Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

I cancelled Claude: Token issues, declining quality, and poor support

nickyreinert.de

Post reply on HN
Print view View original

441–450 of 604 posts

Re: I cancelled Claude: Token issues, declining quality, and poor support

#441

Post by bryan0 »

I see a lot of people struggling to work with agents. This post has a good example:

> “you can’t be serious — is this how you fix things? just WORKAROUNDS????”

If this is how you’re interacting with your agents I think you’re in for a world of disappointment. An important part of working with agents is providing specific feedback. And beyond that making sure this feedback actually available to them in their context when relevant.

I will ask them why they made a decision and review alternatives with them. These learnings will aid both you and the agent in the future.

Re: I cancelled Claude: Token issues, declining quality, and poor support

#442

Post by xlii »

syhol wrote: post #33
zkmon wrote: post #21

Yesterday was a realization point for me. I gave a simple extraction task to Claude code with a local LLM and it "whirred" and "purred" for 10 minutes. Then I submitted the same data and prompt directly to model via llama_cpp chat UI and the model single-shotted it in under a minute. So obviously something wrong with coding agent or the way it is talking to LLM. Now I'm looking for an extremely simple open-source cod…

https://pi.dev/ seems popular, whats not open source about opencode? The repo has an MIT License.

+1 for pi. I used claude and opencode but pi is the first agent tool that made me excited about the whole thing.

Re: I cancelled Claude: Token issues, declining quality, and poor support

#443

Post by cyanydeez »

hadlock wrote: post #437
cyanydeez wrote: post #65

Earlier quoted context omitted.

Honestly, it sounds like, assuming you have no ethical qualms, you could get by with a Mac or AMD 395+ and the newest models, specifically QWEN3.5-Coder-Next. It does exactly as you describe. It maxes out around 85k context, which if you do a good job providing guard rails, etc, is the length of a small-medium project. It does seem like the sweet spot between WallE and the destroyed earth in WallE.

Seems like AMD 395+ is only about 16 tokens/s which is 25-33% the speed of SOTA models. Break even on a $3000 machine is ~15 months

thats pessimistic. do the calc assuming Cloud provider X changes your nondetermistic output every Y Months by Z probability and increases prices by 10% every 6 months.

slow and steady is worth exponentials. keep slopppping it my boid.

Re: I cancelled Claude: Token issues, declining quality, and poor support

#444

Post by Exoristos »

throwuxiytayq wrote: post #439
Exoristos wrote: post #273

Earlier quoted context omitted.

The number of devs I've worked with who can't touch-type and don't use or know their way around a proper IDE is depressingly large.

This isn't about touch typing or IDE tricks. I'm an IDE power user and - reasoning aside - I used to run circles around my peers when it comes to raw code editing efficiency. This is increasingly an obsolete workflow. LLMs can execute codebase-wide refactors in seconds. You can use them as a (foot-)shotgun, or as a surgical tool.

So many are masters of AI marketing, it's thinkable one of them has mastered AI.

Re: I cancelled Claude: Token issues, declining quality, and poor support

#445

Post by linsomniac »

wg0 wrote: post #202

I write detailed specs. Multifile with example code. In markdown. Then hand over to Claude Sonnet. With hard requirements listed, I found out that the generated code missed requirements, had duplicate code or even unnecessary code wrangling data (mapping objects into new objects of narrower types when won't be needed) along with tests that fake and work around to pass. So turns out that I'm not writing code but I'm r…

>Then hand over to Claude Sonnet.

Have you tried Opus 4.6 with "/effort max" in Claude Code? That's pretty much all I use these days, and it is, honestly, doing a fantastic job. The code it's writing looks quite good to me. Doesn't seem to matter if it's greenfield or existing code.

If code is harder to read than to write, you're doing yourself a disservice by having the output stage not be top shelf.

Re: I cancelled Claude: Token issues, declining quality, and poor support

#446

Post by satvikpendem »

sidrag22 wrote: post #274
satvikpendem wrote: post #240

Earlier quoted context omitted.

How is that an absurd bar? If you're handwriting code, you'd need to know what you actually want to write in the first place, hence you understand all the code you write. Therefore the code the AI produces should also be understood by you. Anything else than that is indeed vibe coding.

its an absurd bar if you are being a uncharitable jerk like i was, the layers go deep, and technically i can claim I have never fully grasped any of my code. It is likely just a dumb point to bring up tbh.

I saw your reply to another comment [0], I see what you mean now. By "understand each line of code" I meant that one would know how that for loop works not the underlying levels of the implementation of the language. I replied initially because lots of vibe coding devs in fact do not read all the code before submitting, much less actually review it line by line and understand each line.

[0] https://news.ycombinator.com/item?id=47894279

Re: I cancelled Claude: Token issues, declining quality, and poor support

#447

Post by hunterpayne »

Aurornis wrote: post #316
Exoristos wrote: post #273

Earlier quoted context omitted.

The number of devs I've worked with who can't touch-type and don't use or know their way around a proper IDE is depressingly large.

Same with debuggers. I run into people with 10 years of experience who are still trying to printf debug complex problems that would be easy with 5 minutes in a debugger. I think we're seeing something similar with AI: There are devs who spend a couple days trying to get AI to magically write all of their code for them and then swear it off forever, thinking they're the only people who see the reality of AI and everyo…

Right, but then the AI doesn't have a positive ROI. In all fairness, it never has a positive ROI but now its much more negative, to the point the accountants will put an end to the experiment after year end reveals how negative it really is.

Re: I cancelled Claude: Token issues, declining quality, and poor support

#448

Post by coldtea »

wg0 wrote: post #202

I write detailed specs. Multifile with example code. In markdown. Then hand over to Claude Sonnet. With hard requirements listed, I found out that the generated code missed requirements, had duplicate code or even unnecessary code wrangling data (mapping objects into new objects of narrower types when won't be needed) along with tests that fake and work around to pass. So turns out that I'm not writing code but I'm r…

>I write detailed specs. Multifile with example code. In markdown. Then hand over to Claude Sonnet. With hard requirements listed, I found out that the generated code missed requirements, had duplicate code or even unnecessary code wrangling data (mapping objects into new objects of narrower types when won't be needed) along with tests that fake and work around to pass.

Stop doing that. Micromanage it instead. Don't give it the specs for the system, design the system yourself (can use it for help doing that), inform it of the general design, but then give it tasks, ONE BY ONE, to do for fleshing it out. Approve each one, ask for corrections if needed, go to the next.

Still faster than writing each of those parts yourself (a few minutes instead of multiple hours), but much more accurate.

Re: I cancelled Claude: Token issues, declining quality, and poor support

#449

Post by dieortin »

coldtea wrote: post #448
wg0 wrote: post #202

I write detailed specs. Multifile with example code. In markdown. Then hand over to Claude Sonnet. With hard requirements listed, I found out that the generated code missed requirements, had duplicate code or even unnecessary code wrangling data (mapping objects into new objects of narrower types when won't be needed) along with tests that fake and work around to pass. So turns out that I'm not writing code but I'm r…

> I write detailed specs. Multifile with example code. In markdown. Then hand over to Claude Sonnet. With hard requirements listed, I found out that the generated code missed requirements, had duplicate code or even unnecessary code wrangling data (mapping objects into new objects of narrower types when won't be needed) along with tests that fake and work around to pass. Stop doing that. Micromanage it instead. Don't…

Might as well just write the code yourself at that point. And as a bonus, end up with a much better understanding of the codebase (and way better code)

Re: I cancelled Claude: Token issues, declining quality, and poor support

#450

Post by coldtea »

dieortin wrote: post #449
coldtea wrote: post #448

Earlier quoted context omitted.

> I write detailed specs. Multifile with example code. In markdown. Then hand over to Claude Sonnet. With hard requirements listed, I found out that the generated code missed requirements, had duplicate code or even unnecessary code wrangling data (mapping objects into new objects of narrower types when won't be needed) along with tests that fake and work around to pass. Stop doing that. Micromanage it instead. Don't…

Might as well just write the code yourself at that point. And as a bonus, end up with a much better understanding of the codebase (and way better code)

>Might as well just write the code yourself at that point

"We have this thing that can speed your code writing 10x"

"If it isn't 1000x and it doesn't give me a turnkey end to end product might as well write the whole thing myself"

People have forgotten balance. Which is funny, because the inability of the AI to just do the whole thing end to end correctly is what stands between 10 developers having a job versus 1 developer having a job telling 10 or 20 agents what to do end to end and collecting the full results in a few hours.

And if you do it the way I describe you get to both use AI, AND have "a much better understanding of the codebase (and way better code)".

Post reply on HN