Ask HN: Where have you found the coding limits of current models?
I'd love to hear where you've run into the limits of current coding models, whether it's based on lines of code, systems integration, databases, etc.
If you stick to targeted problems with well-described prompts, acceptance criteria, and lots of linting, unit testing, and integration testing, you'll typically get what you want with code that looks okay. And when things start to stray, it's easy to get things back on track.
It's when you start trying to have LLMs write too much without a human review that you start getting unnecessary function chains, abstractions that aren't needed, code that doesn't really match the existing style, duplicate code, missing functionality, hallucinated functionality, tautological tests, etc.
It works best when there's regular feedback in the loop about what's good and what's not good. Testing and linting can fill in some of that, but we still need a human in the loop with "taste", so to speak.
I have found it very helpful to ask AI to review the latest changes - it often finds serious problems in review of code it just wrote.
One somewhat related thing I've noticed from all the major LLMs is a tendency to 'fix' multithreading race conditions by inserting delay calls of seemingly random amounts of milliseconds to 'give things time to settle'.
It kinda makes sense why they do it because I've seen similar human generated code over the years, but it doesn't instill a lot of confidence in unreviewed 'vibecoded' systems.
I had a mistake in which a large back-up file deletion event happened during a robocopy. 600gb of files got 'deleted' (file headers toast etc). Trying to get the LLMs to understand the hunt parameters, what to focus on, what not to focus on - none of them could reasonably come close to doing file content recovery properly. I needed to build a custom solution because the available industry options couldn't do what was required and the LLMs were useless for that (including the latest versions of Claude, Gemini and GPT). They just went around in circles, capped by their apparently weak knowledge of file recovery as a field. That is, creativity was their limitation.
Also any situation where the context window is even remotely close to being full. At 80% the degradation is noticeable enough to make me start from scratch
For example, models repeatedly try and not use my Makefile's (build / run / test / test-file / check) and instead spend several cycles attempting to invoke using system tools and getting it wrong.
I've got to the point where I run most models with an extra path folder which overrides typically bad commands tells them off and redirects them back (more effective and portable than hooks). But then sometimes the model reads the makefile, discovers the workaround, and does the stupid anyway.
It did the above twice in a row around different parts of the API.
The thought that there are almost certainly devs out there merging Claude PRs without the skills or volition to push back on its screwups is not comfortable.
There's also some types of code that I believe is often wrong in the training data that is almost always wrong in the LLM output as well. Typically anything that should have been a state machine, like auth flows, wizards, etc.
When all is said and done I think the main savings come from the high throughput of low-value generic solutions. I don't currently see this changing, and the reason is that high quality products cannot be generated without specifying a lot of details. Of course, we may not want quality.
I'm curious about what you mean about state machines for auth. If it's a state machine, how do you interface with it, what data type do you use, how is it built? What language or framework would this apply to? I think your approach might be different from how other people do it, I'd like to know more about your approach. Could you walk through an example?
But one big limit is the DX. Their choice of DX is usually abysmal - ironically, just like an average devs. They seem to lack the aesthetic instinct for code, so you have to really point them hard into the direction or provide a sample of the expected DX, for them to still fight against it at every turn.
While understandable in a way, as they are trained on average code and most code will now be written by the machines anyways making the DX "less relevant", it's also a giant code smell, as bad DX tends to point towards bad internals and wrong decisions along the way.
So not really a technical limit - they swallow anything you throw at them, even the most complex cases - but more of an aesthetic limit in terms of taste.
Eventually I asked it "hey, are you sending two requests when you could send one?" Claude thought about it for a minute and said, "you're right! Let me fix that." The 429 errors stopped.
I've found it really is more like pair programming than having another fully independent developer. For Jenkins pipelines, I don't care about hardcoded string compares as much. For the core capability of the software, details are important.
But the last few weeks Opus 4.6 seems to have got dumb again. Now it is making way more mistakes and forgetting useful things and recent context it used to manage.
I am guessing this is just Anthropic quietly dialling down the real effort as they either downsize or free up compute for someone or something else.
are you logging access patterns and times to see when the degradation occurs?
- Plenty of API hallucination happening on cutting edge Spark (4.0.0+) functionality, especially PySpark. Spark bares some blame here for broken and incomplete documentation. Takes a human in the loop to realize that the documentation is misleading or wrong or missing.
- API design. I’ve found that, unless specifically steered towards “good” API design (highly subjective), agents tend to just add another endpoint / function to satisfy the exact task at hand, with total disregard to how the rest of the API looks. (Pretty much exactly what a junior engineer would do…)Unless I'm very clear at all times it will write code like the most annoying stubborn junior you've ever worked with. Nothing is sacred, everything can be abbreviated, shortened, made more confusing, made less readable, and concepts like readability or naming conventions are not even considered.
It also adds superfluous nonsense comments that don't explain the "why".
A good API does two things. Firstly it DRYs the code out. Many APIs start life doing only that, as a collection of routines you get tired of writing over and over again. Secondly, the functions are designed in a way that reduces the need to share information. Typically they do that by hiding a whole pile of details in their implementation so knowledge of those details is all in one place rather than scattered across a code base. A term often used for APIs that don't do that well is "leaky", or we say it's a "leaky abstraction".
Perhaps a good API has other more subjective attributes but it must have those two. LLMs suck at both. You can see that in the comments here, when people say they write verbose code. It's verbose because the LLM didn't go looking for duplicated functionality - if it needed it, it just put the code where it was focused on at the time.
If they are bad at DRY then I need a better superlative to describe how they fare at respecting the isolation boundaries that underpin good module design. As far as I can tell, they have no idea about the concept or how to implement it. Let loose, they are like a bull in a china shop, breaking one boundary after another.
I feel like a human would give up a lot quicker and start to learn where the limits are. Claude spins in circles convinced it's finally found a solution. Again. And again. And eventually gets back to where it started.
Also, I see this more as a feature than a bug. In many projects I inherited in the past, I wish the original devs were a bit more verbose. Then again, with every developer using LLMs now, probably the opposite applies now.
I tend to re-enjoy handrolling code more. I delegate the stuff that annoys me.
We have a large multiplatform codebase, the issue seems to be more the time it takes to navigate the code and reason about it, rather than the size. Arguably the size is causing 'them' to be slower in that regards, but I haven't found the limit yet. And with compaction, it's even less of a problem.
my 5c.
Proper escaping of layered syntaxes in Ansible on the first attempt
Writing bare-metal embedded Rust, although this was long ago, so not current models
The reason is that sometimes it spits out something or does a workflow that's pretty sophisticated, and sometimes it fails spectacularly in the most basic ways.
I don't think there is a complexity or domain knowledge limit as there would be with a human. Or at least not in the same sense. As long as it can repeat and remix patterns that it is trained on, then it will do its thing well. The same seems to be true for "reasoning" loops and workflows. It can spit out code that has been done N times before in a similar manner for a large N.
They can still break down because of very trivial issues and assumptions that happen to be baked in, go off the rails and get stuck long loops that are completely insane if you think of them as imitating human programmers.
When I use an agent, I always interview it first about the task. Ask how they would go about it, probe them, give them info that they lack.
Never go from prompt to action. Have them define their approach first, then split the approach into pieces, from gathering data to cleaning it up and so on. If applicable, front-load work that can be achieved with scripts, so you have testable and repeatable steps rather than let it go wild.
So the TLDR is: I think the limitation is simply that it's a non-deterministic token machine that produces useful results enough of the time so it appears to be reasonable.
My current workflow with Mistral or Claude to implement features is to write a playbook (like a developer guide) with them first on how to implement features in the current repository/source/project.
For example, something like:
Implement the feature using a, b, c as a blueprint (architecture, tests, code, documentation, and other things like styleguides, commands for checks or tests, linter, formatter, frameworks / versions and standards etc.). On features: Write the tests for xyz first. Then implement x, run the test for x, should now be green - and so on. Implement a feature means: Write tests, the code, documentation - and so on. Feature complete means: Tests are green, code is formatted and linted, documentation is available - and so on. Good code means ...
The playbook approach works, even if the chat context becomes too large after some time. If I notice that, I have the model reread the playbook. The playbook is also a living document; usually, I ask the LLM at the end if it wants to add any changes or additions.
But the playbook thingy might become an issues, if it get to long. 200-500 lines works best for me atm.
I once had claude code write a python script that emulated the output of my training script, including pretending that loss is decreasing. Why? Because it was unable to install a python dependency.
Everytime I use a coding agent, I need to double check that it's not cutting corners, hard coding things that shouldn't be or straight up rewriting failing test cases.
What I need more is honesty.
Last night the chatbot was unable to make a thing work, so it ended up spiraling and doing things like ripping all authentication out of a service. Then it just waved it off as, "it's just a dev service it'll be fine." I didn't tell it to rip all the authentication out, it just got frustrated. Yeah, I'm anthropomorphizing, but that's essentially how it went.
When the bot says, "I'm ready to throw something. Just rip it out and I'll deal with it another day." I can't help but sympathize.
Its like the agent must succeed at all costs, even if it means doing some insane solution
It needs to just straight up fail sometimes but its like the models are not trained to allow that