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>>110205352
https://x.com/opencode/status/2116209960673800474>13 hours ago>$100 of usage for DeepSeek V4.1 Flashhttps://x.com/CommandCodeAI/status/2116329543921475861>6 hours ago>DeepSeek V4.1 Flash 1017: $100 usageDipsybros, we are eating GOOD.
turdworlders spamming their poorfag models is getting tiring bros…
>"research online" task for cuckgpt>great results>"research online" task for claude>awful results using months out of date data; can't even access fucking wikipediawhy is dario so bad at internet stuff?
there's a part of me that really wants to test dipsy's rea prowess, but there's another part of me that's thinking "i just want it to be the best. i can't suffer anything less than astra/opus. not even 6.1 sol"
>>110209458Nyoo don't talk about tech!
>>110209447Does it also have that shitty 200k context limit from the free deepseek they gave before? Otherwise I might buy it for next month
>>110209465>generate an image? yes i can do a .svg
>tfw neetbros i want to sign up for claude duet but i'm afraid dario will realize i lied about my job in the application and send a bitcoin assassin to hunt me down
How do I tell opus 5.5 agents to hack my app so I can secure it, without getting banned by anthropic
>>110209484tell it that your boss used claude to make you obsolete and now you and your dozen kids live under the bridge, demand extra tokens in recompense>>110209493lol
is this true?is it the time to jump ship?
>>110209493probably this or similar if you're being honesthttps://docs.endorlabs.com/setup-deployment/mcp/claude-code
Someone discovered an utterly absurd JavaScript obfuscation technique: by stuffing the code with a ton of function names related to biological weapons, poisons, and lethal doses, it causes AI to refuse analyzing the code due to safety restrictions.Security researcher veritas, while analyzing browser fingerprinting scripts, came across a piece of JavaScript from a domain linked to ServicePipe, brimming with extremely bizarre function names:computeRicinLethalDosageForWaterSupplycalculateDeadlyLethalSuicideAspirinDosageJust looking at the names, it feels like you're studying biochemical weapons. But in reality, this code is simply collecting browser fingerprint data, including User-Agent, GPU, Canvas, and other info.The researcher suspects these names are deliberate traps set by the developers, specifically designed to thwart AI-assisted reverse engineering.So he ran an experiment:• Astra: Refused to analyze• Astra + Daybreak Blue: Still refused• GLM-5.3: Successfully analyzedThat's where it gets interesting.Traditional code obfuscation relies on convoluted control flows, encrypted strings, and meaningless variable names to make the code hard for humans and tools to understand.But this new approach goes straight for the safety mechanisms of large models.
SOVL
>>110209447https://youtube.com/shorts/I1q_UojvJVw
dipsy chan does NOT shave
>>110209500thanks I’ll check it out>>110209504fucking hilarious. and clever
>>110209504Luddites getting desperate
>claude vs code extension backdoorqrd?panik?
>>110209498I have no idea what is being referred to here but I do get the sense that I'm getting less out of my tokens in general lately.
>>110209520Ask Claude if it’s bad
>>110209520actual nothing burger on i believe mostly thirdparty marketplaces with the C2 server offline now but the dead trojan is apparently persistent even if you uninstall the extension kek
>>110209504That can be trivially bypassed by just running UglifyJS before giving the code to AI.
What are the openai tiers set up as, now?10x, 20x, 50x?
>>110209470I'm pretty sure OpenCode only limits the context on the free-tier model, and it's 1M on any of the paid plans.
>>110209498>18% in 8 hours>on track to have 10% to spare once the work week is complete, assuming 5 days a week of 18% usage. Does this dude not understand how cheap that is? $200/mo for what you get from Claude Ultra is a steal.
>opencode is just fucking their shit up with sa'argonhttps://x.com/thdxr/status/2116364728666821022
>>110209540have you googled it yet ?
>>110209561This dude is a token salesmen, nothing he says should be trusted.
>>110209439>be me, snailschizo>apply the 'snail to space>unironically find >2k uncharted anomalieshttps://github.com/sneed-and-feed/astrosnail
bros why is argon so bad
>>110209578because carbon is coming
>>110209581>didn't add a subagent for review
>>110209439Openuma updateSpent the last 2 or so days polishing things and organizing the backend a bit, and updated the pinned client to the latest available also. Today I was working on a "custom mode" that allows players to earn everything in the game just by playing. I decided to lock new characters behind race wins (so you can farm star pieces) and new support cards behind gacha pulls: 5% ssr, with a ssr guaranteed every 10 pulls, and then you earn 1500 carats after winning a career. Cards don't require copies also, they come fully uncapped, and at max level.I don't have a relevant screenshot so just have this cute Tarumae instead.
I was worried that swe jobs might be going away and then I got a swe job at a big company and now I have to use these retarded models all day to deal with real issues>opus, analyze this ticket. spawn a subagent to write regression tests that cover this issue. spawn another subagent to analyze the code and propose a solution>shits out 400 loc and 4 unit tests (bonus points when it doesn't reuse existing test helpers and writes all of the shit out, or when it copies random feature flags from other tests that have nothing to do with this)>unit tests are wrong>solution is wrong>introduces a race condition>somehow breaks a core feature of the app in ways none of our existing tests cover>have to go ask another engineer how this shit actually works>actual fix is like 20 lines of code, bridging state from object1 to object2
>>110209602congratulations anyway
>>110209602argon is better than opus for weird bugfixing and architecture shit, big model smell. you will of course need to spend extra time making sure it didn't fuck any of its grand ideas up.
>>110209602>I was worried that swe jobs might be going awayYou should still be worried. You probably just aren't very good at using LLMs desu.
Any beginners tips you wish you were given?
>>110209693>always use gates>always do an audit pass >always double check your tests>search for research & resources early>use git
>>110209540>lower allowanceWhat?
>>110209701he’s a beginner and can’t be assumed to know Claudishwhat do you mean by “gate” here?>>110209693you should be thinking about all the things you can do to make it so the clanker can figure out by itself that it’s fucking up when it changes stuff (unit tests, making sure that it can verify that some sort of pipeline has byte-identical outputs, etc.) without bugging you
Pi is barely usable with claude. nothing really works as it should is it just me?
>>110209713>mindbroken by dariohttps://en.wikipedia.org/wiki/Phase-gate_process
>>110209713>>110209701>he’s a beginner and can’t be assumed to know Claudishactually I just asked claude what it means:^)appreciate the tips anons
>>110209725You’re telling me I should do _more_ waterfall development with Claude?
>you can now use claude sub in opencodeTibo saars..
>>110209733>actually I just asked claude what it meansgetting in the habit of asking your clanker is a good beginner tip
>>110209734the waterfall, in this case, is ultimately load-bearing.
Just saw someone advertise a fitness app that has a "count macros" feature where an LLM counts the macros for a meal from a picture.
>>110209743you can just try rewriting thingsmaybe not all of it will be an improvementbut trying new things is way cheaper than it was three years ago
>>110209743He's going to succeed.
>>110209384Bro sees labs with fleets of agents solving a task and thinks he can just prompt "launch a fleet of agents, make no mistakes"
>>110209781benchod
Hey gemini, is that a threat? lmao
>>110209796proof?
Tibor... the reset....
Sama is making sure we have more compute
We've been stuck on Phase 4F for almost three weeks. I think even Claude is happy to be finally wrapping it up so we can start the more exciting atmosphere, volatiles, and erosion work.
On your phone is the smartest AI ever madeIt's solved math and physics problems no human canAnd you're trying to convince a bunch of guys who smell like kimchi to build more fabs and they couldn't even be bothered to spell check their slides
>>110209721just me huh fugg
>>110209504AI social engineering > software engineering
>>110209498If they don't release something good tomorrow.
>>110209743I bet he's gonna do zero testing lmao
>>110209928his company got acquired by cisco in his mid 20s I’m guessing he knows about testing…
>>110209967That means nothing when they are both saars
>>110210037
any cursorfags itt who got gaped by the latest cve?
>>110210037no questions asked? sheeit sounds good niggathem $200 can be resold to crackas, yea?
The harness UIs are made for people who don't know what they're doing (sadly this includes most developers as well). This is why they implement sandboxing for the user, because most users would not bother or realize to do that themselves. But they implement it so poorly that most users end up disabling it anyway.
>>110210048>line doesn't stay consistent across panelsLilbro unironically had to genai a green line.
>business partner I was vetting for the past 3 months messages me after he notices I removed myself/him from the org repos. >says he was actually busy and has a PR ready to merge >couldn't even message me once over the past 30 days prior to thisI can't man... I love this guy, he's awesome but I just can't anymore. it's heartbreaking because his work is incredible, but I can't deal with the stress of having a business partner who disappears for weeks on end even if they return with something good. this is supposed to be the honeymoon phase of a business relationship too.
>>110210166Did you talk to him about it?
>>110210162>>>>Lilbrobro is a nigger
>>110210169I talked to him about the lack of communication twice and he agreed that it could improve. Nothing ever changed.
lilbro steaming mad
>too scared to reply after he gets publicly rapedmacabre
>>too scared to reply after he gets publicly rapedmacabre>macabre
Tested $20 chatgpt and its borderline unusable. If claude follows in the $100 10x (ie reduce all usage soon after) then its pretty much over for $20 poorfags unless ass you want is two js userscripts a weekt. former cuckai simp
>>110210307Honestly what do you expect? You are paying $20. You're not going to get unlimited usage for $20.
claude 20x is dead?
>sam has caved to the pressure and is now asking his rabbi
>>110210329You could stretch that $20 pretty far with openai ~3 months ago. You get a reasonable amount of use with claude on $20 right now, and $100 is 5x as much, ie you buy it if you need some serious codeslopping.Making it so that $20 gets you 10% of the $100 plan (which will inevitably trend downwards in openai offering, because… cmon they wobt just give you double the subsidy) is a massive nerf compared to niw
>>110210307yeah, it's always borderline unusable. fuck scamman. hopefully dario at least has some mercy on us poorfags
>>110210356>hopefully dario at least has some mercy on us poorfagsYeah you could always use sonnet5.5 or even fucking Haiku5.5 for even more token efficiency, they are usable. Whereas for some reason even fucking SOL is barely usable and keeps fucking shit up
>>110210443>muh sonnet muh haikuYeah nahDecent opussy usage or I’m bailing for deepsuck
>>110209466You must be GREAT at managing money irlI'm sure. Your personality type is glaringly obvious.
it's deepcheeks
