>one small convos with opus>is Claude getting jewwer
>>108451893Stop asked if it to textually suck your fag cock and pound your bussy
Programming is for richfags now. Deal with it.
>>108451893Is this for your transgender fetish fan fiction?
>>108451893no, because it was already maximally jewish from the startuse local models
>>108451934never will be this is just retard tax
>>108451963b-but local model is not.. le smart than daddy dario’s models
>>108451934Weird i can still do it for free. Must be a you problem.
>>108451984>>108452088Do you still believe you are more valuable to your company than this guy?
opus is jewish. need to use sonnet
>>108451893pay up cheapfag
>>108452185>jew1 is jewish>need to use jew2
>>108451893>the service anthropic provides at a loss is not enough>no, I don't want to pay real API prices, that's too expensive>no, local models are bothersome and dumbI hate these threads. The entitlement is unbearable.
>>108451893>using Claude Code>on Pro plan>ask 1 (one) new question on a new conversation>it cuts off half the reply because LIMIT REACHEDWell if this isn't a bug and the actual new limits there's no point in it. >j-just use Sonnet!I'm not working on a fucking grocery app in a JavaScript framework, Opus is barely able to not fuck everything up Sonnet is just off limits.
I thought you mean Opus the audio codec and I was thinking "no way it uses that much bandwidth, and compared to what".
>>108451934>slavewaging is for richfagsdo proles really?
>>108452436Oh, you are underage...
>>108452117what do you so when your agent is doing stuff? i don't really want to context switch but waiting for the chatbot to complete is also not something i want to do.
>>108451893It's always like this during peak hours. They want you to use it at less busy times>>108452117what a ghetto office
>>108452117whoever made this video doesn't even know what POV means when it takes five seconds to figure this out
>>108452579women's brains are physiologically unable to understand pov
>>108452117Yeah, I've worked with a guy leading a project and basically doing this within the last 6 months. Once they were about 4 months into it, fixing bugs and adding features demanded by the founders started to become a nightmare since it was something like 100k lines of badly structured mess with many pointless entry points, duplicate docs spread slapdash all over the repos, extensive tests that did basically nothing of real value to validate changes, literally hundreds of database migrations to address shortcomings of table structures that should have been obvious to anyone who spent a couple of hours before they started thinking about how basic things were going to be handled. Promises about test environments that should perfectly match production behavior that didn't actually work once we tried, The whole thing was scrapped as unmaintainable.
I have been using these tools in my dev work and I've found two bottlenecks>multiple agentsUnless the tasks are mutually exclusive, there is always a slight chance of one agent fucking over the other. Even then, you can only review one agent at a time so the advantages of concurrent agentic development are diminished. >context switchingThe idea of doing other shit while an agent runs sounds appealing, but you can't truly context switch efficiently if you are unsure of when the agent is gonna finish its task, and when it does, it forces a sharper context switch than the one you did when moving away from the agent.So yeah, the main weakness of the AI-enhanced dev workflow is our single-threaded mind. But despite those shortcomings, I have felt the boost in my productivity for tasks I know how to do but it would have taken a lot longer by hand. Hobbyists don't have use case for it that justifies a subscription, though, so stay clear unless you can produce value equivalent or higher than the subscription(s).
>>108452579are you really questioning the intelligence of tiktok video makers?
>>108452661Any opinions on maintainability?
>>108451893same fucking bullshit herethey also lobotomized it
>>108451893Why bother
>Create a whole generation of vibe code monkeys that have no idea how to code yet can generate massive projects automatically> Turn the screw once the VC money starts drying up essentially making them all useless
>>108453719Maintainability is inversely proportional to how much you know about the codebase and the business rules encoded in the app. If you tell the AI to do stuff in a codebase you are not familiar with, how are you expected to review/fix/debug AI-generated code on it? I think responsible use is important here to avoid falling into the rabbit hole of ending up with an unmaintainable app like >>108452635 describes. The illusion of productivity is strong when using these tools. You ought to have a critical and skeptical mindset at all times when using them.A practice we adopted in our team is creating rule specification files and adding them to the codebase (e.g. Cursor support those). The agent grabs the rules on chat initialization and boots up with the full context of your team's coding standards, practices, and patterns. We've noticed this helps the AI generate more predictable, readable code that is easier to review and maintain. When something changes in our codebase that we judge should be part of the rules, we tell the AI to generate the new rules or edit the existing ones. We have also created role files that execute on commit, fire off multiple agents in parallel with different roles (security eng, senior dev, software architect, ...) and generate a report on potential code smells and suggestions, or nothing at all (yes, we explicitly tell the agents to shut the fuck up if they have nothing relevant to suggest). Humans still have the final say in these code reviews, but these code review agents have caught smells we would have missed otherwise.So yeah. These tools work, but you have to use them responsibly, integrate them into your workflow, and understand their limitations. People who get this will survive future layoffs when the C-Suite guys realize mindless monkeys typing prompts are a liability for the bottom line.
>>108453941>Maintainability is inversely proportional to how much you know about the codebase and the business rules encoded in the app.Brainfart. I meant "directly proportional". The more you know, the easier it is to maintain.
>>108452635You need to read the code and to maintain good documentation for your agent or this happens.Also me in the video.
>>108451893So what's happening? One of>yet another bug that the vibecoded slop produced>they're testing to see how much they can turn the pressure up before people fuck off>there's no plan, the free money has dried up, $100/million tokens is the new normal and here to stay (this pops the AI bubble)
>>108454088Pentagon put a gun to their heads and demanded 80% of the computing power. I mean, surely it's not a coincidence that this is happening just as the US is planning a ground invasion?
>>108454175I pray for the poor fucks that will be on the ground. I bet the retards don't even know that despite what context limit may say, LLMs severely degrade after ~20k tokens in context or so. If 5000 marines die landing on a shitty tiny island within FPV drone range of the Iranian mainland it will be the cherry on top of the whole AI slopcake of the last few years.
>>108452117Yes. And being able to program without even internet will put me ahead of most devs.
>>108452117so this is why all software has been getting shittier and shittier huh
>>108452117literally me (i get paid a lot too)
>>108452345fucking real, opus barely became enough for what i am working
>>108453941>>108453950But bugs are often one-liners, subtle logic errors, race conditions, ... the type of stuff where you really need to understand every single line of code, not just the high level business logicSure if you have a detailed set of business rule requirements the AI can do a pretty good job drafting out a spec but when it comes time to debug a subtle issue, aren't you afraid the thousands of lines of code you haven't read yourself will be impossible to debug or reason about? Or is that not something you worry about.We are okay letting code quality slip if it allows for faster shipping
>>108454201you write of things like operation varsity?
>>108453741ChatGPT can't figure this one out either. Gemini had no problem.
>create service 1>create service 2 that's "better" but also more expensive>people feel enticed to try service 1 even when they originally had no interest in said servicewhy do people fall for these tricks
>>108453741you got a bootleg version of opus mayhaps
