…and I still don’t get it. I paid for a month of Pro to try it out, and it is consistently and confidently producing subtly broken junk. I had tried doing this before in the past, but gave up because it didn’t work well. I thought that maybe this time it would be far along enough to be useful.
The task was relatively simple, and it involved doing some 3d math. The solutions it generated were almost write every time, but critically broken in subtle ways, and any attempt to fix the problems would either introduce new bugs, or regress with old bugs.
I spent nearly the whole day yesterday going back and forth with it, and felt like I was in a mental fog. It wasn’t until I had a full night’s sleep and reviewed the chat log this morning until I realized how much I was going in circles. I tried prompting a bit more today, but stopped when it kept doing the same crap.
The worst part of this is that, through out all of this, Claude was confidently responding. When I said there was a bug, it would “fix” the bug, and provide a confident explanation of what was wrong… Except it was clearly bullshit because it didn’t work.
I still want to keep an open mind. Is anyone having success with these tools? Is there a special way to prompt it? Would I get better results during certain hours of the day?
For reference, I used Opus 4.6 Extended.
producing subtly broken junk
The difference between you and people that say it’s amazing is that you are capable of discerning this reality.
What I don’t get, though, is how the vibe code bros can’t discern this reality.
How can they sit there and not see that their vibe-coded app just doesn’t do what they wanted it to do? Eventually, you’ve got to try actually running the app, right? And how do you keep drinking the AI kool-aid when you find out that the app doesn’t work?
They’re the same people that copied code from stack overflow that you had to tell them how to actually fix every PR. The difference is the C suite types are backing them this time
Vibe code bros aren’t real programmers. They’re business people, not computer people. Even if they have a CS degree, they only got that because they think it’ll get them more money. They lack passion and they don’t care about understanding anything. They probably don’t even care about what they’re generating beyond its potential to be used in a grift.
I graduated college not that long ago and my CS classes had quite a few former business majors. They switched because they think it’ll be more lucrative for them but since they only care about money they didn’t bother to actually learn the material especially since they could just vibe code through everything.
So much this.
After working in tech companies for the last 10 years I’ve noticed the difference between people that “generate code” and those that engineer code.
My worry about the industry is that vibe coding gives the code generators the ability to generate even more code. The engineers (even those that use vibe tools) are not engineering as much code by volume compared to “the generators”.
My hope is that this is one of those “short term gain, long term pain” things that might self correct in a couple of years 🤞.
It’s insane that companies are going back to metrics like LOC (or tokens generated), when the industry figured out decades ago that these are horrible, counterproductive metrics.
“The hard thing about building software is deciding what one wants to say, not saying it. No facilitation of expression can give more than marginal gains.” - No Silver Bullet (1986)
I do apps that work, i do patches that are production quality. Half the cs world does… I do full stack ai debugging of esp32 projects.
It’s a powerful tool, you just need to learn it’s strong and weak points, just like any other tool you use.
Half the cs world does…
What’s the basis for this claim? I’m doubtful, but don’t have wide data for this.
Rough estimate from my personal connections only. Some work places where ai is not possible, but all that have made an effort report good code. You need to work with what it is - a word generator that sometimes gives correct results. Make it research and not trust training. Never let it do things on its own, require a plan and reason. Make it evaluate its own work/plan.
Most issues i have stem from models beeing too eager. Restrain them and remove the “i can do this next…”behaviour.
Context is king - so proper mcp and documentation that is agent facing. I use serena as i can get lsp for yaml, markup and keep these docs like that
No, I think you do get it. That’s exactly right. Everything you described is absolutely valid.
Maybe the only piece you’re missing is that “almost right, but critically broken in subtle ways” turns out to actually be more than good enough for many people and many purposes. You’re describing the “success” state.
/s but also not /s because this is the unfortunate reality we live in now. We’re all going to eat slop and sooner or later we’re going to be forced to like it.
Or maybe we will be forced to switch off LLMs and start solving the bugs introduced by their usage using our minds.
As a professional software developer, I truly hope that is the case (and I plan to charge at least 10x my current rate after the AI bubble pops when I’m looking for my next job as I expect there to be a massive shortage of people skilled enough to actually deal with the nightmare spaghetti AI code bases)
Fun times ahead.
Maybe the only piece you’re missing is that “almost right, but critically broken in subtle ways”
Sure, but you have to note that it reaches that point in minutes. Sometimes on a task that would take humans a week. The power is not that it creates correct stuff, it’s that it creates almost correct stuff 100 times faster than human. Plus the typical machine benefits: it never gets tired, demotivated, etc.
So then the challenge becomes being able to be that human, who can review stuff extremely well and rapidly, being natural in probing the stuff LLMs tend to be wrong about. Sort of like the same challenge that every tech lead had before LLMs too, but just subtly different, because LLMs don’t exactly think like we do.
Their usual (crap) defense is:
a) you’re not paying enough, so of course it is crap
b) you’re not prompting right, you need to use detailed, precise language…
c) that is just anecdotal evidence, you need to do an actual study, yadda yadda.
d) it will improve…
(any other anyone has noticed?)
English is cheap to replicate, there is no science to prompting it’s asking the goddamn question.
If AI companies are so keen in keeping humans like dumbasses, that’s an issue on their part not my fucking English.
Vibe coding, in the sense of telling the model to make codebase changes, then directly using the output produced, is 100% marketing bullshit that does not scale beyond toy examples.
Here’s the rub: Claude is extremely useful as an advanced autocomplete, if and only if you’re guiding it architecturally through every task it runs, and you vet + revise the output yourself between iterations. You cannot effectively pilot entirely from chat in a mature codebase, and you must compile robust documentation and instructions for Claude to know how to work with your codebase.
You also must aggressively manage information in the context window yourself and keep it clean. You mentioned going in circles trying to get the robot to correct itself: huge mistake. Rewind to before the error, and give it better instructions to steer it away from the pitfall it fell into. Same vein, you also need to reset ASAP after pushing into the >100k token mark, because the models start melting into putty soon after (yes, even the “extended” 1M-window ones).
I’m someone who has massively benefited from using modern LLMs in my work, but I’m also a massive hater at the same time: They’re just a tool, not magic, and have to be used with great care and attention to get reasonable results. You absolutely cannot delegate your thinking to them, because it will bite you, hard and fast.
For your use case (3D math), what I recommend is decomposing your end goal into a series of pure functions that you’ll string together. Once you have that list, that’s where Claude comes in. Have it stub those functions for you, then have it implement them one at a time, reviewing the output of every one before proceeding.
I rarely use LLM’s for generating code. Usually, by the time I’ve provided all the necessary context, I might as well have just written the code myself. I do use LLM’s for doing research. As long as it’s understood that the response is only as accurate as the source material, they often do a decent job of distilling down to what I’m actually looking for.
Have you been coding professionally long?
I find that the only time I can use these chatbots for a task I really need to already know what i’m doing so that I can read the output and fix the issues. This is like having junior devs on your team and being a code reviewer more than being a full time coder. They get a lot of things wrong but there’s so much usable that you can save a ton of time over doing everything yourself from scratch.
Just like with junior devs, you can send them back to fix what you know is wrong and give them feedback to improve various things you would prefer done another way. There’s no emotions though, so you can just be blunt and concise with feedback.
Nice comparison, but the bugs created by junior software developers are usually much easier to find than the bugs created by LLMs.
-
Did you have MCP tooling setup so it can get lsp feedback? This helps a lot with code quality as it’ll see warnings/hints/suggestions from the lsp
-
Unit tests. Unit tests. Unit tests. Unit tests.
I cannot stress enough how much less stupid LLMs get when they jave proper solid Unit tests to run themselves and compare expected vs actual outcomes.
Instead of reasoning out “it should do this” they can just run the damn test and find out.
They’ll iterate on it til it actually works and then you can look at it and confirm if its good or not.
I use Sonnet 4.5 / 4.6 extensively and, yes, its prone to getting the answer almost right but a wrong in the end.
But the unit tests catch this, and it corrects.
Example: I am working on my own fame engine with monogame and its about 95% vibe coded.
This transform math is almost 100% vibe coded: https://github.com/SteffenBlake/Atomic.Net/blob/main/MonoGame/Atomic.Net.MonoGame/Transform/TransformRegistry.cs
The reason its solid is because of this: https://github.com/SteffenBlake/Atomic.Net/blob/main/MonoGame/Atomic.Net.MonoGame.Tests/Transform/Integrations/TransformRegistryIntegrationTests.cs
Also vibe coded and then sanity checked by me by hand to confirm the math checks out for the tests.
And yes, it caught multiple bugs, but the agent automatically could respond to that, fix the bug, rerun the tests, and iterate til everything was solid.
Test Driven Development is huge for making agents self police their own code.
-
I only use AI for generating ok looking UI.
Anthropic says Methos will find bugs on FreeBSD, Bank system etc. What a bullshit.
Oh, it will ‘find bugs’ alright. And then flood FreeBSD’s bug report system with bullshit bug reports that turn out to be nothing, but require expert human review to discern that.
It’s easier finding bugs than producing correct, readbale, maintainable code though-
“Almost but not quite” is exactly my experience with Claude.
The only time I’ve had real success is telling it to do a simple API change that touches a dozen files. It took a while and I’m not sure it was faster than doing it manually, but at least it was less boring.
Possibly important context: I only started really using it a few weeks ago.
I used Opus 4.6 Extended
Stop being cheap, OP. You clearly just need to shell out multiple billions of dollars for access to mythos /s
You’re probably done with this. But if you give claude a test case or two (or have it try to make them) you can have claude run the test case, and then it will iterate.
Also, aggressively use plan mode and if claude screws up more than three times do /clear, explain that it’s screwing up to it and then give it new instructions.
The solutions it generated were almost write every time, but critically broken in subtle ways, and any attempt to fix the problems would either introduce new bugs, or regress with old bugs.
This is part of your problem right there. The correct word there, instead of “write”, is “right”. You emotionally typed out a message, got your dopamine hit, then felt satisfied, and now the rest of us have to figure out what you meant to say.
Which is fine, but now imagine that not only you can do this, but AI can do it as well…
If you want something done correctly, then you must do it yourself.
I tried using Claude to convert some bash scripts to docker compose files, and it made several mistakes with case-sensitivity and failure to properly encapsulate certain path declarations that had spaces in them. if it could make such incredibly simple mistakes in converting a script to a markup language, I wouldn’t dare trust it to actually compose anything in an actual programming language like Python or Rust or C# or Swift whatever you’re using.
my experience with LLM’s and numerical computations like with MATLAB or GNU octave, has been poor. I assume its more of an issue that the data isn’t there, MATLAB has it’s own proprietary AI (which I don’t believe is trained on users code) and Octave has no AI associated on it’s end so the major LLM’s only get trained by the data it is prompted by users online or otherwise. Which is why if you prompt it to do a 3D plot, it will almost always pull something out of it’s ass.
your feeling of a “mental-fog” is my experience with AI in general, the language model explains the ideas well, but then the code editor does some obscure move that makes no fucking sense. also, because you’re not programming it and learning from your mistakes it makes you uncertain of your code. its unfortunate to see search engines are going to shit because of AI, because AI is not ready.
I have yet to be able to vibe code anything relatively involved. The closest I’ve come is a ffmpeg wrapper script to edit out scenes from a video with a fade in/fade out title card. But even then, I ended up at some point having to debug and add my own arg support because it kept screwing things up. The first draft did do something, though.
I find at this point that it’s still only useful if I have a very clear goal in mind with a lot of context on the area I need to make changes to. That lets me get a more specific prompt, and then I’ll still need to review the output. I have only ever gotten a successful one shot like this with tests.
