced models like Sonnet/GPT-5.4 or better since they are actually thorough. Also, having a main file that they read for context on the codebase and style is also extremely important to prevent this exact failure.
And when you do the best practices for AI agency, it is extremely powerful and codes as well as professional programmers, with the consistency of a robot (assuming you do your part in writing clear, well-written prompts for each request).
It is a skill to be built.
Considering I’ve witnessed how poorly written some prompts were, I understand this too well 
(and providing context to a scenario + maybe the debug menu if there is one is probably a good thing, but I wouldn’t know that considering I can’t code anything)