Comment by pydry
16 hours ago
I'm amused at how many hacker news accounts saw this and immediately jumped to the conclusion that you either were using the model wrong or using the wrong model.
ive been regularly seeing this exact reaction online since November 2025, sadly :/
It's just a little baffling to see someone describe a level of performance I haven't experienced since 2025, despite frequently using the tech, as being a frequent concern.
Might be the type of projects you are working on and how much you care about performance and code quality.
Working on more complex, logic heavy projects with strong performance needs I find the models to be useful but certainly not 'one shot' on pretty much anything. And often incredibly frustrating and genuinely bad code that collapses performance and bloats systems - like what a really bad junior might write.
When I'm working on large standard crud web projects with already decent architecture and a good harness and skills, honestly they work pretty well a lot of the times, the code looks good, does what it needs and fits in with the architectural style.
I really think a lot of hn people just write simple repetitive software, and a small portion works on complex, weird, dense, and novel'ish logic projects. The two obviously don't have the same experiences.
To a first approximation, since 2025 was at best a year ago, it's not reasonable to be so surprised that the improvements in that time weren't enough for some use case. There's only so much anything can improve in a year.
> I'm amused at how many hacker news accounts saw this and immediately jumped to the conclusion that you either were using the model wrong or using the wrong model.
Yes, so I'll disclose it: I was using Opus 5.5, on medium effort, in Claude desktop, which has full access to my entire repo.
Now everyone can officially lambast me for "using the model wrong or using the wrong model," exactly as you say. But I find it quite interesting that one of the commenters here assumed I was using Opus 4.6, because these mistakes sound like that old version! I'm using the version released just four freaking days ago!
I expect some commenters will now say, "Oh, you should have been using Fable, you old boomer." To them I say: "Yeah, well my employer doesn't allow me to use Fable." And, in jest: "Now get off my lawn."
Exactly this. If you were using the exact model today that 6-12 months ago people here were telling you "absolutely does not make this kind of mistake anymore" they'd tell you the same thing again, just replacing $OLDER_MODEL with $NEWER_MODEL. Why did $OLDER_MODEL not make this mistake 6-12 months ago, but it does now? The answer to that question is obvious, but AI-bros cannot understand that.
It's basically impossible at this point to take these people seriously anymore.
If you're tired of shenanigans pulled by these AI companies then you should host your models locally.
My model weights don't change unless I change them.
I still do upfront planning and then do careful review of all AI content. The behavior the parent described above hasn't happened for me since around Opus 4.6.
The most telling one is counting function invocations wrong, because that's simply not how models work anymore. They use terminal commands and Python scripts for research like that (if not an LSP, if one took the time to set up their tools most effectively).
Combined with their attitude, I have little doubt that the parent has disabled tool calls, is working in some janky Harness like chat/Duo/Juno, or is using a severely reduced or outdated model.