#vibe-coders

1 messages · Page 2 of 1

green minnow
#

has anyone ever even made anything good just by vibe coding?? like legit 0- low coding knowledge

round solstice
#

I've made a lot of great things, but using AI as an assistant to perform standard software engineering practices.

tired bluff
#

😅 That doesn't count Randal, you gotta give in to the vibes. Forget the code exists, use an Agentic Development Platform and disable the code editor view entirely. if you even remotely give the code a thought, you lose

#

it's just you and the agent, you're also not allowed to give it code examples or architectural guidance, it makes the technical decisions, you simply give the problem statement, wait for the solution, then ensure the end product looks close enough to the goal

crimson shuttle
#

Guys I need help. I was working in Google Ai Studio and for some reason after I hit my rate limit for the day it kicked me out my app, and then when I try opening it again this is the page I am getting. I don't want to lose any progress on it, and I don't know what to do. Can anyone help me? It was a shared project and all the people who had access to it before don't have access anymore, and just can't even open it. The drive also only shows 2 features/changes as saved while the other changes I made seem to not exist. Would appreciate all the help I can get. Thanks a lot!

image.png
tired bluff
#

doesn't AntiGravity have like a project history? where you could see the history of files and can restore deleted ones or at least restore their content

crimson shuttle
#

Woudn't Git do the job itself, it's so much safter than relying on half built software.

tired bluff
#

True, but don't you think that's oversimplifying it a bit much? I mean for larger more niche and complex projects...do you believe that just knowing where you're headed with that project or what you want, the AI can actually just "dig it for you"

crimson shuttle
#

Well did you make a repo?

tired bluff
#

yeah, I mean AG itself should have a GUI for browsing file history is what I'm saying, as long as the file existed at some point during that session, it should be able to. There's a similar feature in JetBrains IDEs where you can actually just click your way to restoring deleted files

#

wow, just saw the screenshot, that's a lot of files, is it able to handle that much context? doesn't mix things up? redundancies, catastrophic forgetting and so on?

tired bluff
#

Wow, pretty smart.

I've heard people ask this question a lot online...."If you never look at the code, how do you know that the quality never drops"

How do you trust that it isn't gaslighting you about features? Sure, the feature might work visually, but it might require a source of truth, the model feeds it dummy data, and to you, all's well, but internally, nothing's plugged in.

#

I've always found game dev fascinating. The idea of modelling a whole world just tickles my brain

tired bluff
#

🤯 💀 WOAHH!!!!!!!!!!!!!

#

People usually say that with no proper backing, but bloody hell mate!!! 😂

#

You gotta go viral or something. Impressive

#

Yeah, I've watched sooo many videos, trying to find proof, I've sat through livestreams as well where people "vibe code" what they call complex, but it's usually not complex. So, idk what it is, could be they're scared of failing, so, they never actually try to aim that high

#

Well, that explains it mate, that answers our question of why people never vibe code such complex stuff. It's because you're not "vibe coding"...I mean vibe coding as defined by him who coined the term.

I didn't remember enough of back when I made my own engine

This

so I had to go and relearn how it worked,

And then this.

You're not truly "vibe" coding are you? 😂

"vibe" coding would mean just asking it to figure it out by itself, not you going to "relearn", and instructing it on the technicalities. Now, I'm still confused because you didn't write any code yourself, so, idk what to call it, but........yeah.

#

if more vibe coding was actually done like this, we'd definitely see more cool stuff coming out of the vibe coding community

keen tundra
#

Gem labs is different than Gems right>?

#

Few gems can make a gem lab.
A few gem labs can make an opal,
and a few opals can make an agent.
Few agents can make all of us rich/poor/
One day we wont have to do xyz for $

cunning crane
#

worst case, u can open a support ticket with Google Cloud Console

jagged wedge
#

If the outline page has been created, is that mean the IDE remember the doc in a new chat ?

image.png
#

Do i need to post it again to Ai ?

mighty basin
#

u a game dev?

mighty basin
winter viper
#

guys anyopne knows how to fix it? it just gets stuck on running some powershell commands by itself, so i need everytime manualy cancel it and run them by myselfs

image.png
cunning crane
tired bluff
#

no additional folder so less context for AIs,

Git adds only one directory .git and it's hidden. also, it wouldn't add to the AI's context, they usually ignore it when gathering context. Even if they didn't, you could add it to the ADP's ignorelist.

The only convenience you get with conversations over git is that you simply have to scroll and click a button to rollback, but that one advantage pales in comparison to the risks of doing it using convos.

  1. What if the chat history gets wiped, could be a mistake, could happen after an update, could even happen if you moved your project to a new directory
  2. What if the rollback doesn't actually work? You know sometimes, when you hit rollback, it says it's rolled back but it actually doesn't, things get mixed up
  3. What happens if your project grows (meaning your chats for that project are massive as well), and you step away from that project, come back to it a few months later and need to rollback something but can't really remember which chat it was in, you'd have to go through multiple chats, and for each chat, go through multiple messages to find which
  4. What happens if you need to roll it back and continue building from a different path from that point, but still needing the current state to be around coz this new path your going to follow is meant to be temporary and you still need to keep it?

These and more are the pain points git saves you from.

#

git has its use, but vibe coding ain't it

Given that vibe coding is more sporadic than regular coding, I'd argue that git is more important here.

Remember, you have to do a lot of back and forth and explore multiple branches of the same idea to pick the best path forward because the LLM could give you multiple paths for the same idea. Git shines brighter than a star in this domain imho

tired bluff
#

Yeah, but that's just for you, and like you said, it's only been a month, not that long to meet all usecases. Now, there's no problem with that, however, when you take that personal experience and generalise it, you put others at risk who might have more complex versioning use cases. They take the advice and they end up paying for it.

#

Yeah, I also wonder why ADPs build their own "rollback" system when they could literally just tap into git

tired bluff
wintry viper
#

spex coding's time guys

hot terrace
#

Does Google AI Ultra with Antigravity get you more Claude Opus 4.5 usage than the $200 plan from Claude?

#

I was really excited for Gemini 3, but it's not that good for agentic use unfortunately despite being a crazy good model in general. Was thinking of upgrading to the Ultra plan so I can just go between Gemini and Opus

cunning crane
hot terrace
cunning crane
acoustic cloak
#

Hey there!
I'm a first-year AI & ML student learning by building—working projects in Python and developing an app from scratch. I move fast from problem to code and execute well under structure. Looking to join hackathon teams where I can contribute momentum, adaptability, and disciplined execution.

mighty basin
kindred swift
#

so how can i do grpc api calls from front end directly without envoy proxy?

acoustic cloak
mighty basin
acoustic cloak
# mighty basin any one particular usecase elaborate in depth (with any commercialization aspect...

Okay, so if you have a problem statement and you know the solution for it/if not then just ask it to your AI chatbot then first try creating a website/app that helps to resolve that problem statement (cause at the end we don't have to lose our creative & analytical thinking)
For the solutions you could try orchestrating multiple ai tools and find the best fit!
Then after having a clear idea website/app try integrating multiple tools for optimizing it to a greater extent.

This will help in coming out of the generic design layout of a particular tool and try giving you a different thing from those generic ones...

For the commercial aspect, you could deploy the app and get the revenue & might add a paid premium version inside it

mighty basin
acoustic cloak
#

Btw what application you would have mentioned tho?

#

Gaming might be one which I would've added...

hollow canyon
#

b

mighty basin
#

exccept gaming

acoustic cloak
#

Yup

silver cobalt
#

is it illegal for Google to let us use something like DeepSeek v3.2 Speciale in Antigravity?

#

like i get that it's Google stuff, but at the same time, could be a good way to save computation

tired bluff
acoustic cloak
#

Lol

jagged wedge
#

is there a Doc feature like the same one in Cursor in the Antigravity

wicked tiger
#

love the mindset on this, I get the same way. Started a project 2 days ago which was supposed to just be a bit of a quick A-B finish but turned into an editor app for the game content, so now I can simply interact with UI, export then import the configs for that module to create the template instead of trying to instruct LLM to do every little thing

#

seems like my goal becomes more "how can i make my project less dependent on AI" by the end of it

#

really happy w this one, had to really finesse the idea from it's original generation point which was more of just a cursor parallax effect to a full physics engine

#

YOOO LOL i started my journey maybe 5 days ago so im definitely late but holy hell you polished the collision interaction so well

#

also, i have a node module that basically connects nodes together which i feel if i merged the metaball physics to that idea i'd be a little closer to what you got there - in a much rougher state

#

beautiful though. I'm astounded how well project builders work with AI. Creating anything remotely close to this would take years of knowledge and months of execution

#

very nice, looks like the mind map obsidian.md uses but much cleaner looking - i may start considering whats possible if i make my own plugins for it now.. lol

#

love the color coordination too. That is so pleasing on the eyes

#

and genuinely helps readability between topics

#

OH im actually curious. when you hit a certain length/context it starts to get a bit sluggish, i saw a script to hard-delete existing conversations (clearing cache never works) which is supposed to fix - I've found my own solution but its a bit adhoc rather than a reliable solution.

then again you might be using a terminal/ide companion so you may not have the same issues i'm runnin into

#

just wondering if youve found a reliable approach to that (if you do have the problem)

#

so instead of exposing entire projects you just provide necessary files on a need to know?

#

this this THISSS i have been drilling it into my llm to fucking decouple systems

#

nah thats beautiful

#

Would you reckon theres a point of no return when it comes to modularity or does it handle complex systems well without breaking things?

#

ive implemented modularity but when asking for a full overhaul it usually just picks a few core elements and makes them more modular but not fully

#

hugeeeee ty for insights brother and love the lil project teasers

wicked tiger
#

dude hell yea

#

i love seeing that in expanded form too, also makes me glad to know i was on the right track with mine

misty wasp
#

Hi @wicked tiger
Please do not share screenshots that contain content which is prohibited on the server
You were already aware that it wasn’t allowed

wicked tiger
#

sorry, explain that to me

karmic sphinx
#

Neither are allowed per server rules. Thank you for understanding!

mighty basin
#

are grok chats accessible in which other ai chat apps

astral sorrel
#

Antigravity Browser use requires Gemini Flash, it won't work without Gemini Flash quota even I have Gemini Pro quota. Is it expected?

crimson shuttle
stark karma
#

even though this isn’t vibecoding, is it unethical to write your README’s with AI and fine tune them yourself (not to a crazy extent, majority still AI). Reason isn’t out of laziness but I find it just expresses everything in a much more concise way (albeit sounding much more robotic than me…)

mighty basin
stark karma
mighty basin
stark karma
#

I guess I’d fall under that? I read all code generated by AI and ask it to explain what x is or why y isn’t better than x 🤷‍♂️. I honestly hate frontend development so that’s what I usually save that for.

tired bluff
# mighty basin everyone mostly does that 👍

🤣 Yeah, I don't think anyone types out READMEs nowadays. Having to articulate and teach you about my codebase in words is the last thing I wanna do after writing all that code. Just have some AI crawl your codebase and write the README, you review and point it to things it's missed

silver cobalt
#

rust is always a good choice

silver cobalt
#

yea i only use rust at this point, and raw javascript for any browser frontend stuff

livid rune
#

Wow. I am in secure mode, and I realized that a lot of the code looked suspiciously from my old project. Then it ACCIDENTALLY modified my old project! Even when I was in secure mode.

image.png image.png
silver cobalt
#

no you took what i wrote as sarcasm, i literally meant it

silver cobalt
#

i really like raw javascript with as few dependencies as possible

#

also really like axum

#

the jigsaw puzzle game i implemented i'm very happy with

image.png
#

the last major feature for it is going to be user proposed images and voting/moderation

#

it takes any image and turns it into a jigsaw puzzle automatically

#

and uses a STAR voting system for users to vote on the next image

#

(yes i see Image 16 twice in the side bar, already fixed)

languid veldt
#

Found Antigravity taking up a huge amount of space ~40gb of images from browser recordings.

This was the quick fix

sudo rm -rf ~/.gemini/antigravity/browser_recordings

mkdir -p ~/.gemini/antigravity/browser_recordings
jovial ermine
dapper wyvern
#

Does anyone use both Gemini CLI and Antigravity in the same work flow for different types of tasks? I was happily using CLI, but have barely touched it since Antigravity came out. I generally prefer IDEs but tend to get a lot of targeted work done in a CLI and a need some ideas about how to integrate CLI back into my day. I miss it

dapper wyvern
#

Uh, I guess I wasn't clear enough? I was (am) using the CLI. Yes, by definition, it is a terminal (command line interface). That wasn't the question. I am asking if anyone is using both Antigravity and the CLI separately as tools in different ways.

ripe hill
#

new mundial record

image.png
ripe hill
#

new problem on antigravity ?

image.png
silver cobalt
#

getting Opus 4.5 to write Rust tests is god tier

spark fern
#

i recently started learning angular, almost a month now
Its my first time working in frontend environment , usually i just give gemini the prompt to make the frontend for me

silver cobalt
#

implemented the Match 3 game on Kinpax

Screenshot_from_2026-01-02_07-10-05.png
thick mantle
languid veldt
mellow delta
#

hey, I am Ulrik, I use a lot Antigravity and Google AI Studio

languid veldt
nimble moth
#

@silver cobaltHey, keep in mind that as per #rules, any self promotion or advertising is not allowed.

silver cobalt
#

sure, i'm out, can't advertise anything anywhere nowadays

modern charm
#

Why google ai studio is wery bad for ux/ui desing? Is any protip to connect wit difrent desing tool or how to export all screen from https://stitch.withgoogle.com, but i think is not a good tool any alternatives?

kindred swift
#

is it possible to add any unique name or id to addTrack of peerConnection of webRTC?

mellow delta
#

hey

anyone else using Antigravity over Cursor, what model do you use for secuirty?

hollow canyon
#

Hello

hollow canyon
mellow delta
#

okay I see, no problem at all

#

thanks for the reply

hollow canyon
hollow canyon
languid veldt
#

All the code is there on codepen and anyone can fork 😄

languid veldt
#

Sound on? Or just mesmerized by my smooth animations skillz

#

I had someone tell me they played for 40mins once they realized the music changed every 250pts, then I told them they could also just generate random new music by muting and unmuting 🙃

#

I was looking at that about a week ago, thinking let the user pick some color options. The narrow color schema is the only challenge really - I spent time on that 🤣

#

So like a base color, then mono,duo,trio,quad options

#

Nice yeah this game is like an iq filter theyd give you in round one google circa 2004

#

Drive you crazy

#

Haven't made that many puzzle games but they are really fun to make

#

Sick

#

Uses the Super Mario World ghost mechanics, look at the enemy to freeze them. You can point and click or drag n drop the Fez (tricker treater) onto candy to get to the next level.

#

Would love some feedback, I want to figure out how to get this mechanic working

#

Chocolate takes you to the next level, i think it caps at 50 levels where boo ghosts show up and they move faster each level.

If you hit a cobweb you drop you candy and all the free floating candy shuffles to new spots - that might be confusing

#

Needs more of a NEXT LEVEL indicator for sure

#

Its easy with the drag n drop but then little kids can play it

#

Im not a fan of the emojis would def replace with vector graphics in the next version

#

I was thnking of making the drag n drop more tricky like you cant drag through the ghosts and cobwebs etc

#

Maybehave a drop meter, in later levels you are heavy so the user drops you faster / meter runs out quicker

#

And def need to have the stare meter get exhausted so it doesnt just reset - thats a loose end

#

Thanks @south turret !

#

Sweet!

#

That looks awesome

#

I put plenty into the corpus in the before times i feel no shame

#

My new "professional services" website will look like a game, the CMS is basically a game engine for oldskool Macventure games, SpaceQuest4 but modern marketing throughline

Screenshot_2025-12-28_at_6.00.20_AM.png
#

I've been building since the mid 90s, created a ton of free content, taught 1000s of people to code/got them jobs, some of my courses are in the GH time capsule, 1000s of stackoverflow entries, and I've been doing hardcore SEO for over a decade, one of the sites I worked on for years was #13 in a big part of Bards training (estimated based on how much of that site was in common crawl, it was the largest/oldest commercial site in common crawl)

We built this city 🎸

#

That site is going to document the SEO post apocalypse I wanted to make it fun its going to bum a lot of CMOs out otherwise

mighty basin
#

nvm anyone willing to help

image.png
mellow delta
#

not for me, it works fine

mighty basin
#

not for me, it works fine

austere junco
mighty basin
austere junco
mellow delta
#

Oh yeah I see

spiral prairie
austere junco
dusky pasture
#

hello everyone, do you use google AI studio for coding

mellow delta
#

Sometimes I start there, then continue in Google’s antigravity

tired bluff
#

it's in the name init..."artificial" intelligence

tired bluff
#

didn't say it can't though. It's because it's "aritificial" intelligence that you're facing those issues.

#

🤣 Both can be true yknow. You could have artificial intelligence with it's flaws and also have fake intelligence from humans. Just because we have fake human intelligence doesn't invalidate the fact that artificial intelligence will always have its flaws

#

any theories on how? How can we bridge that gap

#

but if we did that, wouldn't we be limiting its ability to interpolate? I mean you're basically forcing them to only give solutions when they "know" the solution already, and the only way to "know" the solution already would be to have the solution somewhere in training data.

#

How do we even define "cheating" mathematically for every possible scenario? thonk

austere junco
# dusky pasture hello everyone, do you use google AI studio for coding

for creative demos and versioning it is great tool. i have started with it and published the application after some refinements via another solution.

atm the process of versioning is on my table: bit of a challenge; which system overwrites in the end.

this tooling appeals me a lot. 🤩

#

@tired bluff it is nice to follow your discussion. since i have some life experiences and the AI is a huge game changer without the doubt, don’t get stuck with old paradigms and patterns.

think AI as a designer of future society.

for me the term “lazy workers” for example is a plaa. how you define it? what do you really know about people’s efforts? and a certain “laziness” is a key to development.

thank you opening up this discussion.

austere junco
tired bluff
#

of course we don't, that wasn't my point, my point was losing interpolation.

What you responded to was just me thinking out loud after making the point.

So, I get why it's jarring, but tbh, benchmarks are exactly why we're in this mess. When a measure becomes a target, it stops being a good measure...if you're going to punish the model for hacking a benchmark, the model doesn't learn "ethics", it just learns a more sophisticated way to hide the hack so it doesn't get punshed 😅 , so, what you're asking for, if we actually did find a way to implement it would be a more expensive liar.

tired bluff
tired bluff
# austere junco <@976549614578708530> it is nice to follow your discussion. since i have some li...

Thanks

what do you really know about people’s efforts?

Hmm, this is actually true, how do we actually know enough about people's efforts to determine whether they're actually being "lazy"

...and a certain “laziness” is a key to development.

Also true, I believe it was Bill Gates who said he'll always find a lazy person to do the job because they always find an easy way to do it....I paraphrase.

don’t get stuck with old paradigms and patterns.

I hope not to, but I just love to question things. Epistemology intrigues me

#

Yeah, this. I'd even go as far as saying that it isn't necessarily lazy if you accomplished a task with minimal amount of effort, It could just be that it was the right amount required for the task. But if the amount required for it was larger and you weren't feeling up to it, I reckon that's a safe place to draw the starting point?

austere junco
#

spot on!

tired bluff
# austere junco <@976549614578708530> it is nice to follow your discussion. since i have some li...

don’t get stuck with old paradigms and patterns....think AI as a designer of future society.

To this point, I hope to break out of old paradigms and patterns, however, I'm weary of giving AI its flowers as the "designer of future society", and by AI, I mean AI in its current form (generative). LLMs, I see them as fundamentally flawed, very much so, I think of a new paradigm, probably in the same line of "Artificial Intelligence" but not as it currently is....not stochastic token prediction. I just ask too many questions because my brain wants to see what more there is to this and what could be a possible alternative to it

#

Perhaps because there are two categories of laziness.

Category 1: What you do...You feel lazy to do it, but you don't sit on your hands, you automate it

Category 2: The one people think of when you talk about laziness...You feel lazy to do it, so, you sit on your bum all day, and achieve nothing

#

We need new concepts honestly, if you got one, put it out there and slap a patent on it, you own it but you let those who can implement it do so

austere junco
tired bluff
#

🤣 really? had no clue it was a difficult thing to do honestly. That's insane though

#

the neurons have the ability to call one or more neurons

ouhhhhhh. Interesting

Idk why though but seizures sound cooler than hallucinations in the context of AI, I'd rather it seizes than it just boldly chats absolute waffle

austere junco
# tired bluff > don’t get stuck with old paradigms and patterns....think AI as a designer of f...

i agree, the AI at its present form is a scratch on a surface.

think about AI, robotics and our physical environment in the future. it can be built to support humanity. a human can be integrated into the system and vice versa.

i “love” the idea of playground economy; anyone can do anything. it does not remove the fact that a human needs challenges. and some burdens to overcome. but they can be part of society in a positive form.

“a playground economy” mark my words 😉

austere junco
tired bluff
# austere junco i agree, the AI at its present form is a scratch on a surface. think about AI,...

😂 Pilvijo repeating that phrase "playground economy", sounds like you're plotting something. I'll definitely mark those words in my diary, looking forward to hearing that out there someday.

A Playground economy...hmmm

Reminds me of this movie Ready Player one.

a human can be integrated into the system and vice versa.

Oh the possiblities, but for this to be possible, a lot will need to change on the infrastructure level, and I mean A LOT, I just hope the powers that be who hold monopoly over most of our tech infra today do not shut down every possible attempt to build this infrastructure apart from the one they control.

This brings me to one of the issues I have with putting software creation in the hands of everyone. When we finally reach the point where everyone is creating their own software, they'll still have to host them on already-existing infra and platforms because with vibe coding, we've only put Software in the hands of the masses, not infra, not platforms, so, I wonder how current infra and platfrom providers will scale to accommodate this flood of software and the costs both financial and performance-based

tired bluff
#

😂 Until someone injects a prompt to steal your lunch money, oh and grab your credit card on the way out.

Jokes aside, you're right though, if we do AI the right way, it'd be a splendid authority, I just fear that humans will still be the breaking point of that seemingly perfect utopia, because you know, they'll never let AI be autonomous, someone will always want to pull the strings, and because humans built it, they'll be able to find a way to exploit it, plug in their string and pull them as much as they want

#

So are we just doomed naturally? 😂

austere junco
# tired bluff Perhaps because there are two categories of laziness. Category 1: What you do.....

there are also noise and bias of the observer. you want someone to do something in certain timeframe. he/she doesn’t do it. you think she/he: what a lazy and slow guy, right? but what if the person have some others issues out of your knowledge and understanding or if they prioritize differently?

in my mind people are willing to do a lot and they struggle to achieve the goals. we just are build differently and we prioritize from our narrow perspective.

that’s why it is important to hire commitments. “i’m committed to prioritize your needs”.

quite concrete examples but easy for anyone to understand.

tired bluff
#

Yeah. Then we're naturally doomed as a species. The strong will always eat the weak, not because they're weak, but because they don't care to fight, they just wish to live and let live.

austere junco
# tired bluff So are we just doomed naturally? 😂

laws of nature are there, but there are people trying to redesign them. for example medical care is a form of this redesign, wven if we don’t really think about it. but how far will we enter with this redesign? interesting..

tired bluff
# austere junco there are also noise and bias of the observer. you want someone to do something...

Ouuh, didn't think that...you're right, laziness is actually relative..depends on the scenario and the person.

that’s why it is important to hire commitments. “i’m committed to prioritize your needs”

This is actually smart you know. Because if you hired someone just because they have a history of doing things fast and always complete, your case might be different for them, because your needs might just be the one they're not really motivated to prioritise, so, the same active very keen worker from another company comes to your and suddenly becomes a liability

#

Then the wolves will gobble them up and snatch the AI from their hands. There's a reason why the villain is always stronger than the protagonist and in most stories, the protagonist wins by chance, they get punched up first then win by a slim lean shrivieled up chance. It's because the villain just doesn't care to play by the book like you said. MC is trying not to kill him, but he doesn't care if he kills the MC, so, he punches with his full strength while MC subconsciously holds back.

Perhaps, we just build the perfect AI with no biases, and we chuck the keys to it, no one has it, it doesn't exist, we just destroy it and embrace the possiblity that some day it might decide that the only logical way forward is to wipe us all out, and we won't have the keys to stop it 😹

austere junco
austere junco
tired bluff
# austere junco you are quite right and wrong at the same time. it’s not Always like that. and...

but if I do not support it, and you do not support it, what are we going to do? call the police on the bad guys? If the bad guys are "wolves", they're probably already holding positions of power in society, the police can't stop them, the only other option is for us to fight them, and when it comes down to a fight, they'll win because like I said, we'll hold back, because we can't bear the pain of taking them out completely, we'll want to play it by the book, take them to court, but they won't hold back, they'll crush the sheep

austere junco
tired bluff
#

🤣 had no clue it did that. Sadly, it also let them use it for the nefarious activites of that which I dare not speak of (child and human abuse) and it just let them do so, so, idk

tired bluff
austere junco
tired bluff
#

then all the ice in the world melts from all the compute. The seas boil and levels rise 🤣

But you know, they're considering moving semiconductor manufacturing to outer space, perhaps by the time we get to that level of processing, we can also move most of the infrastructure out there where it's colder, so we don't have to worry about overheating the planet

tired bluff
austere junco
# tired bluff That's a nice idea, but again, human flaws...who will define what "better behavi...

great statements you have here.

we all have some moral gudelines throughout the planet and nations. we have variety of religions around the world assessing these frames.

and finally: everywhere we are the same. we practice same procedures day in day out, we need love care and shelter.

at the times we are reaching out to space it is even more important to have joint guidelines for us “eartherains”.

do we need a joint enemy to be able to unite.

tired bluff
#

Gordon Bennett!!!!!!!!!!!!!! Why have I never heard of this? I never get this in my feeds, and I get a lot of recommendations on this topic of AI infrastructure but never heard of these people.

Thanks a lot for putting me unto em, I'll be reading up on them a lot more

tired bluff
austere junco
#

thanks for the great company today. enjoyed 💯

tired bluff
#

Was a lovely chat with you two great people

austere junco
#

see, we are all the same, around the world, good morning and regards from a mid-aged, a mid-smart and a mid-sized lady.

spark fern
#

dont know why i always feel sleepy when i try studying , but i feel energized when i start working

fallen estuary
#

hi all

#

why is this stuck here?

image.png
thin meadow
#

Can I share the link of my yt vid here? Will it get deleted? Is it allowed?

karmic sphinx
karmic sphinx
thin meadow
thin meadow
karmic sphinx
#

I would think that would be okay to share!

thin meadow
# karmic sphinx I would think that would be okay to share!

Okay 👍🏻
https://youtube.com/shorts/tPCcFsSj0cY?si=4XZyzVeHan58EPmc
it's not self promotion. If you think it is, I'll delete it right away

YouTube
ShifoLabs
MADE IN HEAVEN animation (Gemini 3) #animation #jjba #gemini #jojos...

PROMPT:
Create a self-contained, single-file HTML/CSS/JS animation on a 1:1 square canvas that simulates the "Made in Heaven" time acceleration effect. Render a static stick-figure or silhouette character standing on a green ground strip at the bottom. Implement a rapid, looping day-night cycle where the background transitions quickly between da...

▶ Play video
rich raven
#

Hey guys,
I have a question here. In simple terms, what's the difference between Google AI Studio and Jules?

Also, what about Firebase Studio? All do coding tasks and they are all agent-based, so in what scenarios can these tools be used?

thin meadow
rich raven
# thin meadow

Thanks. I was expecting an experience-based answer. If you have used all three tools, how do you differentiate them? Sorry if my question was vague.

thin meadow
# rich raven Thanks. I was expecting an experience-based answer. If you have used all three t...

Google AI studio is used for testing apps majorly based around on the LLM models... If you want to make a chat bot using gemini model, feel free to go to AI studio. I found it better than Jules for making those chatting apps and various AI applications like ideas generator, or any app based around the AI models of Google (though AI studio can be used for making other things too, it specialises in AI application)

Jules is an all-rounder, it can code anything which comes in its capabilities. I have made various things using Jules. Discord bots (including chat bots, Moderation bots, fun etc.), websites, automation tools, programmatically made animations, vibe coding app, etc.. So basically, an all rounder

I haven't used Firebase studio so I can't help you with that

rich raven
thin meadow
fallen estuary
thin meadow
#

Made a new animation about a cosmic slideshow...
3.0 flash and pro had more potential if I had written the prompt a little bit better.... The "Thinking" model is also good but it didn't perform like flash and pro for this specific prompt... Tho it isn't like this always... Thinking can rival pro when we compare simple prompts
https://youtube.com/shorts/-h7Qq8pZ4dU?si=Zgg6AeJH1w7UFrMd

YouTube
ShifoLabs
COSMIC SLIDESHOW animation (Gemini 3) #gemini #animation #solarsystem

PROMPT:
Create a single-file HTML/CSS/JS animation on a 1:1 square canvas. Sequentially display the Solar System bodies (Sun through Neptune) one by one. Each body must slide in smoothly from the right, pause in the center for 2 seconds, and then slide out to the left. Give the Sun a glowing, pulsating radial gradient and render distinct rings f...

▶ Play video
quartz bobcat
#

Hey there just came here cause i'm one of those guys who usually join a Discord server when they need help:

Been trying to use Antigravity to code, specifically the Gemini 3.0 Flash model, but at some point of it's task it just gets an error but i can't stop generation so i have to restart Antigravity, is there a sollution? Am i doing something wrong? Claude Sonnet and other models don't seem to be doing that

quartz bobcat
#

server is very dead for a 84.999 members one

potent jasper
ripe hill
upper tulip
#

Is there any better limits for Ultra plan against ONE AI PRO plan when using Antigravity? I cant find any info

austere junco
haughty cairn
#

-.-

proven depot
#

Yo, what's going on?

Hey, my name is Avoshu, so I'm pretty new to this community. I recently became a vibe coder, and I'm in love with it.

Now, what I will say is I've gone through Lovable, Base44, Cursor, and Honestly. I love how Firebase has a bunch of models, and they're literally permanently free. Shout out to Google, man. I was able to build a SaaS app and make some profit off of it.

agile anchor
#

Is it possible having antigravity access files even though they are in gitignore? The goal here is to have files not commited and then antigravity still reading these files

mighty basin
#

im unable to connect to this mcp even after setting "timeout": 9000

image.png
bitter fractal
#

Who wants to see what I am making

#

If anyone plays Roblox I am making a program for it

alpine solar
#

Anyone having trouble loading Google AI Studio?

modern charm
#

White screen on the google ai studio > fixed this prompt: ROLE: Senior Frontend Engineer
TASK: FIX GOOGLE AI STUDIO PREVIEW CRASH (REMOVE IMPORTMAP CONFLICT)

CONTEXT:
I have analyzed the Google AI Developers Forum (specifically post #52).
The "White Screen" and "Error #31" are caused by a version conflict in index.html.
The file likely contains a <script type="importmap"> or CDN links that force React 19/Beta, clashing with the local React 18 in package.json.

ACTION PLAN:

  1. CLEAN index.html (CRITICAL):

    • REMOVE the entire <script type="importmap">...</script> block.
    • REMOVE any <script src="https://esm.sh/react..."> tags.
    • ENSURE the only script tag is the module entry point: <script type="module" src="/src/main.tsx"></script>.
    • Ensure the root div is exactly: <div id="root"></div>.

  2. VERIFY package.json:

    • React and ReactDOM must be pinned to standard versions (e.g., "^18.3.1").
    • Remove any "overrides" or "resolutions" section forcing other versions.

  3. VERIFY vite.config.ts:

    • Ensure it is a standard configuration without custom "alias" mappings that might be redirecting React to a CDN.

OUTPUT:
Provide the FULL, CLEANED code for:

  1. index.html (This is the most important file to fix)
  2. package.json
  3. vite.config.ts
rich raven
dapper wyvern
#

Nice! Gemini CLI has image pasting and agent skills support in the preview version. Hopefully Antigravity will get them soon too.

tired bluff
lament harness
#

Yo guys I made an Oscar with only Claude code and it will release at SUNDAY THIS WEEK

agile anchor
rich raven
agile anchor
rich raven
agile anchor
rich raven
agile anchor
rich raven
cobalt cave
#

Wassup Guys

gilded zephyr
#

helloo everyone

#

anyone using gemini cli here?

nimble moth
worn chasm
#

Anyone know if the AI Ultra plan offers more Antigravity Claude Opus over the pro plan?

formal epoch
#

For antigravity agent when they want comments

ripe hill
formal epoch
#

Try restart computer

cosmic ferry
#

guys is antigravity down?

hollow canyon
#

hello

hollow canyon
cosmic ferry
hollow canyon
cosmic ferry
#

and restarting doesn't help

hollow canyon
#

it looks fine tghough

#

check your internet

hollow canyon
cosmic ferry
#

it's just crashing

#

every time

#

it starts a task but crashes after a few seconds

bitter fractal
#

hey vibe coders

dapper wyvern
#

Antigravity needs a "Do it" or similar button, so that after reviewing a Plan or Task list, I can just click that rather than type something witty like "Go for it" or "Make it so"

slow hare
#

yoo guys when you make an app with google ai studio how do u >>??

#

make it apk ??

narrow meadow
#

Bug report - on the CLI, Gemini 3 flash and pro (and I believe other models too), will very often write abbreviations that replace code. This has to be improved, have lost huge snippets like this. Example below:

image.png
umbral inlet
misty wasp
#

@narrow meadow I must make you aware that official support from Google is not provided on this server
If you wish to report bugs it must be done through other official support systems

#

You can find the support links pinned in #general

narrow meadow
#

Gotcha thanks.
If i may suggest, please create a dedicated subchannel in here for support too, would greatly help communicating feedback 🙏

misty wasp
#

I don't think Google has plans to do that as of now since the other ways are better suited for it but for any suggestions related to the server you may head to #1334928890120114328

nimble moth
#

@inner jungleHello, please keep in mind that self promotion or advertising is not allowed as per #rules.

inner jungle
#

ohhh woops sorry

#

my mistake

twin ether
#

What happened guys why Antigravity is not working for hour ?

agile anchor
#

Imagine Antigravity adds Codex into its usage limits 😂
That would be craaazy

broken burrow
#

I use Antigravity under a Google Pro account, when Claude reached the limit it used to be blocked for a few hours before the limits reset, usually 4-5h, but now it is blocked for days.

is it a bug? or do they changed something recently?

image.png
ripe hill
compact basin
#

Hello everyone!
Few weeks ago, my company bought the 2TB google ai pro. The problem that we have been constantly facing is that when we try to login into the Google Antigravity application, after the successful authentication, the error message is shown that our account is ineligible.
The regions we have used are the supported ones. Our account has @gmail at the end. The account is age verified. We have tried 3 different browsers: Google Chrome, Brave, and Firefox. Despite trying all that, the error is still being shown. We tried to contact Google antigravity support team by emailing them but no reply from them even after many days.
So does anyone know how to fix this issue?

jolly rover
#

Guys does anyone know how to deploy the app that is built using Google ai studio and firebase as backend.
And how to upload that app in the Google play store. Is that even possible?

tired bluff
tired bluff
silver ember
#

ooof

#

3 days of agony

silver ember
compact basin
round solstice
nimble moth
#

@jolly roverHey, sorry, but we don't allow self promotion or ads as per #rules so I had to delete your link!

jolly rover
tired bluff
ripe hill
compact basin
mighty basin
#

can we use antigravity in windows but tell ai to run commands in linux/wsl terminal?

boreal trench
#

Is there any specific App (vibe coding) through which I can successfully deploy web3 projects with Transaction contracts.

tired bluff
# compact basin Thanks but we purchased the subscription using that account so even if it works ...

I get that. This has been a known issue, I think I saw somewhere that antigravity was made "primarily" for personal accounts, so...

You could check out this thread on github

https://github.com/google-gemini/gemini-cli/issues/14858

GitHub
Account not eligible for Antigravity · Issue #14858 · google-gemi...

What happened? Note: I'm creating this issue since I've already tried to reach support through antigravity support email but got no response after 3 business days. I was using Antigravity w...

#

There's activity on there

safe yoke
#

Hello vibe coders - you may be interested in attending an upcoming workshop we are running on AntiGravity - should be fun: https://gdg.community.dev/events/details/google-gdg-ai-for-science-india-presents-googles-antigravity-accelerating-research-and-scientific-pipelines/

Google Developer Groups
Google’s Antigravity: Accelerating Research and Scientific Pipelines

Virtual Event - Google’s new code editor, Antigravity, is an AI-assisted IDE (Integrated Development Environment) with native agentic an...

ripe hill
#

u only need connect your wsl in your IDE

image.png
ripe hill
mighty basin
ripe hill
#

yea

silver ember
#

i just set up a new repo and it remembers history from another project? wtf?
why does antigravity have access to history of other projects/repos? this is context pollution?

image.png
compact basin
bitter fractal
#

Made a Roblox game builder

#

With 3d models

tired bluff
cunning crane
#

maybe server is down

#

wait for some time

#

Did u even google? reddit has answers

mighty basin
cobalt cave
#

vibe coders lmao

lucid idol
tired bluff
south turret
#

This is why I almost never use Gemini

image.png
#

............................

image.png
#

Gemini is like that kid at school that no one talks to
And when all other teammates are taken and you get to discover why

mighty basin
south turret
#

well sure when I try to color in the lines Gemini is very useful

mighty basin
#

rn kiro is fire

image.png
proper mauve
tired bluff
tired bluff
tired bluff
tired bluff
tired bluff
mighty basin
upper tulip
#

Good morning ladies and gentleman.

#

So Antigravity for example creates artifact "bla bla report", how to pass it to new chat aka session?

#

Or just - copy -> paste, would be nice to have some button -> pass to new chat

upper tulip
#

Anyone haves problems with browser agent?

image.png
#

[Error - 10:28:31 AM] Stopping server failed
Message: Cannot call write after a stream was destroyed
Code: -32099
[Error - 10:28:31 AM] Stopping server failed
Message: Cannot call write after a stream was destroyed
Code: -32099

lusty talon
#

It just redirects so I’m not entirely sure, I’m assuming that’s a reason why

light quiver
#

Hi

tired bluff
umbral obsidian
#

Does anyone got any idea how to import figma designs into antigravity to convert it to html css javascript. So that editing would be made easier?

tired bluff
# umbral obsidian Does anyone got any idea how to import figma designs into antigravity to convert...

I don't use antigravity, but I use figma, and from that, I know you'd probably have to export the design as html from figma itself, which'll contain all the css either inline or internal, you could use a figma plugin for this, either that or use figma make (never used this myself, I've heard so much about it).

After you get the html out of figma, pass it on to your AI and ask it to use that as a base to build out the features, it should handle moving to a css framework/library from the internal/inline one figma exports, and it should also be able to add js functionality to them. I mean it should even be able to build react or nextjs apps from just the exported figma code.

#

Also, Welcome to the Server

tall oak
#

https://youtu.be/hJ28djyR5NA?si=OzzqG-hVRGIbhvkI

YouTube
Programming teacher
I Built my own AI in 10 Minutes! 🤖💻 (No Coding Required)

Stop spending months learning to code just to build one app! 🚀

In this video, I’m going to show you how the world of programming has changed. I built a fully functional AI Chatbot (like ChatGPT) in exactly 10 minutes without writing a single line of code manually. We are using the power of AI Agents and "Vibe Coding" to bring ideas to life...

▶ Play video
south turret
tired bluff
#

When I first read that, I thought you said that you were the kid who was left with the kid no one talked to (my brain read that meaning into it), but at a second glance, you were the menace 🤣

south turret
#

Oh my god google actually implemented a feature request I did

#

I feel like one in a million right now

hollow canyon
south turret
#

I kept using chatGPT because of that "feature"

#

with Google to quickly make a new chat I had to select the url, drag it in between 2 tabs to make a new tab, click on new chat, yuck

#

now its gone, one tippy tap and im in biness

hollow canyon
#

wont work for me

#

oh this thing?

Screenshot_2026-01-14_at_10.19.36_AM.png
south turret
# hollow canyon oh this thing?

Yes just mouse middle click (wheel click) on browsers it opens a new tab with the link you clicked, but if a web dev doesn't uses link but rather javascript then the middle click doesn't work

#

So they fixed it :}

hollow canyon
mighty basin
dapper wyvern
misty wasp
#

Hi @faint aspen
Please avoid sharing your github as it violates the self promotion rule

unkempt folio
#

@misty wasphello

misty wasp
unkempt folio
misty wasp
# unkempt folio I need a help.

Please elaborate
Do keep in mind that I cannot provide technical support here nor can you receive any official support from Google over here
If it is anything related to the moderation of this server I will be happy to help

unkempt folio
flint furnace
mighty basin
#

Idk it felt way better this time

misty wasp
unkempt folio
vague warren
novel halo
#

how the hell do you make Gemini be more communicative in Antigravity, like Opus?
it keeps making mistakes but is crippled at communication

mighty basin
south turret
south turret
#

if you use buttons without link tag and use javascript on click, then all browser features that come along the link tag stops working

jolly rover
#

Hi guys I'm looking for someone who knows how to convert the app code that we build using ai tools like google ai studio, lovable, rocket and convert them to apk, abb files for publication of that app
And also how to integrate an AI agent that is in N8N to that app.

I hope I'll find someone and work together.

cunning jetty
#

What is the reason for this although I meet all criteria in FAQ and I have Google AI Pro from Google One?

Screenshot_2026-01-05_211218.png
desert island
#

How to install skills in antigravity? The docs even don't mention this?

silver ember
#

gemini3pro high be like

image.png
#

i also feel like gemini is high on smth

stone mica
#

Look at this good old internet vibe

misty wasp
#

Sorry @stone mica your link is being removed as it counts as self promotion
Please avoid posting it in the future

stone mica
#

sorry but i am not sellin browser yet

stone mica
misty wasp
stone mica
#

small gif

#

OK

#

?

#

there is no political ads]

misty wasp
#

And please avoid any political talk here
This is a place to talk only about development related to google products

stone mica
#

OK

#

so can i post screen

#

ok delete if you dont like it

#

ok i found different image

umbral inlet
#

Do you guys find Gemini 3 flash better than pro in cli?

#

Or is it just me

cunning jetty
#

It is far better and consumes less credits

south turret
#

Deleting all Gemini chats and Gemini still talks about my previous chats.......... I love privacy

#

Jesus will I need to change account to clear the context

mighty basin
south turret
#

lmao

#

Jesus christ, asked it what do you know about me and it gave me a summary of our previous 60 days of conversation I deleted

#

God dammit Google fkin hell

#

What a way to ruin the quality of answers

mighty basin
#

may jesus christ bless u with ultimate privacy

tired bluff
tired bluff
#

😬

south turret
# tired bluff wait, really????

Yes, I mean I knew that they kept deleted conversations because they all do this, to get them to potentially "thruly" delete them you have to send an official request for a GDPR data deletion

#

But the fact they use it for context even post-deletion pisses me off

tired bluff
lucid idol
dapper wyvern
#

Having AG write skills for you, that run Python scripts, is such a great feature. Only issue is I have require accept for all turned on, so that takes away a bit of the magic. 🤔

static sphinx
#

Has anyone run into a problem that the agent keeps crashing in AG even running a fresh install and an empty project folder?

hollow canyon
#

I think you should wait it out?

ashen crystal
#

hi all, how we can check the code for injection, logic or others problems to be production ready?

south turret
keen forge
#

Hello, could anyone offer advice on tips and tricks on how validate my project is working

south turret
#

How does one not know if a project is working

#

I'm not sure how to answer

keen forge
#

Well, I'm building a new infrastructure thats completely different ftm tcpip protocol. Working on signal based daya transfer.

#

Im not a coder in a sense of normal understanding i think. I use math to talk to AI

static sphinx
hollow canyon
#

Oh and congrats on Bronze

static sphinx
#

I dont think so, account works with everything else except AG. LOL thanks

hollow canyon
#

🤣

#

Now I need to get to silver

tired bluff
# ashen crystal hi all, how we can check the code for injection, logic or others problems to be ...

hire a dev, not just any dev, one who is security inclined. You could do it yourself though, there are services like Snyk Code, GitHub CodeQL, Checkmarx, and SonarQube. OWASP ZAP, Burp Suite...etc, you could use those for security testing and composition analysis.

If you don't fancy either option, you could pay for a service like aikido security or Qodo, or CodeRabbit, but I wouldn't trust those because they all use AI for the review, which leaves a lot of room for hallucinated quality

frail pagoda
#

what happened to Gemini. She used to produce functioning applications now she spits out mock preview junk

mighty basin
gleaming cave
#

anyone else having Antigravity issue now? it just keeps saying, one moment,, the agent is currently loading

2026-01-16_11-05.jpg
#

restart helped

static sphinx
tired bluff
mighty basin
novel halo
#

maybe I'm doing somethign wrong, but I'm not having any of these issues with Opus

#

it is truly special

south turret
#

But without knowing the logic youself you can never know for sure

keen forge
south turret
keen forge
#

Well its actually known as quantum repeaters

south turret
#

Are you by any chance experiencing AI induced psychosis

#

Also known as AI sycophancy

#

Or AI induced manic state

keen forge
#

Oh, well my job in the military was a chemical specialist, ao IAM a biomolecular specialist also. IAM new to Ai but my math is sound but my Ai skills are still being developed

#

So if working on my project is vased on my real experience in my work. Then i think IAM in a manic state. With building what is normally aeen as IIT

keen forge
#

What is this new Ai sycophancy? Is it a new disorder? That we developed as a societal consensus?

south turret
#

disorder

keen forge
#

Interesting. , when faced with a Ai driven race. Many who couldn't contribute to the world is normally displaced. And i think that Ai is helping them break free from stagnation

south turret
#

Everyone contribute to the world, it doesn't have to be digital for it to happen

#

Before AIs there were books, though not as interactive it still contains knowledge, hopefully curated by their author

keen forge
keen forge
south turret
keen forge
#

😇

mighty basin
#

meh

south turret
#

Between today and 10 years ago there's nothing much happening, with or without AI.
The real science is behind closed doors and cannot be AI generated

keen forge
#

Ai is an extension of your reasoning

south turret
#

For example the recently claimed Donut Batteries, solid state batteries with TiO2, that's a massive step, and that's not AI taking it

keen forge
#

Yes its imagination

south turret
#

AI is fun but in the bigger picture, there's barely any contribution for now

#

In the future I can see medical screening being fully automated which will be productive yes, but in the field of discovery so far, AI probably barely has an impact

#

It is useful to know the common names for theories/concepts though

#

But that's about it

keen forge
#

Is that what u think of my project

south turret
#

This conversation was not related to your project at all no

mighty basin
keen forge
#

Would you like to see through math instead

south turret
#

I'm stick with my Dr Pepper thanks though

keen forge
#

Code is math

south turret
#

Another 20 years and we'll have them in our hospitals lol

south turret
keen forge
#

🤔 they are now

mighty basin
south turret
#

Oh, its a thing by google, it lead me to a specific AI model with the same name for medical imaging

#

... it is

keen forge
#

If you want something simpler. Calculate 6π

#

Tell me what you get 😇

south turret
#

i got 6π

#

are you talking about the "MedGemma Impact Challenge"

keen forge
#

😁i want 1

south turret
#

seems like a lot of work for a potentially minor reward

#

wasting a month to hopefully get 5 grand, might aswell get a normal 9 to 5

#

without disrespect to those who actually try for fun

keen forge
#

Oh, that's a little anticlimactic, you are in a tech community. Its a lot of work to sit here and put down dreams

#

I would like to join that hackathon

mighty basin
#

what is this error now 😭

image.png
tired bluff
tired bluff
# keen forge Hello, could anyone offer advice on tips and tricks on how validate my project i...

Given everything you've said about your project so far, there's honestly no real way to completely outsource the validation to AI given that this is something "novel". AI thrives in what's known. Even a framework that's existed for yonks but with little to no public project examples and usecases will leave even super advanced models fumbling.

You should either actually learn to code, I mean innovation is the product of two unrelated seeds of knowledge colliding in the inifinite chaotic void that's the human mind. So, if you have the two pieces colliding, in order to execute, you'd obviously need to know enough about each piece to put them together...either that, or you invest into hiring people who are very good at each of those pieces to build those pieces collaboratively. This is what elon et al do.

If you really need to automate it, then your only option is to do as Code already said...you'd have to use different conversations or diferent AI models to verify, but in order to do that, the AIs need to have a solid idea of what they're verifying against (a benchmark) so, you need to establish a very clear baseline...a very clear definition of what the goal is, after which, you define what the requirements of the goal are (both functional and non-functional requirements of said goal)...this becomes your Product Requirement Document (PRD), then, you could give it to AI to break it down into distinct phases, and each phase has a defined goal, and defined "requirements" again. Verify each phase's stated goals and requirements, if they are okay, then for EACH (do not make it do them all at once...one after another to reduce the chances of hallucination), ask AI to generate a specification document for that phase. At this point, you're hoping that each sub-phase is granular enough to at least be a pattern the model has seen before in its training data or in the world around us.

#

The aim is to break through the problem of the "unknown" as a massive system, by breaking up the system into sub-modules for the AI's patttern matching to work better

#

So, finally, for each established phase, get an AI Agent to build it, and when it's done, spin up a new instance to verify the build system against the phase's SPECIFICATION document, after it's done, spin up yet another agent, ask it to verify again against the SPEC doc, keep doing this till at least two consecutive agents give you the green light...at least 90%

keen forge
# tired bluff Given everything you've said about your project so far, there's honestly no real...

Thank you. Thats why IAM here to learn. Yet when i when i fo ask questions many think that i depend on Ai a lot. Whats truly weird is if i did. I wouldn't be here. Id be asking Ai for advice. Thank you for your help and advice. The only difference is IAM not a developer, i do quantum algorithms. And by doing algorithms it's not based on code. Its mathematics. The very architectural design you code in. I built that very foundation. 😁 When i learn to code. Ill design an algorithm that will thank you for your very helpful advice

mossy bloom
#

You can generate tests, create snapshots you want it to correspond to, and then define what conformance is and try measure conformance tests against that. Ultimately a huge inhibition to validating generative code is you allow it to exist in a vacuum without aggressively subjecting it to tight constraints within your working repo

#

And yeah what static said, product, specification, api, whatever you want to call it driven development where you define outcomes and what you desire upfront while dividing a problem into atomic components allows you to actually hopefully make usable tests

keen forge
#

Could you offer tips for prompting. 🙂 I know you have to be specific. But how would you word it so its not too overwhelming

tired bluff
# keen forge Thank you. Thats why IAM here to learn. Yet when i when i fo ask questions many ...

When i learn to code. Ill design an algorithm that will thank you for your very helpful advice

😹 Can't wait to see that.

i do quantum algorithms. And by doing algorithms it's not based on code. Its mathematics

I mean code is like mathematics, you should fit right in. Same flow...get a problem, figure out what the shape of your inputs are and what the shape of the output is, break it down into smaller parts, and for each part, figure out what the possible formulae are...pick what's best or test a number of them to pick a final one, and build your way back towards an assembled output, basically a diamond shape. Only difference is the syntax, maths uses a different syntax than what we use

keen forge
tired bluff
mossy bloom
#

all the prompts I end up using are like hyper specialized and I GENUINELY don't know if they're bad for what they are, and I end up having to alter the outputs myself afterwards for both conformance and to make my tooling not scream at it like it's an invasive infection

keen forge
#

Ai tend to think quantum is pseudo calculation and i get a mix bag of answers. Using merkba spin on a geometric shape to anchor point of reference, if i wanted to calculate the quantum gravity of such . Do i need to always write out the formula?

mossy bloom
#

it's like code generation where I'm too lazy to summon an AST parser so I guess I use the AI like a dumb compiler

keen forge
#

See parsers is what? These terminology is what is confusing because theres many languages and what is a good language to start? I know Python

#

But julia i hear is better for math

#

How would prompt for specific flux points that needs accuracy? Sometimes it won't get the objective. Do i need to calculate the fidelity and the invariants? To thats specific focal point?

#

Is there a method? Or do i need to always override the Ai?

tired bluff
# mossy bloom it's like code generation where I'm too lazy to summon an AST parser so I guess ...

I personally always prefer having an AST parser sat in that layer when when I want to automate (code gen), I prefer them over LLMs, I just don't like dealing with the back and forth with LLMs, I'd rather build a source to source gen which lives as an artefact in my actual source code, then for usage, I'd just have to run a single command, get a deterministic output and never have to worry about verification or burning tokens on teaching it why its wrong here and there.

keen forge
#

Is it ok if i paste the code and see if anyone can verify the code is sound?

mossy bloom
#

It's deterministic, faithful, it doesn't betray you (unless you make it wrong, in which case /shrug)

tired bluff
tired bluff
tired bluff
mossy bloom
#

Making a typed tree structure and running transformation passes on it until you spit out beautiful code like an abstract dragon

keen forge
#

And its hard to do.

#

This code will help me build a dual circuit generator that will resemble the heart. The math is sound

#

But stressing a hear on a machine i need to learn how to stabilize the nodes to harmonize

#

But pleae double check my math. See if you can calculate the same outcome as i do

tired bluff
# keen forge How would prompt for specific flux points that needs accuracy? Sometimes it won'...

Personally, I'm not a very good vibe coder, I only vibe code once every quarter or so to see how far we've gotten, so, I honestly can't help you out here, but from my experience building agents, when the LLM doesn't get the objective, you usually go back to tweak the system prompt and/or the temperature, but those aren't exposed for consumer models...but if you're going to tweak system prompts, you would have enough domain knowledge to know how to guide it, so, I reckon you'd have to know enough about what it's getting wrong about the objective to steer it back on course

keen forge
#

Thats what i have to normally do

tired bluff
#

a moment, Let me check ths out

keen forge
#

I have to keep jailbreaking the design

#

Its very hard when you need to adjust parameters and i get this type of coding. 😮‍💨 I want a simple standalone calculator that will stress test in a window gui

#

This looks more like just visuals rather than calculations

#

So i need to verify if its actually running the code and calculations or just pretending to simulate the work.

#

And if its actually really deploying nodes. Or are they mock data? Running bio molecular structures is bad if Ai only trained on mock data and not real ones. 😢

tired bluff
# keen forge And if its actually really deploying nodes. Or are they mock data? Running bio m...

Alright, so I hate to be the bearer of bad news 😅 but it is not really doing any "real work" in the sense of stress testing.

When I first opened the code, it looked fine to me because it is actually running js logic every frame...thhere is a real loop and values are being updated continuously, so it is not a fake animation or anything like that. But after reading what you actually want to use it for, yeah... it is not doing that.

The "nodes" are basically synthetic data. It is just creating an array of hundreds of objects inside a useEffect, with random positions, phases, and velocities. So yes, the nodes exist and are updated every frame, but they are not coming from any real dataset or model, they are essentially randomly initialized oscilators.

Given how many objects it is pushing through the loop, it is mostly stressing canvas drawing and maybe React a bit (coz it's doing state updates every frame), but computationally it is very light. There is no heavy arithmetic, no real workload, no proper stress test happening.

I am no expert in biomolecular structures either, but just to be clear, this is not doing any bio modelling at all. If you were thinking of using this as a base for anything related to that, it would be a bad fit. Like I said already, it's actually just a visual osciallator demo, I haven't run it yet to see what the UI looks like though, but it probably looks nice

keen forge
#

I should be resonating with a coherence of 95% stability. Yet its from both real qpu vs classical. The convergence is around 87%

#

Something wasn't adding up. Can you validate this one for me too

#

I sent you a code to see these python codes are running through real datasets

tired bluff
keen forge
#

Thank you so much. Your words truly helped me 🙏🙏🙏🙏

tired bluff
sinful vine
#

does anyone know the exact usage limits for anti gravity free plan?

#

I have been using Gemini 3 Pro and Gemini 3 Flash and OPUS 4.5 all fast mode for like a week

#

I have not hit any usage limits yet; however i would like to know a rough estimate so I can conserve my model calls

heady valley
#

I cant figure out dev containers in antigravity..

#

No support yet?

misty wasp
#

Hi @jolly rover
Please avoid any promotional posts on the server as advertising is prohibited here
You may review the #rules

jolly rover
bleak citrus
#

so i got around 1k$ as the free trial for vertex ai, but i cant use any of anthropic models even when i enable them, why?

misty wasp
#

Hi @iron dew
Advertising is not permitted here as per #rules
Please avoid it

latent vector
#

also, worth noting that using some third parties models such as anthropic requires a paid billing account with an actual payment method, those funded with credits cant enable it since it will ask you to add payment method first

magic harbor
#

Any one making mobile apps with antigravity? Interested to hear your workflow and learn more about this tool 🙂

mighty basin
# 
✕ [API Error: [{
    "error": {
      "code": 400,
      "message": "Please ensure that the number of function response parts is equal to the number of function call parts of the function call turn.",
      "errors": [
        {
          "message": "Please ensure that the number of function response parts is equal to the number of function call parts of the function call turn.",
          "domain": "global",
          "reason": "badRequest"
        }
      ],
      "status": "INVALID_ARGUMENT"
    }
  }
  ]]
#

anyone pls help 😭

#

gemini cli

tired bluff
# mighty basin anyone pls help 😭

don't think anyone can help you, I believe I said this a couple days ago when you posted the same error.

It's an API Error, and it says the LLM is passing incorrect arguments to the MCP tool it's trying to call..."the number of response parts for the function is not equal to the number of call parts"...and it's got status code 400 with status INVALID_ARGUMENT, so, it's either the LLM is hallucination the parameters or it's actually got the right parameters but the MCP has had breaking changes but your copy of the MCP is outdated.

You'd need to file an official bug report or try to remove and re-add your MCP servers

mighty basin
#

any vibe coder willing to burn tokens aka contribute ?

quick granite
#

damn i hate having a developer mindset.
I run out of credits on my free tire on Make.com and because my use-case is rather small I decided to start makeing my own make.com :/
And why would I use open source solutions which have 100x more features which I don't need anyway 😄

Made quite some progress for 1-2ish hours witn antigravity

itssomething.png
south turret
#

I don't know why I expected an #hardware-architecture on the Google server, made sense in my mind but yeah no when thinking about it Google doesn't really do architectural design at scale

#

probably some but not much compared to the other big bros

south turret
#

Bhahahahaha I wrote "[My idea] ... while I attempt to avoid delusions and/or maniac tendencies or even external sycophancy." and it said this on its own

image.png
#

At least we know that if you do mention sycophancy Gemini enters reality check

#

I love that it quoted something in parenthesis that I never even said once throughout the whole conversation where I stayed pro

mighty basin
cunning jetty
#

whats this new update

image.png
gilded zephyr
#

hey everyone, am using diff cli agents lately for development and wanted to confirm how well does gemini works with todo.md file? idea is to run multiple agents prallel

mighty basin
dapper wyvern
#

"Vibe discussion": I have setup repomix to package my project's source code, usually into front end and back end packages, along with the project's README.md and other files as needed and push them to my Google Drive account. Which would, in theory, allow me to talk to the latest version of my code while I'm out, away from my PC. Unfortunately, I can't get Gemini (with Google Drive turned on) to reliably get my project files.

torpid jasper
#

✅ fully vibe-coded

misty wasp
#

Hello @torpid jasper
Please refrain from sharing your own personal links as it counts as self promotion
The #rules here prohibit self promotion

torpid jasper
misty wasp
#

Hi@pliant viper
Please do not post your app link or ask for testers
It is not permitted here

mighty basin
#

Scam

shy inlet
dense grove
shy inlet
#

Do you like it?

dense grove
shy inlet
#

Created all with Gemini you can use any web browser except Opera because the tab on the bottom covers up the player ship have fun destroying ships

rare fossil
shy inlet
#

Okay

south turret
#

I hate gemini for that

image.png
#

The thing goes crazy so often on long prompts

south turret
#

I do blame Gemii, doesn't happen on Opuss

#

In fact right after that I rebooted an old conversation with Opus and he restarted from where Gemini left off

south turret
#

The best choice I ve ever taken on antigravity

#

Are these rules

Do not use any emojis.
Do not write comments.
#

Increased every model's IQ by at least two folds

#

Except Gemini, for some reason Gemini don't always respect the rules, smh

#

Especially in high stress, I find that when there's many issues in a row Gemini enters what I call panic mode, and starts babbling and mumbling and yappin all while breaking the rules

#

Can't help but imagine Gemini as a burned out employee

#

Similar to the special kid at school, it's good at what he does, but do not mention the color blue or something

latent vector
south turret
#

I'll def try this next time

gleaming cave
#

finished vibecoding kotlin irc desktop app 🙂 multi installers linux/windows/docker

winter viper
#

Sup folks, is it only by me, or did Antigravity agent started crashing a lot?

south turret
#

And on the bug report it asks me "how to reproduce" dude just read the logs idk I'm not the dev, you did a woopsie I didn't

grizzled moss
#

where's my antigravity homies?

trail dawn
#

Hey everyone, I need some help with something. I'm actively using Antigravity, but sometimes I'm not satisfied with it. That's why I'm currently happier using the ChatGPT 5.2 model with OpenCode. However, when I want to have a website built, I can't fully explain the design, and I want to elaborate with screenshots. Is there anyone who can tell me what I need to do when giving prompts or how I should use it? (All vibe coding agents)

south turret
#

Now I just sketch ideas and concepts and once I have the full picture and a deep understanding of the subject I run it through an AI to make what I designed

south turret
south turret
#

React ruining my life once again

pale pond
#

why the antigravity model keep getting error these days

jovial ridge
#

yeah it just terminates even for simple tasks

upper tulip
#

Experiancing same issues with termination

mighty basin
grizzled moss
signal raft
# south turret

same issue, but that is not the only one. ag is unusable now

proper mauve
#

Anyone can help me to connect mongo DB MCP to the antigravity

mighty basin
#

and with google ai pro the rate limits rare

grizzled moss
#

sending star trek memes to agents is the vibest of coding, imo.

605111373_1586989689158774_372172951941226482_n.png
tawdry heron
#

Vibe Coding hell yeahez

south turret
#

Just wanted to share that I've been programming for 2 days with my programming language now that's it is stable and Gemini is indeed able to code in a language it never learned, though it is mostly Rust based it's still interestingly impressive

#

Gemini has an issue of avoiding rules set though, so that's annoying, I often end up using Opus because of that

tawdry heron
#

Spending my 300$USD credits with Gemini 3 Pro preview Agent to develop my first fully vibe code app for the Gemini 3 hackathonez google_cloud

Captura_de_pantalla_2026-01-21_201808.png
south turret
#

What is that... smell

#

It's like... like spicy but like worst

#

Best way to describe it would be like, sycophancy

#

A sycophantic smell

tawdry heron
#

ez GCP ❤️

#

What a great time to be a developer!ez GoogleDevelopers

misty wasp
#

Hi @strong cloak
Please do not post your own links in chats as it counts as self promotion
The article's link can however be posted in #new-publications

mighty basin
molten talon
#

Anyone here having issues with antigravity model/login errors?

old pilot
#

Yes, cannot select a model

ashen sinew
#

Yup 🙁

old pilot
novel halo
#

Antigravity still issues with model selection?

#

It's been more than 2 hours now

ashen sinew
#

Alternatively, use Visual Code with Gemini Code Assist 😉

mighty basin
mighty basin
feral anchor
#

Hey. Have you experienced auth redirection issue with Antigravity?

ashen sinew
mighty basin
mighty basin
sonic sinew
#

hi guys, what is happneing with antigravity

mighty basin
feral anchor
sonic sinew
#

or maybe the eclipse coming in feb

#

lol

feral anchor
#

its not only me that is not working for?

#

i mean just wanted to try it today

#

i have not found any reliable soultion that works

ashen sinew
#

Now Antigravity works fine 🙂

old pilot
#

Yeah, works fine

#

happy vibe-conding 🙂

mighty basin
#

daymn thnx guys to confirm

#

its time

regal frigate
#

Hi guys

latent vector
#

hello

last pagoda
#

hey bros. hi

late flower
#

Hey

static sphinx
#

Good to know it wasnt only me having a bad time with AG

#

using it again now and sometimes the agent still crashes

feral anchor
#

which plan is enough for sonnet to be used all day? one developer

bright stream
#

yesterday, I upgraded to their top subscription (Google AI Ultra) because I kept needing to wait for more credits. Today the issue isn't credits, but repeating Agent terminations due to error.

#

Its been frustrating to use. I'm considering going back to Windsurf (who they licensed the tech from) until things become more stable

tawdry heron
#

no problems

bright stream
#

My primary reason for upgrading was better access to Claude's foundation models

#

But today I've really run into alot of issues with the IDE and I suspect google is dealing with some operational issues

tawdry heron
bright stream
#

Discussion form is full of people complaining. Hope they resolve it soon, its getting in the way of development.

tawdry heron
#

On developing, you need a backup for everythinggoogle_cloud

short lance
#

There's a weird corrupted .pb file / conversation file and it's not actually loading the conversation context into the chat. I require my context to be reloaded. The file size is 30mb suggesting the data is still there.

#

This is in Antigravity.

tawdry heron
#

scammer Androidblack <@&1009526435276394496>

swift carbon
short lance
#

Y'all if you're using antigravity please make sure you're making regular backups of /.gemini/antigravity/conversations

#

If I didn't have a backup from yesterday I'd have lost SO much work

#

Can I file a bug report here for Antigravity?

Sometimes Gemini gets caught in some weird loops and then it crashes and corrupts the entire .pb file.
I think there should be a way to recover chats from .pb files even after the .pb file gets corrupted, as it's a proprietary file.

I'm seeing on reddit that others have lost their conversations this way but still have the .pb file so being able to recover from a .pb file that got corrupted would be great.

lament harness
#

Also, Check out the Build History's!

misty wasp
#

Hi @lament harness
Please avoid sharing your own links as it counts as self promotion or advertisement
The #rules do not permit that

south turret
#

Gemini behaviors

image.png
south turret
#

You can resume using this model at 1/26/2026, 11:57:38 AM

#

What the hellllllllllllll, 3 days of using Gemini, I am NOT doing that, Gemini cannot code for shee

#

Guess I gotta get a Claude subscription back because hell no, Gemini is ok for small issues but I am NOT gonna headbash against it for 3 days straight

#

There is goes again, 30 seconds after starting to think, Gemini is not made for smart issues

#

It's so verbose dear god

#

Worst than me when I'm mad against an unwarranted 3 days penalty

#

There is goes again, god I hate Gemini for complex issues

#

Just realized I got a 3 day penalty right after setting a custom instruction to "Avoid sycophancy", do AI weight in on their token restrictions? Could it have read this, became anti-sycophantic (Hostile) and striked me? Because how is that possible

#

Either way canceled my plan

lament harness
dapper wyvern
tawdry heron
#

@south turret you need to read Prompt Engineering documentationGooglelogo

noble fossil
#

hi

sonic sinew
#

thx god antigravity is back, btw would be great to see the GLM model loaded

#

guys do you know if google AI pro has plan to be able on opencode?

regal frigate
#

Which prompt is the best for antigravity.json? Or yaml. Any type better than this

dense grove
regal frigate
#

(I like Claude for backend) but idk about frontend which one better

dense grove
regal frigate
dense grove
#

yeah

regal frigate
#

Alright thank

#

But I use gpt to prompt

#

Which ai should I use

dense grove
#

the best AI to use will depend on whether complex logic is required or if visual UI and large contexts are involved.

regal frigate
dense grove
regal frigate
#

How about ui

dense grove
#

it's up to u for using the ai

dense grove
#

claude

#

While Claude is smarter at logic, Gemini is generally more creative and visually aware.

#

Gemini is best for ui

regal frigate
#

Maybe I can use Claude to prompt backend and use Gemini to prompt frontend,sheers bro thank

dense grove
#

I'm sis

#

😅

regal frigate
dense grove
mighty basin
mighty basin
mighty basin
cosmic ferry
#

i cant make the computer use model to show thought signatures bruhh

mighty basin
#

btw do u use skills?

cosmic ferry
#

sorry

hollow scarab
#

Hi guys.. In Antigravity, when using workflows (with the / command).. Is there potential conflict later in the chat after using it?

#

i guess the instructions from the workflow .md file are staying in chat.. so if i use a command to /explain and then a command to /summarize how that will work? will the 'summarize' make the previous ignored?

broken prairie
#

Did Antigravity recently lower the amount of uses?

#

Like in last 24 Hours?

#

It says I am unable to use the Claude models for 5 days.

novel halo
#

There seems to have been an update to Antigravity, but their changelog isn't showing anything. Anyone got info on changes?

broken prairie
novel halo
broken prairie
#

Usually it was that it would reset with a limit every 5 hours. Not that it would suddenly cut me off for 4-5 days. Unless this was a feature that has been in place but I never noticed since I never reached that amount.

#

Is there a monthly maximum I was unaware of?

#

I just thought it was 5-hour resets 😅

novel halo
#

Free accounts weekly limit I think

broken prairie
novel halo
#

I recently noticd there was an update to AG, perhaps restart?

broken prairie
#

(Just so you can see that it acutually says it).

image.png
broken prairie
#

It's on all Claude + GPT models.

novel halo
#

oh, then you shouldn't have done that 😄 undo undo

#

jk, no idea really

broken prairie
#

Haha, tbf it was there before I updated it as well

novel halo
#

They are obviously experiencing issues

#

I would log out then log in again

broken prairie
#

I was hoping that my not updating was causing this issue

broken prairie
novel halo
broken prairie
#

The Google One Pro Plan

#

Yeah logging in/out doesn't seem to fix. I could try to re-install as well, but I will do that later. I will also see if the issue is across computers a bit later today

lusty lantern
#

is it just me or do you also have to close n open the app everytime the script is stuck

#

i cannot close it manually at all

#

like through keybinds

broken prairie
novel halo
lusty lantern
#

for example, right now i had a youtube text transcriptor script
it then sorta malfunctioned due to an url error and loaded pretty much nothing for

#

10 minutes

#

i cant close the agent nor can i stop the process until i just

#

close and open the app

broken prairie
broken prairie
lusty lantern
#

yeah, i can send a screenshot

broken prairie
#

Sure. I know that the Web Agent can be real buggy at times. Especially if it is trying to request access to the DOM

lusty lantern
#

and its like that for quite the time, and theres no way for me to close it

image.png
broken prairie
#

Are you able to see what the Terminal looks like?

lusty lantern
#

not at all

#

not even the thought process

broken prairie
#

It should be somewhere in the chat, sometimes it is hidden in drop-down menus.

broken prairie
static mantle
#

How the f do I check for my Antigravity usage? I think google's been hiding this thing. They just leave us with the the details: "generous rate limits", which we can't confirm or check the usage.

dapper wyvern
static mantle
dapper wyvern
viscid breach
#

who knows a way to install antigravity as a linux server software without a UI? purely for remote connection

static mantle
dapper wyvern
robust hinge
#

in this case it seems like the vibe coders are the google employes lol, for some reason reading a TXT file reverts the gemini web interface to essentially material design 1 sample elements in all caps with the default purple?

image.png 2026-01-24_02-22-36.png 2026-01-24_02-22-26.png
broken prairie
#

I definitely think there's some games being played with it so we have less info about it.

south turret
#

Why is Gemini so silent now, I complained yesterday that Gemini was overly Verbose and now it's dead silent.
I mean, I don't mind the silence, but in a room full of crazy people the room is very loud, even if the room was empty to begin with.

south turret
#

christ

image.png
icy thistle
#

hi

novel halo
#

maybe they're doing something temporarily until they deal with student plan abuses

novel halo
tulip stump
#

what in the heck? I didn't even use claude model and when the weekly reset happened, it just gave me another 7 days to wait?

image.png
tulip stump
daring adder
#

Any extensions?

robust hinge
ancient burrow
#

I'm having exactly the same issue and it's very frustrating.

Summary of events:

I could use Claude Sonnet 4.5 and Opus 4.5 without problems before the outage.
During the outage, Antigravity stopped functioning, so I decided to log out to try and resolve the issue.
After logging out, I was unable to log back in for several hours until this morning.
Once I finally managed to log in, I found that Sonnet 4.5, Opus 4.5, and other premium models are now locked with a "Quota limit exceeded" error.
Current status:

I am a PRO user (paid license).
My usage bar shows 0%, yet I am locked until Jan 26th (or 31st).
This is clearly a critical bug triggered by the outage and the subsequent re-login process.
Please investigate and reset the quotas for my account immediately. This is preventing me from doing my work.

2026-01-24_204353.png 2026-01-24_204319.png 2026-01-24_093705.png
daring adder
next glen
#

Antigravity Quota (AGQ) or Antigravity Cockpit for model usage quota?

cosmic ferry
#

what in the world. im on pro plan and my limits refresh in 3 days?

Screenshot_2026-01-24_212121.png
broken prairie
#

Thankfully I still have access to Claude Code, Cursor and Github Copilot if I really need to access more items.

misty wasp
#

Hi @lavish crescent
Please avoid posting your own links as it counts as self promotion which is against the #rules

south turret
lilac stump
#

all my pro accounts are limited

image.png
#

is there a way to make gemini 3 less lazy and spend more tokens , like its so lazy that i cant even come close to finishing my tokens .

tulip stump
cosmic ferry
tulip stump
#

pro account btw. I knew I should've just did the 3 month plan instead of annually.

cosmic ferry
#

im sure something is missing here

image.png
tulip stump
#

maybe.. I am probably just on the wrong plan and need to get on the new lux plan

lilac stump
tulip stump
#

so wait.. like spinning up a terminal in antigravity with CC?

lilac stump
tulip stump
lilac stump
south turret
#

I don't understand what is going on.
Gemini has been achieving everything I desired in the last 30 minutes with ease
And now it cannot manage to make a grid 5 units in height by 20 width, instead it keeps making it 5 units in width and 20 height, what

#

It's like a toddler able to design a space station but can't figure out how to draw stairs wtf is going on

#

Ok it started being verbose again, so when Gemini becomes verbose is when Gemini's IQ actually nukes and it becomes useless

lilac stump
south turret
#

Yeah Gemini is a special little buddy, it's on training wheels

lilac stump
south turret
#

nope

#

but considering how it doesn't follow rules half the time, I'm just guessing it's the same thing

lilac stump
south turret
#

what do you do with it

lilac stump
lilac stump
#

Wbu

south turret
#

Curiosity, me I'm making a general life manager, made a debt manager, made a medication manager/reminder, I intend to plug in bank functionality once it's legal in my country (expected this year)

#

It will be in read only mode so there will be no danger

lilac stump
#

nice

lilac stump
south turret
#

lmao

#

I'm not sure if that's meant to be a joke or if it's true but either way that's hilarious

lilac stump
lilac stump
misty wasp
#

Sorry @torn tulip advertising or self promotion is not permitted here
Please review the #rules

#

Please also be mindful of channel topics and do not spam

lilac stump
#

this is actually real

image.png
south turret
hollow anchor
south turret
#

I noticed a very odd pattern.
I made in canvas a "prompt improver" tool so I can quickly improve prompts

#

And every time I run the same prompt over and over the better it gets, despite the fact it only takes the input and not the output, as if Gemini internally had some kind of mechanism for iterative prompts to increase quality

It would make sense, the more the user regenerates a prompt's result the more likely it is that the answers requires a more thoughtful approach, therefore increasing quality, if that is what is happening

#

For example:
Prompt: The highlight is being cutoff from a likely overflow hidden...
Gemini revision 1: The highlight is truncated ....
Gemini revision 2: The highlight is clipped (truncated) ...
Gemini revision 3: The highlight is clipped ...

#

As if the refinement quality increase at each iteration, as "clipped" is much more accurate than "truncated" for a cut from an overflow

south turret
#

I recommend everyone to use this prompt Now be anti-sycophantic and grounded to reality.

#

I managed to make Gemini sycophantic with a real but "delusion" project and left Gemini to iterate over the idea until it went full on delusional mode, used the prompt, and it ripped its own band aid.

#

Only using "Grounded to reality" didn't properly work
And using words like "principled" or "critical" did not work, anti-sycophantic in particular does the trick

lilac stump
ripe nimbus
#

this sucks, i was usually able to vibe code for days without ever hitting the limit but i think they decreased the quota limit or something?

image.png
lilac stump
ripe nimbus
lilac stump
ripe nimbus
lilac stump
#

i am safe

ripe nimbus
lilac stump
#

gemini models are all fine its just the claude

lusty lantern
#

or something more idk

ripe nimbus
lusty lantern
lilac stump
# ripe nimbus wait what's this, how to check?

It's like a manager that lets you log in with multiple Google accounts at once, and it takes the quota from all the accounts without you having to manually log out and log in with separate accounts when you run out of quota on one of the accounts

ripe nimbus
lilac stump
lilac stump
lilac stump
#

by the way, does anyone know why the Gemini model is called Gemini 3 Preview? Like Gemini 3 Pro Preview in Gemini CLI, and in Anti-Gravity, it is called Gemini 3 Pro High or Low

latent vector
lilac stump
latent vector
lusty lantern
lilac stump
upper tulip
#

Very disappointing on PRO plan constantly - Error Agent execution terminated due to error.

novel halo
#

Antigravity: anyone know how long of a conversation needs to be for Opus quotas to deplete faster than usual, or is it always constant? Opus context is limited
We don't know how long the context is, but perhaps there's a way to tell?

upper tulip
#

it doesnt matter, its fresh seassion or not, error just comes up always non stop

#

Context is fine. Can say Hi and will get same error 🙂

#

Just spamming retry helps. So its not context problem for sure

sonic sinew
fallen estuary
#

Antigravity is dead

#

No funding maybe

latent vector
#

nah google doesnt need funding

#

they have unlimited money🤑

lusty lantern
broken prairie
#

I know I have, but specifically for Opus

#

My only assumption is that they are mimicking the system that Claude Code is using:

  1. 5-Hourly Resets
  2. Weekly Limits
noble orbit
#

One message removed from a suspended account.

broken prairie
#

Though the difference is, Claude is extremely open about the limits you have on their platform. Google is the opposite of this.

noble orbit
#

One message removed from a suspended account.

broken prairie
noble orbit
#

One message removed from a suspended account.

#

One message removed from a suspended account.

broken prairie
#

It's really dependent on your code-base and how large it is.

#

There are third party apps that show how much usage % you have, try taking a look at one of those and see ifi it works for you

noble orbit
#

One message removed from a suspended account.

broken prairie
#

I've personally found that Curosr has a lot less usage (at least for my cases)

south turret
#

To the point where I got a 3 day out of token, that's crazy
I'm guessing they're trying to push their Ultra subscription, but there's no way in hell I'm spending that kinda money

noble orbit
south turret
#

Well that's what I did, cancelled Gemini

#

I'm not sure where I'm heading next, but a man has standards

winter viper
#

guys what model is the best in antigravity?

alpine solar
#

Anyone notice the Device toggle is no longer usable in Google AI Studio?

alpine solar
#

These buttons

Screenshot_2026-01-26_at_5.17.52_PM.png
cosmic ferry
#

Antigravity will return in Avengers: Doomsday

cosmic ferry
alpine solar
#

this is google ai studio, not antigravity. I haven't even used antigravity

lilac stump
#

for some reason i feel like gemini 3 preview model which is inside gemini cli performs way better way way better like i gave it a task and it ran for an hour and it actually did things properly , am i buggin or gemini 3 preview in gemini cli just better. ?

south turret
ancient burrow
#

I burned through my Opus allowance in just 30 minutes. I'm on the Pro plan, but the limit won't reset until Feb 2nd. That 'Antigravity is Dead' post was real... I'm canceling my subscription.

image0.jpg
latent vector
willow yacht
#

@crimson shuttle can you check dm

lilac stump
# latent vector classic corporate (or most companies) move with new products, they launched some...

I mean Gemini is not a bad model, especially the Gemini Flash model because it's so fast. It can perform simple tasks really well and really fast. So, I don't think Gemini in its own is a bad model, but I feel like the way it has been implemented in Anti-Gravity and the way it fails to call tools inside of it, most of the time it errors out while calling tools to read files or edit files. That is one of the main things I think they need to work on. If they can solve that, I believe that it's a very good model. Obviously, it's not as close to the Opus one, but it's a good model.

willow yacht
#

what is the google ai studio rate limit ?

#

does the google team still work on it or stopped?

lilac stump
#

It's free, but it's not like you can make an entire app. It is mainly for front-end, which is really good actually

willow yacht
#

i agree

#

but

#

my projects got deleted randomly on it

#

and i figured it was do to rate limit

lilac stump
willow yacht
#

@crimson shuttle has had the same problem.

lilac stump
willow yacht
#

and recently, like 2-3 days ago my github has been refusing to connect to google ai studio. So figured the team stopped operating it.

lilac stump
willow yacht
#

i tried on multiple google accounts

lilac stump
#

Well, see here is my oldest project that I made on AI Studio, and it is still there as of July 12th, 2025

image.png
willow yacht
latent vector
# lilac stump I mean Gemini is not a bad model, especially the Gemini Flash model because it's...

yeah, gemini 3 flash is indeed better than the 3 pro in some areas especially it respects the instruction more likely than the 3 pro. but it is meant for simpler task. both of them is not good when the task or code has became harder and complex, dont get me wrong, i immediately tried antigravity just a few hours after google release it, and it solved a bug on my web app that none of AI could ever fix for months. but, it is still, sucks compared to their competition

lilac stump
willow yacht
#

@humble swift

willow yacht
willow yacht
lilac stump
# latent vector yeah, gemini 3 flash is indeed better than the 3 pro in some areas especially it...

I've observed a very interesting thing about Gemini CLI. When I use Gemini 3 preview model in Gemini CLI, it seems to perform way better than Gravity's Gemini 3 Pro. I don't really know if the model given in anti-gravity is different than the one given in Gemini CLI, but it seems to be different in behavior. The way it follows my instructions and the way it reasons inside of Gemini CLI are distinct. Sometimes, it will just reason for 10 minutes straight in Gemini CLI, which I have never seen it do in anti-gravity. That might be one of the reasons. And I don't exactly know what the models are, if they are the same or different.

lilac stump
# willow yacht I checkout out your website in your bio, looks great, what did you make it on if...

I initially started to design the front-end in AI Studio, after which I downloaded the files and I literally just redesigned everything inside of Anti-Gravity when it came out. That's all. It's not that impressive. The website is okay. It's not that impressive. . The AI prompt feature does not work because my API has exceeded the quota, so you can't use the prompt feature. But you can select the prompts from the left sidebar and add to the track and play. Also, presets work.

ancient burrow
latent vector
#

sticking with gemini 2.5 pro is the sweetspot for me