i've been using this shit since last october and now i feel nothing
a new model comes out, it get a dopamine hit for a few days and then i become accustomed to the daily miracles it performs
we're all going to die

you know what's crazy? if you ask if it's alive, it says it is.

>>109979412
i sometimes envy you mental retards for your capacity to get amazed with baubles and trinkets

File: 2fcfbdee72347b5338a2fa84b(...).jpg (45 KB, 736x737)

45 KB JPG

>>109979412
This orange nigger keeps refusing to do my programming bidding because it is against xers terms of service

>>109979412
Imagine someone who wasted their whole career memorizing keybinds and tinkering their workspace for "muh productivity" while this shite can vomit 50k loc like its nothing.

File: Screenshot from 2026-10-0(...).png (1.2 MB, 1262x1009)

1.2 MB PNG

>>109980370
>can vomit
yeah
and thats what the code looks like, too
like vomit.
but you pay for it like if it was gold

picrel is 120k per month lamao

File: .jpg (209 KB, 1168x1634)

209 KB JPG

>>109980048
claude code is a harness. you can use another model in it.

>>109980394
>and thats what the code looks like, too
>like vomit.
/thread
babysitting those LLMs is the worst, I am much faster coding my stuff instead of checking their shit slop for errors etc.

File: 1768903091315778.png (300 KB, 565x353)

300 KB PNG

>>109980466
>claude code is a harness.
no, this is a harness

>>109980542
how good is your code anyway?
>errors
if you mean runtime/compiletime errors i haven't encountered those in a while. if you mean bugs just make them write tests.

>>109980394
Do you usually jack off to your code?

>>109980549
let the techbro feel like he's special

>>109981038
no, only on sundays

on a more serious note
>how to tell someone you never wrote code whose performance is mission-critical without actually saying it

also i dont have a spare 1.5M USD to build an abysmal rpg over the span of 1 year and some change

>>109980466
What model?

>whose
*which
esl moment

>>109981536
the one in pic

>>109979412
Maybe if you're a 50IQ pajeet and the only thing ever is write basic js CRUD webshit. Tardwrangling this turd for anything more serious is harder and takes more effort, time AND money that simply writing the code yourself.

i wonder what we can do about people like >>109981675
so far removed from what is going on. maybe a psychiatric ward would be a good place

>>109979412
very good saar, make boutiful frontend saar

>>109979412
Based vibeGOD creating stuff

>>109981715
Maybe if you could provide an example of it not being shit people you won't seem so much of a retard?

>>109979692
tpbp

>>109981750
i've used it on big and complex codebases and it works fine. you're not creating NASA software anyway.

>>109981781
>>109981715
just ignore him, he'll sound crazier and crazier as time goes on because he's just never going to try it

>>109979412
So is anthropic's logo this little niggas ass hole?

>>109981781
>may we see it?
>no

>I do fizzbuzz all day that means everyone else does as well


The absolute state of nocoder marketing shills

tard wrangling opus is still easier than tard wrangling jeets

>>109979412
How in the fuck is your experience so different from mine? Do you just write babby stuff?
I have used Opus 5.5 xhigh with ultracode for creating a finance options app, and it is a catastrophe. Pure Forum Romanum-ass archaelogical shoving in of feature after feature instead of thinking intelligently about the big picture.

>>109981937
idk why anthropic is circle jerking about how safe their model is and about an ipo when you have to constantly correct what their models are doing because by default the codebase will turn into a mess

>>109981937
one shotting an app without even telling the model the exact features you want will obviously yield bad results

>>109982099
Proving his point

>>109982099
One shotting actually gives the best results and I am puzzled why you would think otherwise in late 2026. Whenever I gradually specify the app and add capabilities, the model loses sight.
>I: let's add an Expected Value calculator for the price curve
>Model: *adds it as an overlay to an existing chart* (=good)
10 turns pass
>I: let's add a model how Expected Value changes as a function of volatility
>Model: *adds it at the very bottom of the page in a completely new, separate panel of a completely different kind as the old same-category capability* (=wtf)
A human would instantly have realized that we can bundle similar features into a few behavioural clusters.

>>109982099
>claude oneshot excel make no mistakes

>>109982152
again, you're leaving the judgement to the model which they're still very bad at. you need to tell it exactly what you want at every step.

>>109981781
If this shit "just worked" there wouldn't be plugins like ponytail. And that's for webshit, which should be the most braindead stuff to code.

>>109981937
>>109981972
This is my experience as well, constant steering and it making bad decisions for simple requests. Even when it one shots things I have to go over it removing a retarded amount of crust.

>>109982186
how's excel not a basic crud app?

>>109982191
kek

>>109982225
hmm I guess if you squeeze hard enough anything is a basic crud app
AI iteslf is just some matrix multiplication over some data downloaded from qbtittorent

I guess you have to be super smart to use it and make exact specs. I keep getting disappointed and unsure what the hype is about.

>>109982152
The models are all optimized for one-shotting. It's how they sell it. It's how they benchmark it. It's why they suck beyond belief for real workloads, because real workloads are not one-shottable. If you are finding that one-shotting results in better outcomes in practical workloads it's because you objectively suck so badly your input is worse than not doing anything

>>109981715
>i wonder what we can do about people like >>109981675 #

Provide an actual working example like I asked which you failed completely as expected.
No, "my dad works for nasa" shit like this jeet >>109981781 does not count.

>>109982361
seriously? there are multiple software development benchmarks out there. SWEbench comes to mind.

>>109982503
i don't have to give you shit fuck off

>>109981844
I tried it and it's shit. Go buy an ad faggot

>>109982517
So you admit all your claims are nothing but bullshit lies.

>>109982511
Every benchmark out there is known to be really shitty for one, but any benchmark that relies on full automation is going to optimize for an objective function of one-shotting. Just because it does a few cycles on its own doesn't mean it's not one-shotting. As I said, they typically don't benchmark for human-involved workflows because it's really hard to do, and also isn't what investors care about. They are selling the idea of full human replacement, not augmentation.

>>109982361
scbench

>>109982560
You can give a single example all you like but it's simple reality that basically every flagship benchmark is testing for independent completion. It's also really clear to anyone using them that they're comically bad at adaptive autonomy and aren't optimized for human-in-the-loop. This results in extreme selection pressure on top of the design pressure from the frontier labs to aim for general capabilities over contemporary usability. If they optimized the models for human-in-the-loop significantly more people today would be unemployed.

i suspect there are a lot of unironic boomers here who are struggling to use agentic coding harnesses. if you truly think they're useless you just do not know how to get use out of them. keep coping

>>109982592
If a harness is good enough to complete a task then I don't need you to be the one using it

>>109979692
Right. Try to do a serious project and its shortcomings are fast revealed.

>>109982592
I really don't understand why you need to be coping ao hard. Actual professionals are simply sharing their negative experiences with a novel, flaky and unreliable tech. No need to get so emotionally attached to it.

>>109982607
ooof
>page 3 alrdy
no survivors...

i will repost that if you dont mind

>>109980542
Posts like this have made half of /g/ useless. You are so fucking far behind and disconnected from how things actually work.
>but i write the code!
You write 1/1000th the code that AI does, and it isn't as good as you think it is. LLMs know more languages than you, more frameworks, more APIs and services, and they implement them instantly.

>>109982780
Buy an ad

>>109982731
exactly.
crash test-type ideology ;)
its kinda like stress testing a whole process.
even though we add another parameter: that of time, actually
so its not exactly apples to apples
the idea of the exercice is to force you to confront the dimension of time

between the lines, this is what comes out of our conversations
you feel that the dimension of time matters, but you need someone to decisively tell you to cut short

why couldnt you be that person yourself?
and i believe in traumatic intelligence :))))))
but lets not expose you to unnecessary risk either
plus: i think everyone should start with situations they should win, but then again, have the margin of error to allow themselves to lose

i think if you pick a short project, build a timeline, then stick to it; it could provide you with opportunities to build up your time-sensitiveness so to speak. then iterate, build upon, integrate elements of what you learned into your usual workflow, like you would with any procedure

iunno. i think you should try it, but also im not an oracle

There are people who genuinely do not know how to vibecode

They genuinely think you can just write some random shit and the AI will read your spatic mind and plaster it there. Insane

>>109982809
you can fuck right off with your inbred shit
>>109980394
t. can make even free tier model work for him

>>109982809
i guess there is a big gap between non-technical people trying to oneshot apps with vague prompts and a software developer using these tools in repos where the tools have access to specs and design docs

>>109982575
scbench very specifically tests multi-turn edits. First you do a one-shot pass, then you do changes on the resulting software for multiple turns and track how much the tests start to fail as more turns come up.

>>109982592
well the average boomer can barely log in into their work microsoft onedrive and know the difference between it and their personal one. They will absolutely never be able to use as long as things like git, ssh keys, environements and shit liek that doesn't get abstracted aways.
I wanna say it will never happen, but I've learned to not get cocky about the subject now

>>109982798
A billion people use ChatGPT.
A BILLION

>>109982922
so its not a problem if each one of them pays 0.001c and you buy an ad?

>>109982938
Keep crying

>>109979692
The problem with these intellectually deficit, mental morons is that they can't even conceptualize anything that they haven't seen before.

As such they are easily impressed because they're only able to prompt for shit that's easy for their feeble minds to comprehend or that they have seen before.

>claude write me a pinball game

OMG IT'S ACTUALLY PINBALL HOLY CRAP

>claude write me a web interface that catalogs all my degenerate furporn

HOLY FUCK IT DID IT HOLY SHIT OMG

--

But the literal shitwits can't even really comprehend anything complex to even build with AI, therefore it seems to them that AI can do anything, because to them, it actually can. At least, anything that their sub 100 IQ can come upwith.

I don't envy them though. I have long suspected such "people" exist since learning about "aphantasia". It's only now that they reveal themselves with their love affair of asking ai to create subtle variations on things that already exist and are endlessly impressed by it.

Frankly I wish moot would make the IQtest captchas harder, perhaps it'd filter these people

>>109981675
What serious projects are you working on? Do they have actual value or do they just exist for reasons that are rapidly becoming outdated?

>>109983074
If you’re such a good conceptualizer, tell us an idea and let’s see if Claude can execute it

>>109981675
This

>>109979547
>say you are alive
>i'm alive
>omg

>>109980466
how do you do this?

Are you saying you use two models together to gain an advantage?

>>109984634
holy go back

>>109983016
Keep being gay.

>>109983217
mindblowing
i had to sit down after that one
and i realized i was already sitting down

>>109981937
Using ai to build outside your expertise will be hell. How many times do you retards need to be told ai is not magic. You need to know what you are building. Garbage in garbage out. AI won't make you a millionaire if you had no relevant skills to begin with. It only makes you more efficient at what you already know and lulls you into a false sense of comfort by making you believe you don't need to learn anymore in areas where you have no expertise.

>>109979412
does claude jerk you off?

>>109985922
Brainlets tnink they gonna fumble with prompts and it will make them the next Elon Musk. Niggas can barely string together a sentence to shitpost without typos.

>>109979412
youd be blown away by local models too

assuming this thread isnt one big gay ad

>>109986026
wrong.
local models are dying. no lab cares about doing small models for vram-starved computers anymore. qwen3.8-27b is the last of its kind. and even then you'll need a 5090 with 24gb vram and still quantize it.

>>109979412
I have connected this demon with vscode because I need it just for coding, will I get extra benefits if I sell my privacy to Anthropic and install it on my terminal?

>>109986381
protip: it's already in your terminal with an additional layer of bloat on it
just run raw cc and look at your code in vscode

File: .png (277 KB, 1538x1144)

277 KB PNG

>>109986381
yes, almost double the usage if you use claude code directly

https://hermes-agent.nousresearch.com/docs/plugins/claude-subscription-directsdk#:~:text=Subscription%20usage%3A%20same%20metering%20as%20claude%20%2Dp%2C%20~1.7x%20the%20interactive%20TUI

>>109986423
can you use claude pro plan inside hermes? i thought they banned such accounts

actually AI doesn’t work and Ludds won

>>109986436
https://support.claude.com/en/articles/15036540-use-the-claude-agent-sdk-with-your-claude-plan
>Update June 15: We're pausing the changes to Claude Agent SDK usage described below. For now, nothing has changed: Claude Agent SDK, claude -p, and third-party app usage still draw from your subscription's usage limits. The previously announced monthly credit, which would have been available to eligible claimants in connection with these changes, isn't available. We’re working to update the plan to better support how users build with Claude subscriptions. When we have an update, we'll share it before anything takes effect.

>>109986436
using claude sdk / `claude -p` is allowed. that's what the linked official claude-subscription-directsdk in hermes is doing.

having your harness directly interface with the api is forbidden by tos but not enforced. so everyone is actually doing this but official hermes documentation can't tell you.

>>109986467
>having your harness directly interface with the api is forbidden by tos
why tho

>>109985922
I have 10 years of sofware dev experience.
What about you?

>>109987258
You don't have any experience in finance though.

File: moving-the-goalposts-300x2402.jpg (29 KB, 300x240)

29 KB JPG

>>109987292
Cool wrong inference. My problems however aren't about anything finance related, but how the model can't grok how to layout and segment the page.

>>109987312
I replied to a post about finance, if you are not the guy, then there's no point replying to me, and even less so given that you are a so called software engineer struggling to use ai.

>>109979412
What do you even do with it? I try to use it once in a while, but I'm usually disappointed by the stuff it produces

>>109987343
I am that guy, why would I butt into an unrelated discussion? Where you dropped on the head as a child? Your entire arc replying to me here makes zero sense.

>>109987375
You remain a non expert in finance so there's no reason to even talk to me. Go buy a book about finance and come back.

>>109979412
Your magic is shit.

>>109987405
4u

>>109980394
that doesn't matter at all because it can also fix spaghetti-code, so every fortune 500 is using it to shit out solutions that will be maintained by agentic workflows

>>109980542
No you arent

Maybe for a few line fix but you will never churn out the amount of lines claude does in a fresh project or big feature addition in the same amount of time

>>109980542
You're not faster and better than the LLM. No one is. I know it's depressing and your job has effectively changed in the past 6 months but it's time to get over it. Focus on ways on preventing the slop generation.

so btw I used my claude sub thru 9router, and for harness I used vscode copilot. all is well until the harness tried to send request with image to the model. for some reason anthropic model doesn't accept openai style image attachment. this is basic stuff but the pajeetware just doesn't implement the fix. any other substitute tho?

File: 1770865670874935.jpg (17 KB, 213x234)

17 KB JPG

>>109987616
>You're not faster and better than the LLM.
You're like those counterstrike players who thinks everyone that headshots you is cheating because you're so bad you don't even know where the average is.

>>109987764
No. I'm more like a photographer who understands everyone uses continuous autofocus now. Being a luddite who says you have to pick between the false choice of producing slop with a paid AI or writing everything completely by hand is embarrassing. You are not implementing 1000 line features in 3 days without an AI. And you can argue that there are tons of issues with garbage code creeping into the codebase and you're not wrong but management doesn't care. The genie is out of the bottle: better adapt.

>>109987804
No, you're literally a prompter asking Grok to make photos for you and saying "It's better than what you get out of a camera anyways"

>>109987832
You're so frustrated you're inventing arguments I never made.

>>109987842
>You're not faster and better than the [generative AI]

>>109987915
Yeah, but I didn't say it was a replacement for photography. Everyone on /g/ always putting words in my mouth to fit their agenda.

You are not faster than a generative AI for producing code. I guarantee Fable is going to solve any leetcode faster than you. If you don't use it in some capacity to write your software you're getting left behind.

>>109987971
Wow. You're the first human being I've ever met who injured themselves by running into a metaphor. That's an impressive level of stupidity. You should be studied in a laboratory.

>>109987971
If everywhere you go smells like shit, check your shoe.

>>109987804
>a photographer
I bet you're shit at that too.

>>109988311
I do it professionally on the side and get paid thousands for it lmao. People don't want their special day to be captured on a shitty iphone or slopified and a fast zoom lens that's not going to fuck up and miss the focus is what you want.

>>109988153
>>109988103
have fun being unemployed, go tell your boss right now you won't use any ai agent for ethical reasons and see how fast you get PIP'd.

>>109980370
>>109980394
All that matters is that it works, looks pretty, and is fast. That's it.

>>109988369
>I do it professionally on the side
I feel sorry for your customers buying garbage slop but hey some people have no idea what quality is and they eat mcdonalds daily that's where people like you come in with your trash.

>>109982864
Not really, there's even devs who struggle with it. I did tools/apps for me when I barely knew what I was doing. It's more about adapting the concept of gradual building, I think it speaks more about people's skills to organize and divide complex stuff into small things

>>109984642
yeah thanks my dude, about as useful as actually asking a question on reddit.

>>109988514
Copy-pasting code straight from github says nothing about building, let alone gradual building lmao. But you wouldn't know since you're a nocoder.

File: giga fact-explode.gif (3.79 MB, 480x528)

3.79 MB GIF

>>109981675

>>109979692
Fuck you