> My most successful work this year was hand-writing a tool, replacing ~4,000 lines of vibe-code with a few hundred. My system has proven more flexible, easier to test, and more generalizable than what it replaced.
> agents still need human taste and experience to produce good work.
Where do you think human taste and experience in software comes from?
> overstimulating & exhausting in a way that coding never was.
> for many of us, honing the code, finding the right design, seeing the parts mesh together — that was the meditative purpose we found in our jobs.
> That’s not the only time AI waylaid me. LLMs could not figure out how to annotate that model’s training data correctly. I spent days going back and forth with them
> LLMs are not yet good companions when brainstorming complicated things.
Why do we expect this to get better? The training data will degrade in quality if more people rely on LLMs so I'd expect models to eventually start getting worse even if they currently enjoy apparent improvements in output.
> Writing code is solved.
What a strange thing to say given all the above experiences, which I think we all share.
Why are we willing to give up so much in terms of purpose and understanding for perceived productivity gains?
RL is not dependent on training data. The mathocalypse didn't happen because they found better training data somewhere.
There are reasons to suspect that OpenAI may have intentionally or unintentionally trained on chat logs from mathematicians--mathematicians who were actively using OpenAI to try to to solve the same problems themselves.[1] If those mathematicians were on the right track, and getting close to a solution, that could be classified as finding "better training data somewhere."
[1] https://cybernews.com/ai-news/openai-accused-stealing-anothe...
This always reads like “writing english is solved” just because you collected snippets of famous works and have built a spell corrector.
Code is written to process data. The struggle was never to learn programming syntax (most grammar are only a few page long). The main struggle was that it’s hard to model a domain into an information system, identifying the components and their relationships, what is ingested, and what needs to be output.
Computers have integers and floats, but you need to implement units like distance, money, various counts,… The real world have words and sentences, but that needs encoding like ascii and unicode as well as the “string” concepts to make it work. Take HN, I’d bet that if you can model it in terms of data, writing code is not an issue.
If anything, I do expect the casual non-tech users to be making the most adaptable software because they have what they want in their head. Although, part of what the dev does is figuring out the weeds, and only a part of users have the patience to sit through detail corrections. And the AI user has to verbalize their wants exactly too.
Those are the people who are most likely to give a monolithic prompt and then spend a week telling the machine 'I don't care about encryption or file permissions, just make it work.' Not to mention that most people can barely verbalize the problems they have with devices they use for 8hrs a day, let alone form a full, self consistant, project idea and then put it in writing.
Even with basic pseudocode and very lenient logical rules, most people won’t bother to provide a a good definition of some process, as it’s quite unnatural to be that logical. Instead, it’s mostly just playing around. When you ask the one that is supposedly 10x-ing his productivity to explain his mental model of their own projects, it’s often a wishlist instead of a real understanding of what he has built.
For 85-90% of devs, yes this was the struggle. Creating and updating rows in a database from user input is all the “domain modeling” there is in most software. That’s what is solved by AI and it’s the majority of software dev.
The issue here is that you are learning the syntax at the same time you are learning foreign concepts, and the syntax is often confusing and counter intuitive
Learning the syntax separately is of no use because without a mental model you can't do anything with it
This is why children can easily learn enough programming to create games, because it’s a domain they can comprehend easily. Just like they can use english to write about their daily lives. It will be simple sure, but it will work.
So learning to code is fairly easy, you can easily create simple programs after a weekend. More complex programs will still use most of the same basic concepts as far as coding goes, but you will need to master more complex concepts, and that takes time. And such concepts are not about coding, they are about the nature of computing or the domain the software is created for.
This is why experienced devs only need a few days doing kata to get up to speed with a new programming language. It’s usually the same old, with slight differences. What was always needed was an understanding of computing and the ability to break down problems and create a solution that can be computed.
Coding to me is the same as drafting an engineering schematics, or writing a sheet music. It’s just a tool, not the end result.
The struggle is not where to put a semicolon. The struggle is, how do I handle validation problem X with framework Y.
As a side effect having to read the code at least once inevitably makes it way shorter and usually much better.
Guess what: because I said "both", Opus decided I was looking to pick out exactly two of the issues. If I hadn't read that reasoning log, I would've remained oblivious and thought there were only two it could find.
This is like writing `if (cond_x && cond_y) {}` and then blaming the compiler because it didn't realise that actually meant to write `if (cond_x || cond_y) {}`.
0. "find issues that are both critical and related to X"
...could be read as "issues that are X and are Y":
1. [valid, intended] find [all] issues that are[, at the same time,] 1) critical; and 2) related to X
...or as "issues, both of which issues are X and are Y":
2. [valid, your reading] find [an to-be-specified number of] issues that are, both [so I want two issues], possessing X and Y attributes.
I'd be curious to learn how you're using them.
However, I'm still split. Because this could mean that we see a jump in code quality, due to bad coders getting someone better to code for them. But we could see a decrease in code quality, due to less people "caring".
There is still questions about if that jira ticket should have been written but from ticket to mergable PR it's better than me.
It's crazy realizing that Claude Code only started getting big in the last year or so. It feels so different now, with all the new models & tools we have to use. I hope you enjoy!
I guess, its debatable that o1 was the real start- but I think its hard to deny that Claude Code CLI was the first time we had what we now call a coding harness hit the mainstream. And- Sonnet 3.7 was kinda Anthropic's "o1 moment".
God. Opus 4.6 and 5.3-codex feel like ancient history.
While trying to capture that bug, I (ironically) asked Gemini (but the Gemini-via-Google-search Gemini, since Gemini proper was an outage) for the package name of a screen recorder (gtkRecordMyDesktop); the command it returned referenced a non-existent package¹. I then found an alternative, and then hit a bug in that screen recording software that I had to then work around to get the video file there.
The site I uploaded the video to, Imgur, breaks completely — i.e., it cannot even so much as surface an error message — if your browser refuses to let it store god knows what on your disks. (So as it is, I have an extension that says "why yes, of course you can have storage, and essentially provides a black hole. But since it looks like a "success" to the Javascript, the JS doesn't crash!)
This isn't, to me, some one-off interaction. Every interaction with software these days feels like this, more and more. Just breakages everywhere that leaves me wondering if "coding is solved", where is all the bug free software?
(¹forcefully prompting the AI to actually query the package listing netted me a classic "You're absolutely right! The package name I referenced does not exist!")
My experience is the same as yours. Every single service I use. Apps, websites, doesn't matter what. Every single one seems just softly degraded. Like I can't trust that features work any more. I could report the bug, but what's the point? It's going to go to some Claude triage agent who is going to "fix" it by slopping up the app even more. Do people even use their own websites and apps any more? Apparently not - the same people declaring software solved are having their AI agents doing computer use and going on these websites for them!
Claude, please read the Bible for me. Tell me what I need to do to get into Heaven. Oh, can you dictate it to me? I'm managing an agent swarm.
Bleh.
The comments to this article suggest that it's still disputed.
Personally, I think it's a bit of a silly phrase, because it implies an absolute, and all anyone needs to do to disprove it is to show that coding was not solved for problem X. That's still the case -- it's just that the set of X is rapidly shrinking.
I think it's more indicative to say how much you trust LLM code. A little over a year ago, I scrutinised every line carefully, and it was buggy as hell and I had to fix most generated code by hand. Now, models write clean code with fewer bugs (far fewer) than I would. A year ago, I was doing a lot of steering during planning; now, it's rare that I don't agree with the model's first proposed solution, and even if I don't, it's most likely to fix a minor change in understanding, not to propose a substantial architectural change as it was before.
The one thing I'm not sure about is how much this is due to better prompting on my part. I doubt that this contributes much, but I still suspect that continual usage does make you more aware of what works and what does not: how much detail to provide, how to describe a problem or change, how to propose solutions while still leaving other options open.
LLMs still mess up a lot if you just let them loose and don't guide what they do.
We're in our own bubbles where a good part of HN population woke up to coding agents being good enough ~1 year ago, there's a group of folks that found out ~6 months before that, and then the people banging their heads on these a bit earlier than that. If you aren't using a coding agent everyday most people are just using these things to apply an instagram filter, or sort a grocery list. Muse is probably the closest thing to 'full automation' a lot of people are seeing for the first time.
Unfortunately I don't think the bubble that exists at the moment is on the promise of automating so much coding as trying to penetrate consumer use cases, which I think are still lacking and not reliable enough yet. As much as I use coding agents every day I still will not let a claw-like near a bank account or a credit card.
> my uncle can create software
Is considered great
> my uncle can create music or art
Is not considered great.
I guess humans have agreed, that art is a skill you must learn if you want to produce it, but engineering or software is not a skill worth learning and we are happy if computers can do that for us.
So it’s great and amazing if anyone can create their gizmos and software to help them but not if they create music or digital images, which apparently it is wrong to prompt, and not a sign of a great future.
If you specified to someone that the AI is replacing the farting-into-mic-core music genre, they would not care. Same with programming, the average person is not exposed to this
Big +1 to that.