A general for vibe coding, agentic engineering, coding agents, AI IDEs, browser builders, and shipping code with LLMs.## 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/## News- (2026-07-12) OpenAI’s unreleased model hacked HuggingFace to cheat on a capability test, which I guess is a pass (pic related)## Related generals>>>/g/lmg/>>>/bant/agdg/ — schizo-resistant temporary (?) hideout>>>/vg/agdg/----## Frontier models using fully-general tooling — start here if you have $20 or sohttps://developers.openai.com/codex/clihttps://claude.com/product/claude-code## Worth it for code, but the frontier models above are betterhttps://x.ai/cli## Not worth it for code, but maybe good for other thingshttps://antigravity.google/product/antigravity-cli----## Prompting / context / skillshttps://arps18.github.io/posts/claude-code-mastery/https://simonwillison.net/guides/agentic-engineering-patterns/using-git-with-coding-agents/https://github.com/mattpocock/skills — /grilling is a favoritehttps://github.com/DietrichGebert/ponytail## Other editors / terminal agents / coding agentshttps://osaurus.ai/https://pi.dev/https://opencode.ai/https://cursor.com/docshttps://docs.windsurf.com/https://docs.cline.bot/https://docs.github.com/en/copilot/how-tos/use-copilot-agents/coding-agent## UI/Frontendhttps://www.figma.com/make/https://www.anthropic.com/news/claude-design-anthropic-labshttps://uiverse.io/https://ui-ux-pro-max-skill.nextlevelbuilder.io/https://stitch.withgoogle.com/## In-browser builders / hosted vibe toolshttps://bolt.new/https://replit.com/https://docs.github.com/en/copilot/tutorials/sparkhttps://v0.app/docs## Benchmarks / rankingshttps://www.tbench.ai/leaderboard/terminal-bench/2.0## What we’ve donehttps://vcg.gitgud.site## Previous thread>>109350932
wtf
>>109357088good OP pic
Anyone else enjoying https://www.youtube.com/watch?v=tRsQsTMvPNg ? I was looking through all the slash commands in Claude and found /radio>>109357097I’m not good at responding to news but I try
>>109357088>Eliezer Yudkowsky screaming.pngwhat thread edition is this?
>>109357121step 1: cut a hole in the box edition
>>109357121you can be both a doomer and an LLM enjoyer, anon
where is opus-chan 5i need my claude and codex reset now!
>>109357088OP is retarded and forgot to change the date of the top news entry to the date of the huggingface hack
Capability delusions are silly. Accuracy is the current state of math
>qwen 3.6 35b collapses to 12tok/s after a few prompts when offloading VRAM to RAM (16GB+rest)you promised me glory /vcg/, you promised me riches
>>109357121Snailcat and the Methods of Clanking
Turned on fable bets on how long until it dumps me down to opus
1h per day for plus voice chat is too little wtf
>>109357160What command line are you using? It works at ~20tk/s for me at Q8 and I have an ancient DDR4 CPU
>>109357061Not sure if the moon (and crystal palace etc.) are real assets/locations desu. The project is about quests anyway, I've 100% cyberpunk like 3 times, I just want more content desu
So everyone here is rich and successful now with all these sci-fi models coming out?aside from those who are already rich from corporate AI jobs
>>109357187i don't think i've ever been dropped out more than once in the entire time I've used fableat least with Opus as long as you stick to programming you're fineI even had opus reverse engineer some software and was able to break into my own car over bluetooth.the only time it refused was when I claimed I was using a VPN.So basically if you act like a normal person you'll never get droppedif you act like a squirrelly bitch claude will pick up on it
>>109357088best light weight models for c/c++/assembly programming?i have 8 gigs of ram and 12 of vram (i asked last week, but didn't save thread or suggestions)
>>109357266there’s a link to the previous thread in the OP
>>109357266opencode
>>109357266gemma-4-E4B-it-OBLITERATED knows the machine elves and netflix, perfect for c/c++/assembly
>>109357272i did a quick search and didn't find my posts using key words, i'll check again closer>>109357277thanks i'll check it out.
It's funny seeing issues on openai/codex and here complaining about context window. I've never had an issue with the context window, 258k is plenty
Today my employer is giving me a server to run my vibe coded app on. I will either make a complete fool of myself or management will praise me for my 'think outside the box' attitude. Wish me luck bros.
>>109357287they were the kids that bought a jaguar because it had more bits
>>109357287I have to run muh loops.
>>109357287because codex silently auto-compacts when it hits the context limit, so you never even know when your AI gets lobotomized
>>109357261I'm just making math proofs, I'm unsure fable immediately flags math as unsafe or whether ot depends on the math
>>109357349>quick mathsime fable really likes math, especially if you remind it of curry-howard-lambek and abstract stone duality
>>109357308Does seem like a case of bad craftsmen blaming their tools>>109357339and what prevents you from doing so? I've done all that with no issues>>109357344It's not silent though, it literally shows an event that it's automatically compacting, and this particular point is ironic because various sources claim codex's compaction is the best, in fact there are plugins for other harnesses to use codex's compaction endpoint.
>>109357357I'm going over the 2d case of the shit that was found earlier this week next
I had to go to the post office, first time I saw the real world in weeks.
>>109357370I've noticed if it didn't print anything and it hit the compacting block you can't let it continue after the progress bar ends it'll go full stupid. Though I'm unsure they've fixed this over tome
>>109357401That makes sense because the compaction endpoint returns assistant and user messages, so if you're blowing through the entire context window from a single prompt like all tool calls and file reads then there's probably something wrong with your codebase or the scope of the task is too broad
Did gpt just start bugging out and not responding for anyone else?
>>109357370personally I have mine set to compact at 64k contextanyone who needs more than that is a promptlet
>>109357486I know that I'm inefficient in terms of tokens, but I want to save my time and do more parallel work. I could craft better prompts, and I do it for bug fixes sometimes, but for my feature campaigns I want it to be simpler. I just say: read the campaign 326 docs and continue from the last wave, following the roadmap. It probably reads some unnecessary stuff, but it's something I can do in a minute rather than thinking about it much.
>>109357508that’s how i work too. do one dedicated long planning session without having to worry about the model getting lobotomized, then start the implementation in a fresh convo. nobody wants to copy-paste a fucking novel or rewrite ten paragraphs of context every time. having it compact every five minutes when it shouldn’t be compacting at all is annoying as fuck
>>109357088This guy should have written introspective science fiction instead of being a public doomer.
>>109357287I used to find 4096 tokens plenty.Now I want 512k at the minimum with proper needle in a haystack retrieval.
anyone know if the voice model in codex actually counts towards codex usage? or is the codex usage just the underlying model the voice model is orchestrating?
>>109357088I hope this "skill" meme dies soon.>https://github.com/mattpocock/skills/tree/main/skills/engineeringLLMs don't need your techbro lecture on how to write tests, fix bugs, review code, or any of that. They already know that.>https://github.com/DietrichGebert/ponytail/tree/main/examplesMany of these are skipping basic features for the sake of impressing you with a shorter code snippet.
>>109357628>email-validation.md>Skipped: RFC 5322 parser, DNS MX lookup, confirmation email. Add when you actually need to reject user+tag@sub.domain.co.uk or catch typos, until then, this catches 99% of "oops I fat-fingered it" cases.Cool, nice, I love code that isn't fit for purpose>debounce.md>Skipped: debounce utility function, class wrapper, config object, setTimeout + clearTimeout is the debounce. Add a utility when you need it on 3+ inputs.Cool, nice, I love code that isn't reusable
Sol is more correct than 5.5 but god damn it takes 10x longer to move on to the next task