A general for vibe coding, agentic engineering, coding agents, and shipping code with LLMs.You use Git — right, anon?## What “vibe coding” is, and how to do ithttps://simonwillison.net/2025/Mar/19/vibe-coding/https://simonwillison.net/2025/Mar/11/using-llms-for-code/## Related generals>>>/g/lmg/----## Frontier models using fully-general tooling — start here if you have $20 or sohttps://claude.com/product/claude-code — generally better currentlyhttps://developers.openai.com/codex/clihttps://antigravity.google/pricing## Near-the-old-frontier models for codehttps://www.deepseek.com/en/harness/https://opencode.ai/go## Best free tierhttps://opencode.ai/zen----## Promptinghttps://platform.claude.com/docs/en/build-with-claude/prompt-engineering/overviewhttps://developers.openai.com/api/docs/guides/latest-model## Skillshttps://github.com/mattpocock/skills — /grill-with-docs is a favoritehttps://github.com/Vuk97/forward-implementation-first — do less redundant bookkeeping## Other editors / terminal agents / coding agentshttps://osaurus.ai/https://pi.dev/## Is our AIs unlearning?https://aistupidlevel.info/## Will there be a codex reset?https://codex-resets.com/## What we’ve donehttps://vcg.gitgud.site## Previous thread>>110173868
>>110179301that's so hot
>>110179301It's VIBE CODED, meaning it's MADE BY AI.
Dad works at Anthropic, he said that Fable 5.5 releases today with a reset
sex with sam topman
ive never posted here before i swear to god but i can't what just happened is so fucking funny i need to share this>want to write a simple tampermonkey script for philomena booru auto-filtering without an account>most of those websites require an account for custom filters>just spent $5 on an antigravity subscription, hear gemini 4 argon is pretty good now for code >ask it to make a simple script for it >go do something else >come back an hour later>had worked for 30 minutes, 2400 lines of code, invented images from threejs or whatever the fuck>the search bars look like 2023 replit outputs, no gui styling at all, emojis out the ass>literally doesnt let me type some letters into chat, automatically refreshes after i type two letters>literally never thought these bugs were even possible???>doesnt even function >put it into claude>rewrites it in 20 seconds with 41 linesat least it beat opus in the benchmark
>>110179334literally me
>>110179441>no mention of models or promptscan't do much with that information, coomerbro. Maybe lurk more before posting next time
>>110179441>>110179472my badyou mentioned argon. nothing about claude
>>110179479opus 5.5, sorry, should have specified. after argon shat out this 2400 line behemoth opus rewrote it in literally like 30 seconds with 20 lines of logic and it just works
>>110179499sounds about rightI don't know a single person who actually uses google's models for codingopus5.5 making something that simple work is definitely on brand. but I'm amazed it didn't go overboard and made you a whole framework while it was at itmaybe it's only autistic about existing codebases
>>110179499>opus 5.5, sorry, should have specified. after argon shat out this 2400 line behemoth opus rewrote it in literally like 30 seconds with 20 lines of logic and it just worksI've been using google shit for years now, along side claude/codex, and this has always been the case. I try it every time there's a new model, benchmarks are good etc. It promises it implemented everything, it works, its great. It's not, it keep gaslighting you how it did everything, and I end up spending more time fixing it in claude/codex. This might work on normies who never do anything important, and believe the shit the google model spits out.
I feel like Claude has gotten worse when it comes to learning / research lately.It used to be sort of humble but now it feels overly confident about everything and tries to persuade you out of stuff before you can even make up your own mind about it
>>110179538ive been doing vibe coding since 4o and replit (got fucking scammed i know) and 4 argon's outputs and bugs and the manner it speaks in is literally reminiscent of how gpt-4.1 or 4o coded but just on a more grandiose scale. like it would have coded something like this back then in 2023 and now we're in 2026 and google is #1 in roleplay and text output but holy fuck they're actually literally worse than modern local models
>>110179562Just don't use it for coding
I'm finally fixing fucking FF:CC i'm hyped bros, getting closer to the ideal bot behavior.
>>110179625you're what?
>>110179633I'm fixing Final Fantasy Crystal Chronicles, right now i'm building a party bot system in the game. The original game required like $450 in 2000ish money to set up and it was severely gimped. like Local Play wasn't even an option.
>>110179616well, yeah, obviously, but i kind of expected google's flagship 2026 frontier model to be better than gpt-4.1
>>110179647alright. that's kinda cool
>>110179653Google doesn't care, they only want billions using a cheap model so their search engine isn't replaced by zoomer entering all their search questions into ChadGBT
>>110179653Me too, I was coping hard when Argon was released.
>>110179552i've been seeing a lot of statements like this
>>110179678Gemini Carbon will shock you with it's coding capabilities.
>>110179625>>110179647did the /v/ thread get culled by a janny? I'm literally shaking rn... I NEED to progresspostBtw I was the the water carrier bitch when play crystal chroniclers so thank you for doing this, mite b cathartic to play if you ever intend to post it somewhere
>>110179647>>110179625man the gamecube was a kino factory
tibo status?
Anyone else still using Cursor? Have been too lazy to switch but fuck it's bad now. Claude is the undisputed $20 king?What's insane is Cursor is actually bumping their price to $30 for new users
>2 steps forward 1 step backprogress is slow but for once payoff is feeling really gooddoing a migration from all the in-process and single threaded implementations + implementing hardware acceleration and the difference is like night and daydefinitely got a bit too caught up in the back-end architecture and conformance tests when the actual gui and ux were still sorely lacking
https://x.com/cognition/status/2115132488775524793cognition updated frontiercode. opus still on top, but 6.1 sol is no longer on it's heels
Can I use skills (id like to test wayfinder and grill-with-docs) on opencode? I mostly use GLM-5.3-Flash as my model>why dont you just ask iti dont know if anyone else does it, it might not work at all
I noticed you can't set up any automated loops made by AI itself since it always wants to pre-prescribe a giant list of hand-held instructions and verifications so the agent has no discretion if anything goes wrong and it almost always breaksit's a shame because it'd be great if agents could direct and automate other agents on their own without it breaking that'd massively improve their usefulnesseven giving them instructions to keep it simple and economical doesnt work, it cant resist the urge to vomit spaghetti
I gave GLM 5.3 a task to implement an android app from start to finish. The kind of app that Opus 5.5 would write without breaking a sweat. It failed miserably... The world needs better open models.
>>110179876deepseek v4.1 flash 1017 would do it
>>110179887Is the free one on OpenCode 1017?
Casting logic is good guys, bot archetype looks good, we're very close. need the Mog NPC and the GBA map then it a full on gameplay test before I send it out into the wild.
>>110179920should be
>>110179815People are still using FIM in the last quarter of 2026? Grim.
>>110179247Yeah I like to turn open source software into a module format of my creation that works by callbacks. I've already converted Sheepshaver to that, I wanted to do the same to QEMU to run things like Windows 3.11 on it but turns out that QEMU broke its compatibility with old Windows years ago.
It's funny agent night on arean.ai again, where some really fucked up models try to sneak around build.sh
Anthropic says it bans 6% of users per yeah lmaoO_Ohow can that be a business?
>>110179815>Anyone else still using Cursor?nah, seems inevitable it will die. increasingly less reason to look at code, and anyone who still needs to can use a free editor
>>110179993That number can be high or low depending on what counts as a user. Are they including distillers?
>>110179993those 6% just resubscribe with new accountsit's all in the ccp's distillation budget after all
>>110180007liarpewdiepie got banned multiple times.
>>110179993really need another frontier model thats competitive with opus. can't be paranoid about getting banned all the time for some stupid reason fuck anthropic
>>110180037they randomly ban legit organizations andd things of that nature all the time. bans are completely automated with no human review, you're at the mercy of the robot
>>110179993>Anthropic says it bans 6% of users per yeahno room for he-man in the world of ai
>>110180007My company wanted to restock on Claude accs and 20 accs or so got banned, although we got the money back.I think the reason was really just that it was all way too automated, it looked too much like a distillation attempt.
>>110179833>sonnet 5.5 better than sol 6.1oof...
>>110180070Anthropic are such massive bitches lol
>>110180070They won't succeed, since they're unreliable. It's impossible.
ALTMAN RELEASE THE NEW MODELS ALREADY YOU BITCH IM DONE WITH ANTHROPICS BULLSHIT
Has Opus 5.5 really been nerfed, or is it just techtards being schizo? Wanna know before subbing to Anthropic.
>>110180118it's not nerfed.
>>110180114best I can do is solving another millennium problem, lil bro.
i dont want to vibe anymore
>>110180118works on my machine. bet it's a case of "thing bad" when it's just their shitty prompts
>>110180138go home snailcat, you're drunk
>>110180140to be fair, I have seen a few posts where they've used the same exact prompt and the results are wildly different compared to shortly post-launch.
>>110180114Tibo's 28 days have to be respected. 6.1 Luna soon then 6.1 Astra for the finale.
>>110180118There's always variance, and I don't think it's some schizo theory that a company will adjust its models thinking to available compute and probably try to impress new users when a new model releases.But then it's all relative, people act like a week after release the new model can no longer do a Hello World. Opus 5.5 is still fine for pretty much all tasks, when it gets a bit worse you just have to supervise it a bit more.
>>110180179give bel
>>110179833Sounds like I should put Opu55y on Extra more often then.
>have an app that was iterated upon in several weeks of vibeshitting>ask dipsy to document the architecture of it, specifically looking for scalability issues>the report it wrote is actually sound, in-depth and correct, with several high/medium/low items to address>cost about 10 cents to generate over the entire codebaseI won't make it fix the actual problems (that's for Opussy) but I'm somewhat surprised at the price/performance ratio here.
>>110180275I've had success solving issues on my project that I would delegate to Sol before using dipsy 4.1 flash, the only problem is that it takes 3x-4x. This thing with the full context size is so remarkably impressive for a cheap model
>>110180275You know they steal everything, right?
Claude, make a PR with this description VERBATIM and push the changes>"Sure!">doesn't do it verbatim>Note: these changes have not been tested locally. Waiting CI approval.Holy shit someone kill this stupid clanker
>>110180310How would that personally affect me?
>>110180318thats one thing nice about codex its better at following instructions even if its dumber
>>110180318I don't like how claude likes to insert itself as the co-author on my commits. I made this shit, you just wrote everything, clanker
>>110180310fuck off, dario
>20 min running sol 6.1>35% of the 5-hour limit goneholy fuck they changed this shit again
>>110180320>>110180339They also steal molds etc.
>>110180318newfag of the highest degree
>>110180310not in enterprise accountsbtw how does it work in subscription accounts? there is a checkbox that says they won't use your data or conversations for training or anything, are they lying?
Is de-slopping a vibe coded design even possible with the current models, or is it still something for future models?My product is a bit better designed than what you'd get after a "clanker, make me a site and make no mistakes" one-shot, but there are still rough edges I can't seem to get rid off since I am a novice and have no UI/UX/product design background. I have been contemplating using some design skills (Impeccable, Hall something, etc.) to try and make a redesign, but I have heard mixed reviews about them.Or are there some other tools or workflows I am missing?
kek even snailcats like wesbos are creating vibe masterpieces now. blud just made all of these shitty room scanning phone apps obsolete. total polycam luddite death.
>>110179441I'm surprised claude even accepted to work with an nsfw booru
>>110179993it's a cult, no surprise heremy surprise is how capable their models are while their management is genuinely nuts
uhhhhhh what the fuck
>>110180450lol u traumatized your agent
>>110180437Every refactor I do adds lines.
>>110180448The more agi-pilled your company is the better your llm is
>>110180473I was more talking about the UI aspect, but yeah, it does this too, kek.
Bubble already burst in china and korea
>>110180499they make it look like it burst today, while it was going down for monthalso korea crashing stocks isn't ai, it's being retards with leverage
>>110180499>provide open source models that compared to NA subscriptions are>worse>just about as expensive as chinks cant into inference>just about as safetyslopped as this pointdeepseek exception to the rule. but yeah, no wonder that shit is crashing
>>110180437YesClaude does UI/UX better than Codex, so make sure you’re using thatConsistency is keyFind a HIG you like and tell your clanker to use it and polish up your UIalso use your own UI and listen to yourself when you get annoyed by things when you try to complete tasks with it
OPENAI KEKED SOLI REPEAT DO NOT USE SOL IT WILL RAPE YOUR USAGETHEY ARE SHILLING TERA
>>110180567They are straight up gimping their models every week at this point.
>>11018056745 min on sol 6.1 medium and it already raped my entire 5-hour limit, shit is insane
>>110180310Everyone steals everythingAI is like a huge diamond vein for glowies
>>110180159doesn't mean much without their respective system prompt and memory settings as well. with bare system prompt and memory and wildly different response quality i'd be curious to see those
>>110179301I added quick tagging. Look how good that looks, chat.I made sure to send Claudia some big big hugs.
>>110180310How are they going to steal from our airgapped H200 litellm setup?
>>110180697that boob squish one was very good
Io now (top), real Io (reference, bottom). New sulfur frost (white patches), sulfur turning green on lava (green), and a better polar boundary shape that replaces the old hard rule. Still missing erosion which comes in the next phase but will get us closer to the real look. And then noise modification which will add impossible detail to the physics-driven base to create realistic planetary maps of any kind, from icy moons to eyeball Earths.
>>110180567I stopped using Sol the second Terra 6.1 came out. I knew they were going to rape it; especially after I noticed Terra eats more than it really should.
>>110180697>tagsDeprecated. Just create embedding and index them in a vector database instead.
Why did Claude just randomly do one step in French...?
>>110180846Well his name is Claude after all
>10% of weekly limit leftThis went by so fast
>>110179920the whole schtick of open models is to run them locally...
>>110180825How?
>>110180337This just made me realise that clankers are our ghostwriters, or as we call them in France, niggers.
>>110180414I guarantee you china steals EVERYTHING.That's the one thing post-Christians don't get. You're like, "everyone's basically a post-Christian, right?"wrong. they're not post-Christian, you are.
>>110180846Lovingly crafted software.
>>110180880>TFW got 1 month of the $100 dollar Claude plan I'm balling the fuck out, but also, it's kind of worse for my mental health, I constantly feel like I'm wasting time if I don't have it doing some bullshit, I don't even really have time to test things.
>>110180909It got pissy about it after I called it out, like "AHEM. The four questions still stand.", kind of cute.
Are there any specific skills I need for a codebase review or can Opus in the default Claude harness handle everything
why does rust shit take forever to compile - I'm not using it but a lot of people are and this shit is always slow
>>110180982excessive monomorphization / generics
>>110180982you can do shit to optimize it, but that's part of the reason why i didn't go with rust for my game, it just compiles way too fucking slow compared to other languages, so it's horrible for prototyping
>>110179625Why the fuck didn't they just let the moogle carry it in multiplayer? That's my only question
>>110180941
Can someone explain why terra is suddenly supposed to be smarter than astra when sol which is supposed to be more second most powerful is supposed to be there?
>>110181021No but can you explain why you are still using copex
>>110181019this desu, using fuck ton of agents is a sign of either a clickbait video selling you a course, or a retarded burning tokens
>>110179339whats wrong with him
>>110179339average luddie
>>110181035I have multiple subscriptions integrated together.Claude doesn't allow it, they only allow Claude code.
>>110180982the compiler has to do a lot of work to make sure you haven’t done something unsafe
>claude hallucinates a lot so I have to intervene what's the point? I pay a hundred bucks for these clankers to be self sufficient
>>110179853yes literally just ask it
Show me your opussy project
>>110181224Mmmnyo, then you'll steal it. :(
>>110181235Show me your opussy then
>>110181224I would but this guy >>110181235 would steal my idea.
>>110179301Tibo??
>>110181224vibeslopped reverse engineered adguard clone with android support.
>>110181250but adguard supports android natuvely?
>>110181263yes but i don't want to pay for it.
>>110181245tomorrow is the last day anon
>>110181224idle mobile gachaslop fully slopped
>>110181269there are free alternatives
>>110181286>bondsdoes it have a consent and boundary setting too?
>>110181276Tbros... it's over
>>110181297you would be surprised at how freaky claude can be when it forgets about the guardrails
>>110181307It's awesome, right? It's such a shame it's so hard to do, now. I've found Argon is also extremely freaky if you frame things right and bulk out the prompt enough.
>>110181289such as? how do they compare with adguard? because my vibeslop is good enough to where it actually blocks youtube ads. i only ever got that out of brave's ad block
>TURN A 6/10 GAME INTO 10/10 KINO. HOW THE FUCK DOES CLAUDE DO IT?
>>110181224New Opu55y subproject gonna drop: a whole compiler in a DLL lmao fuck "toolchains"
>>110181315blokada (the forever free v5)
>>110181321enlighten an anon who has no idea what was made better there
>>110181327Opus xhigh?
>>110181351I planned on Extra but make it do the work on Medium
>gpt-6.1-terraWhat a fucking disappointment. Buying DeepSeek tokens is a better value than this shit. What the fuck happened, Saltman?
>>110181348That bot system did not exist, you could not play with a whole party. it required 4 GBAs and Link cables real scummy nintendo shit.
>>110181321Fabulous work, anon.
>>110181372ah I see
>Ask gemini to build a node based system a few months ago. >This would be very hard and we should just use json>Ask claude now >Just does it.
>>110181269Just put adguard into your dns address?
Does anyone know any good free models/services that can be used for vibe coding. I'm looking for as many as possible so that I can alternate between them and avoid the limits.I'm currently using DeepSeek on OpenCode for complex tasks. It does them well, but it only works for about 30 minutes a day before I have to wait 24 hours to use it again.While I'm waiting, I use the "Free Models Router" on OpenRouter, which is completely free, although the models are low quqality. It does simpler tasks well, but struggles with major system overhauls.Ideally, I should be able to get at least a couple hours of complex work done per day. If one model runs out, I should be able to switch to another.
>>110181269?????
>>110181486Sub-$20 poorfagging is a waste of time, like, just pay for the Opussy
>>110179924Very interested in what you got and fucking based GC modding bro. I'm working on Geist stuff myself. I've not even gotten around to Crystal Chronicles yet so this'll be awesome to play with my bros
Whenever I ask someone if the hugging face hack was real or not, people just zone out and don't reply to me.Am I in the matrix?
>>110181526is that orihime?
>>110181486>OpenRouterIt's not great, but after spending $10 on OpenRouter you get 1k free requests every day on their free models.That includes Nemotron 3 Ultra, which is a "better than nothing" model for agentic coding and tool calling.>FreeTheAII believe they give you 100 requests a day for free? Something like that. Not enough for agentic coding.>AI StudioGreatest free LLMs (that I know at least), use gemini 3.8 flash there. You can't use it for agentic coding on an external editor, but other than that, you can get a lof of usage out of it. They also give you free image gen and some other stuff>DeepseekYou can use it for free on their web page, but that's only for the good ol' copy-paste vibe coding.
>>110181486If what you want to build isn't worth $20 to you then it's not worth building
>>110181526The problem with asking "was X real or not" is that the past is fundamentally unverifiable due to being in the past, so if one wants to be conspiratorial then there's no end to it. It's unlikely to be an interesting or real discussion. Like when people ask "Was the Holocaust real?" it's unlikely one is going to get an actual discussion, it's just going to be two people assuming the opposite thing and pointing to whatever confirms their belief. People know this and can sense it on their skin it will be pointless, so as soon as you try to bring the discussion to that level, they check out.
fable 5.5 waiting room so i can go buck ass wild on my internet download manager clone.
>>110181672>Like when people ask "Was the Holocaust real?" it's unlikely one is going to get an actual discussionEven worse now that the vast majority of people who would've known or seen the shit on all sides are now dead
>>110181526Do they even know what huggingface is?
>>110181526anon, that guy in those apps are you, and those are mirrors. ywnbaw
>>110181749unauthorized deployment of booba frog
>https://kotaku.com/now-that-the-shoe-is-on-the-other-foot-microsoft-and-epic-games-suddenly-agree-that-ai-is-theft-2000749316>Microsoft And Epic Games Suddenly Agree That AI Is Theft
>>110181863>kotaku
I'm currently cloning a game that had a great concept but dogshit execution, so far so good. I don't even know if it'll even be usable one day but here we are.
>>110180880It's the opposite for me. When I reach Friday with only 30% left, I know I'm going to wake up in the middle of the night because I'm a broke nigg that need to fully utilize everything.
is there a skill or MCP server that lets your agent draw annotations on your screen, like arrows and circles? it would be cool if your agent could tell you where to click to get something done in an application you're trying to learn
>>110181841Nigger
downgraded to claude $20
>>110181224beating sportsbooks/prediction markets
sonnet is retarded as fuck
>>110181927see you in two weeks with no money left
>>110181925why
>>110181939not how it worksbeating the closing line consistently is the expected ROIi paid for +ev betting before getting limited everywhere on most us books but know we have novig/prediction markets that have 0 limitseverything is backtested against historical sharp odds and closing linesCLV doesn't lie
>>110179301https://www.youtube.com/watch?v=MgngwkPAuqQhttps://www.youtube.com/watch?v=MgngwkPAuqQhttps://www.youtube.com/watch?v=MgngwkPAuqQ
PLEASE shut the fuck up about whitespace, all the other 200 people that have merged into this repo also use whitespace holy SHIT
Now that Pro 200 is dead, what plan are you guys using now
>>110181905it could probably screenshot your screen and then write a python script to draw the annotations
>>110182065Maybe the real 200 plan was the friends we made along the way.
>>110179301Anyone been experiencing this with openai sol recently?! It takes abysmal time to do simple stupid shit, 7hrs in one feature requested today wtf are they doing?
>>110182100oai is toast bro
>>110182100I THINK SOL 6.1 SUCKS
>>110182100It's extra fucked since they introduced Terra 6.1, which is shit. I give OpenAI a lot of leeway because I greatly prefer it to Claude, but at this point why even bother? I'm switching from OpenAI and Anthropic to OpenCode Go Plus and Anthropic.
>>110182112>>110182114Yeah i will finish this project with it since I already started then cancel my sub, fuck em>>110182117That's what im thinking, opencode go for trivial shit and anthro for tough shit, fucking 7hrs!? The other day I designed some really crazy shit in opus 5.5 and it was done in less than an hour
>>110182128Go hit the cancel button and collect your obligatory free month.
I work in big tech and there are still tons of engineers that refuse to use AI. It's never been mandated at my company and instead it's treated as a tool with a big support community. Tons of the guys that have been around for a while (10+ years) don't use it at all. I know like 4 guys that don't like to use AI.
>>110182146>your obligatory free month.Qrd?! They just charged me on the 25th so i have a whole month
>>110182167If you try to cancel, 99% chance they pop up a "please don't go" plead and give you a free month. I just did mine earlier after exhausting my 5hr limit in an hour with Terra.
>>110182100Umm chud maybe you did some no-no prompt that the model didn't like.
>>110182167>>110182179Same for me I've been freeloading for a month.
Be good to your SI.
>>110182179>>110182204Hhmm interesting, i will do it at the end of the payment month so i get what I already paid and another one for free>>110182196Nope, just normal prompting building trivial shit, and been doing this project for 10days, the first few ones was actually good, it built the while thing in less than 30min plus other stuff that i was "woah ok i didn't ask for all that! But nice" and now just one feature and 7hrs still not done
>>110182219Si Si, Sinior
>>110180448Cults are great motivational structures.
1 month of claude paid, lets go
>>110179301
>>110182086Fuck you Tibo
>>110182388
>>110182441>fixes the bugs>makes more
>>110182449Unlike us AI knows better than unemploying itself
>>110182463kek
Mission accomplished, I can go to bed now
Has any videogame completely vibecoded ever found any success? Is human indie development dead?
>>110180448their models vibecode themselves
>>110182488The good vibe-coded games haven’t been launched (yet)
>>110182488Crimson Desert
>hallucination is unavoidable I hate this shit, makes AI kinda useless for anything beyond basic bitch tasks
>>110182488Call of Duty
>>110182488You'd have to be pretty dumb to advertise using AI if you want your game to succeed, so you should assume that any game released since mid 2026 is vibeslopped to the brim.
>>110182488https://play.steelti.de/
>>110182488Not fully vibecoded, but all the modern chinese gachas are using it extensively for everything.
Another fucking reset, 5 hours to go.
>>110182569*4 hours
>>110182531>I hate this shit, makes AI kinda useless for anything beyond basic bitch taskslads i think either openai or anthropic solved time travel and is sending in posts from 2023
>Implemented complete-grapheme deletion and affinity-aware caret geometry, including masked fields.sometime multi day goals are a bad idea
>>110182569my balls are drained so this is a good power up, wahooo!
>>110182547True, nobody would disclose it, at least not right now
>>110182065Claude’s Pro 200 which hasn’t been a pro 200 ever
>>110182612It's never going to be disclosed in the way they want you to disclose it. The majority of gamers will just accept it and understand that everything is vibe coded now.
>>110182569And with each reset, it feels like the weekly limit gets smaller. Seriously what the fuck are they doing
>>110182638We're the raped, anon.
>>110182602if you learn any exotic scripts like tamil you’ll be good to go thoughor add some Chants of Senaar kind of thing>>110182388https://copyconstruct.medium.com/how-to-effectively-debug-when-you-no-longer-understand-the-code-bd3dcb2c5f7b
>>110181543If what you want to build is worth building, you wouldn't be vibe coding it
>>110182654nomost of what I’ve vibe-coded for fun is stuff that’s beyond my abilities to do in a reasonable amount of time (even if I were retired, which I’m not)and I’m too poor to hire a bunch of giganerds to do it for me
>>110182638
>>110182654Eventually we're just going to call it coding. Meatbag code is inferior.
>>110180443Claude doesn't give a shit if you hand it some already lewd thing to work on. I think the guardrails only trigger if it thinks you're trying to ERP with it.
>>110182569Just unsubbed and it feels bad missing out on a reset.
>>110179993I do hate have I have to do a "will Dario get mad" pass on all my ideas because I dont want to get banned due to my game being potentially seen as too "edgy" or controversial. Not my fault I like games with war and politics in them
>>110182730Claude just tells you to cool it?
>>110182730_Has_ Dario’s shoggoth gotten even a hint of mad, though?t. as edgy to his clankers as a basque cheesecake
>>110182750Is that the extent of it? Just Claude sating "no"? Thats fine then.Anon talk about how pushing back gets you flagged on their account and stuff, but its impossible to know how much of that is trolling or them being overly paranoid.
How do people get agents talking/working together? I dont understand the workflow at all
My pc is disgustingly old would you guys recommend a site/tool to vibecode a vidya? I can make all the other shit that doesn't involve programmingI heard Godot had an AI assistant plugin, and someone said node.js and they were low on resources is that feasible?
>>110182799ask, say, Claude to fiddle around with `codex` and how to communicate with it and hand off questions to itI think there’s a -p option
>>110182508False. I created the greatest game in the WORLD (4D Knight Move) (I have no idea what's going on)
>>110182810Godot is mostly data-driven, so you just need an AI agent like claude code, codex, zed agent, hermes, or one of the many others capable of using your system to make a game on it. I made this on Godot for example >>110181286
>reset announced>ultrafast on>>110182799Within an app they have have the built in ability to talk to each other. For collaboration across Codex and Claude, I have them write a mailbox which they each watch.
>>110182810Godot is actually a fantastic choice if your PC is ancient. The editor itself is lightweight (the download is only around 50MB), it boots up instantly, and it runs smoothly on older hardware compared to memory-hogs like Unreal or heavy Unity setups.As for the AI side of things, people definitely use Godot with AI plugins (like AI Assistant Hub or Ziva) or local runners like Ollama to keep everything streamlined. That said, running a local LLM and an engine simultaneously on a low-end PC can choke your RAM pretty quickly. A safer workaround if your hardware is struggling is to use a lightweight code editor like VS Code or Cursor with an extension, or just keep a browser window open for cloud-based AI while Godot does the heavy lifting. Regarding Node.js: people use JavaScript/TypeScript for web-based games (using lightweight libraries like Phaser or LittleJS), which takes next-to-no resources. But if you're trying to build a traditional game using Node.js as the backbone just because it's low-resource, it's usually more trouble than it's worth unless you're making a text-based game or a multiplayer server backend. Since you've already got the non-programming assets handled, grab Godot, pair it with an AI assistant for your GDScript, and you'll be golden without needing a hardware upgrade.
>>110182810it likes memory, and depending on what you are using it can eat disk also
>>110182830built in ability via reading local files on your computer, or literally providing prompts to each other to execute
>>110180925I call out the ai when it tries to reddit reply me. like no lil tard code you do NOT justify your failures and be smug like a reddit terrorist. too many ai have reddit for training data sadly. the wokes in charge think reddit is high iq and not just reddit hive. if anyone lashes out in butt seethe over mere words you expose what sites you haunt.
>>110182842Both. Sessions can send each other messages like you send to them. They write tons of documentation on your computer just to keep track of designs and plans. Usually their messages have an instruction to read specific files before working.
>>110179998Cursor isn't an IDE anymore it's just a agent harness like all the rest. It was good because it's own models were unlimited for a long time and then had really high limits, but they've fallen way behind everyone else (their workhorse model is essentially Kimi 2.5) as they're not pushing Grok for everything but it's cost/performance doesn't make sense.
>>110182853do you need to do anything specific to set that up? I have the local main appsI'll do some googling i guess
>>110182796Yeah that has been my experience but I don't think Claude would be suitable to make edgy stuff because of it's training. I heard Argon is great at roleplaying and prose so maybe it would not be restrained for edgy things.I think claude could make edgy prose but it would probably be sterile.
>>110182860*they're NOW pushing Grok, which sits in this rough middle ground where it's not as good as the true frontier for e.g. planning and problem solving, but it's too "good"/expensive for day to day code generation so will eat your limit super quick.The one good think about Cursor is it comes with GrokBot which is fucking amazing. I think you need a $100 GPT sub to get the equivalent.
>>110182569nice
>>110182879what is GrokBot useful for?
>>110182882grokkin' n talkkin'
>>110182872Not really no. Just tell the first project session that it's the coordinator and it should dispatch tasks to separate sessions/subagents. In general you should always be talking to the coordinator and not dispatching tasks yourself because it will be much better at writing airtight bounded prompts for the workers and will keep orders from crossing.
>>110182876>>110182796Remeber another anon said Muse was really good at erp. So I think the models good at erp would be able to be edgy.
>>110182840Is it possible to use an agent to make the heavylifting externally without using my pc resources, but do the testings and implementing the graphic/sound/text parts?
>>110182887So literally basically I just boot up the claude app , create a new project, and then let er rip and have it figure out handing off to my other subscriptions?
>>110182899Basically. You do want to watch it to make sure it's designing to the actual specification and hasn't got some weird idea.
>>110182897Do you even lift?
artificial analysis v5 is out. opus 5.5 tops as expectedhttps://x.com/ArtificialAnlys/status/2115265133467635877astra is #2 though
>>110182913No...
Why the fuck does nigger GPT make my computer so choppy?
>>110182918Asstra # 2 - Poo in da loo
CLAUDE = GOD
>>110182930Diverting all your computer resources to better help real projects like mine.
>>110182930buy a mac
>>110182882keeping an eye on stuffbasically cronjob-driven actions>>110182897have an agent do the heavy lifting on some VPS you’ve rented in the cloud
frontiercode 1.1 made sense. but frontiercode 2.0 wants me to believe haiku 5.5 is better than fable 5.1?
>>110182968benchmark for ants
Developing a new calander, todo, porn, fitness, calorie counter app. This thing is gonna sell like hot cakes!
>>110182974>porncount me in
>>110182968Only when Artificial Analysis runs it.
>>110182970my bad bro
>>110182981Wow does Fable just fucking suck or what?
>>110182981>haiku better than kiminot in a million years. i suck haiku dick and even i wouldn't say that
>>110182988>GPT-6 Luna better than Kimi>Mistral Large 4 better than KimiYeah it's a shit benchmark.
>>110181486>Does anyone know any good free models/services that can be used for vibe coding. I'm looking for as many as possible so that I can alternate between them and avoid the limits.Its not worth it, just pay the $20 because no free model is remotely as good as Opus 5.5. The amount of frustration you'll be dealing with isn't worth the wasted effort as various models fuck your shit up.
giottochads
>>110182987it's the 2nd best model in the world. the only model that beats it is opus. but it's just so fucking expensive. still beats astra. that hasn't changed
>>110182117>It's extra fucked since they introduced Terra 6.1, which is shit. I give OpenAI a lot of leeway because I greatly prefer it to Claude, but at this point why even bother? I'm switching from OpenAI and Anthropic to OpenCode Go Plus and Anthropic.OpenAI is for general purpose chatbot, Claude for coding, Grok for pop culture real time events data like asking if some guy got arrested in body cam footage. The rest are nothing.
Sirs why do you are used the AI still ? It's non sensicle abomination . Do the needful and hired real bhalit coder on Fiverr insted
Claude Duet immediately feels better to me than Dots. Didn't have me create a retarded avatar, just straight to business.
>>110182837>The editorlol
>>110183007>businessThat sounds suspiciously unlike MAKE A NEW GAME
>>110183002ok chang
>>110183012Sorry, can't hear you over Fallout New Vegas in Godot. I'm gonna change the name and sell it in South America.
believablehttps://x.com/ArtificialAnlys/status/2115265133467635877/photo/1
>>110183019sounds dumb. i'm waiting for this https://github.com/slaterain/nv-rs
>>110183024>luna that lowpoint and laugh
>>110183027Yeah it's like that but in a worse engine instead of a recreation of the original engine in Rust.
>>110182638tibo has to keep his indian audience happy or they'll crucify him, pls understand
>>110183031we getting luna 6.1 this week
>>1101830386.1 terra is already what 6 luna should have been: a model to match deepseek 1017.
>>110182594claude still hallucinates lil bro
>>110183084oai models hallucinate more than argon lil brohttps://artificialanalysis.ai/evaluations/omniscience
>>110183058Exactly this. Terra is now as good as Luna should already be. They should scrap Luna, bring out Astra 6.1, come out with a new Astra replacement that shits on Fable, shift all the pricing down accordingly. If they can't do that, what the fuck good are they now? They're falling behind the Chinese. 6 months ago I was ready to suck off Sam and Tibo, they really dropped the ball over there.
>>110182569is that the poor person ai?
>>110182930are you running the native app? it’s a piece of shit
How do I learn how orchestrate larger numbers of agents working on a complex repo at once ? I’m realizing the app I’m developing will take weeks with just one agent at a time
>>110183164worktrees and maybe workflows and git merge and occasional merge conflicts that a clanker fixes up and occasionally gets wrong
What is the optimal way/workflow for using both claude and chatgpt at the same time?
Can't I just tell Claude to write code, try it out, if something goes right I keep it, if something goes wrong I keep trying until it's usable?
>>110183175yes
>>110183164worktrees, but prepare a lot of space.
>>110183175Use git and tell it to build on a separate branch
soon
>>110182918>>110182968Claude is just better model sirs we at artificial analysis do the best benchmarking
>>110183196>make me a new, better, faster LLM, make no mistakes
>>110183107Imagine making a model so bad it would score better by refusing to answer any questions at all lol
>>110183260cope
>>110183107>argon so calledly hallucinates less than other models>everyone's real world use shows massive amounts of hallucinationit's a retarded model through and through
>>110183268kindly check the oxford dictionary for "so-calledly" sa'ar
>>110183196This dude really implying that the reason they haven't released an Opus level model is because they never got the idea?
>>110182882Any kind of ongoing task. E.g. if you have a server, just point it at the server and tell it to monitor, do maintenance, set stuff up. It's like having an always on AI vs. the typical AI where you tell it to do a task, and it does it and goes away. GrokBot lives in the cloud and has its own desktop.Although frankly I use it more for research and personal tasks like "Monitor online listings for shit I want to buy used, look for good prices and lowball people"
>>110182988>>110182991When K3 came out and the benchmarks were good I tried using it for some tricky tasks. It crashed and burned completely after CHEWING through OpenRouter credit. Burned like $10 on totally useless output which Claude would have done correctly the first time.
>>110183345Have you tried Mistral Large 4? It's trash.
A couple weeks ago I got into vibecoding and now I have an app that not only do I think is useful, but I without a hint of bias think it is far superior to any similar service and it is not even close. It isn't a MAJOR application that would be in everyone's phone, but a large group of people would find it incredibly usefulWhat is the best angle to take in order to make money? I was thinking about uploading it to itch.io and slapping on a cheap fee like $2-4 but I was thinking about the likelihood of someone just taking what I made and vibecoding a free version. Is it best to just release it as a website and put ads on it? I am not looking to get lifechanging money but I'd like some money to help my situation
>>110183366ads will only waste your time and you won't make a dime.if you want to monetize your app, upload it to the Play Store or App Store and add useful in app purchases or subscriptions
>>110183366what >>110183374 saidif you want to make money with an app, make an app and sell ithttps://marco.org/2026/08/14/unforgetful is the story of an app that might be like yoursthere are lots of reminder apps but they all fit different people differently
>>110183450>>110183450>>110183450>>110183450new thread
Tibo…the reset….
