Karpathy on Programming: “I've never felt this much behind”
twitter.com549 points by rishabhaiover 3 months ago
549 points by rishabhaiover 3 months ago
I admit to pangs of this, but it's really never made any sense because the implication is that the profession is now magically closed off to newcomers.
Imagine someone in the 90s saying "if you don't master the web NOW you will be forever behind!" and yet 20 years later kids who weren't even born then are building web apps and frameworks.
Waiting for it to all shake out and "mastering" it then is still a strategy. The only thing you'll sacrifice is an AI funding lottery ticket.
Finally a voice of reason. The tools will just get better and easier to use. I use LLMs now, but I'm not going to dump a bunch of time learning the new hotness. I'll let other people do that and pickup the useful pieces later.
Unless your gunning for a top position as a vibe coder, this whole concept of "falling behind" is just pure FOMO.
Same. I only just started using agents a few months ago.
Earlier this year the ecosystem was still a mess I didn't have time to untangle. Now things are relatively streamlined and simple. Arguably stable, even.
I feel behind, sure, but I also don't think people on the bleeding edge are getting that much more utility that it's worth sinking dozens or hundreds of my very limited hours into understanding.
Besides, I'm a C programmer. I'll always be several decades behind the trend. I'm fine with that.
Doing small project for customer. They have explicit instructions that I can't even use some unapproved AI... So well they are paying. So until it is actually forced I see no pressure to move there.
And rest of my field. Automated tools do part of work. AI probably some, but not enough of actually verifying findings and then properly explaining the context and implications.
Yeah Karpathy is engaged here in more hype creation. Software engineers pretending they just smashed some particles together and there is a whole lot of new data to math out.
It's high dose copium. Please keep the good times rolling! Buy my books! Sub to my stack!
Meanwhile, with local models, local RAG, and shell scripts, I am wandering 3D immersive worlds via a GPU accelerated presentation layer I vibe coded with a single 24GB GPU. Natural language driven Unreal engines are viable outputs today given local only code gen.
Karpathy and the SV VC world thought this would be the next big thing to pump for a decade plus; like web pages and SaaS. But the world is smarter, more adept at catching up that it is just state management in a typical machine. The semantics are well known and do not need re-invention.
The hilarity at an entire industry unintentionally training their replacements.
>> Meanwhile, with local models, local RAG, and shell scripts, I am wandering 3D immersive worlds via a GPU accelerated presentation layer I vibe coded with a single 24GB GPU. Natural language driven Unreal engines are viable outputs today given local only code gen.
what drugs are you using?
What on Earth have you used to get reasonable results out of a local model?
I've tried at every new model release (that can run on my 24GB card) and everything is still entirely useless.
I'm not writing web stuff though.
Yeah that's my view too. It's definitely fine to wait a couple of years (at least), and see what emerged as most effective and then just learn that, instead of dumping a ton of time now into keeping up with the hamster wheel.
Unless you're in web dev because it seems like that's one of the few domains where AI actually works pretty well today.
Or if you like learning new stuff. Personally that has been best part of being programmer.
I love learning new stuff, but for whatever reason the AI stuff doesn’t interest me. So I learn other stuff, only so much time in the day.
I like learning new stuff, but not if it's going to be completely obsolete in 6 months.
> Unless your gunning for a top position as a vibe coder, this whole concept of "falling behind" is just pure FOMO.
???
The person you're quoting has a point. Everyone is losing their minds about this. Not everyone needs to be on top of AI developmemts all the time. I don't mean you ignore LLMs, just don't chase every fad.
The classic line (which I've quoted a few times here) by Charles Mackay from 1841 comes to mind:
"Men, it has been well said, think in herds; it will be seen that they go mad in herds, while they only recover their senses slowly, and one by one.
"[...] In reading The History of Nations, we find that, like individuals, they have their whims and their peculiarities, their seasons of excitement and recklessness, when they care not what they do. We find that whole communities suddenly fix their minds upon one object and go mad in its pursuit; that millions of people become simultaneously impressed with one delusion, and run after it, till their attention is caught by some new folly more captivating than the first."
— Extraordinary Popular Delusions and the Madness of Crowds
Thank you for the subtitles, it's not like I didn't understand the lingo, I just couldn't make sense of the implied meaning.
And here I am partying (coding) like it's 90s (C++ desktop apps) and web never happened... :)
It's pretty nice that C has garnered such hate because there's apparently very little focus on getting LLMs to write good C. It's all Rust and Python and whatever this month's fad language is. LLM fans mostly leave us alone apart from the "C bad rewrite the world in rust" crew.
I'm very happy being decades behind the curve here. C's slowness is perfect for me.
Eh, for myself as a middle-aged software engineer, it feels a little like the last chopper out of Saigon. I feel less and less confident that I can make as good a living in software for the next decade as I have for the last couple. Or if I want to. The job is changing so fast right now, and I’m not sure I like it. When I worked in big tech, I preferred being an IC over an EM or tech lead because I like writing code. Now it feels increasingly like you can’t be an IC in that way anymore. You’re now coding through others, either humans or AI.
Sure, I can write code manually, but in my case I’m working full time on my own SaaS and I am absolutely faster and more effective with AI. It’s not even close. And the gains are so extreme that I can’t justify writing beautiful hand-crafted artisanal code anymore. It turns out that code that’s “good enough” will do, and that’s all I can afford right now.
But long-term, I don’t know that I want to do that work, especially for some corporation. It feels like the difference between being a master furniture craftsman, and then going work in an IKEA factory.
You'll make more money than ever cleaning up AI generated messes.
I had few projects like that this year and I can say it how messy and demotivating its to cleaning up mess.
And its actually not well paid because client now has the expectation that mostly everything is now done, you have to just only fix few things and you even have AI at your disposal so expect that you just write a better magic prompt.
I think actually often its faster and cheaper to start from scratch or at least rewrite whole module (of course still with AI with just better vibe engineering rather than vibe coding).
It's similar with house renovation - often its just cheaper and faster to tear whole building down rather than fixing it.
Would you be able to share any more details on the clean up projects you had to do? Like, wasn't front or back end, which tech stack, where were the LLM code issues etc.
I'm just very curious where we are at the moment with in this profession.
the project was iOS app and vibe coded in Claude Code - it was around half year ago so maybe things improved. Client actually knew some coding so actually quite impressive how far they did manage to go along.
However it was just adding pile of feature after features without taking time to refactor it. Client most likely did some few different attempts to add some specific feature or fixing something and there was a lot of dead code that haven't been used. This dead code actually confused AI and often tried to modify part of code that have been abandoned.
There was completely no tests. No performance tests. And some part of my job was to improve performance (cv/ai model inference) and robustness (crashes, memory leaks).
I think AI is fine and useful but whats bad with such vibe coded project if somebody hand over to you is you have completely no clue what part of the code are written/designed properly with good foundation if previous developer didn't test extensively and didn't refactor continuously. Even worse if you cannot talk to previous developer responsible for the project.
Not OP, but I’ve spent months cleaning up sqlalchemy models that were written in isolation using AI. Project was just not scalable.
First, I’m highly skeptical of that, especially over the course of the next decade.
Second, do you actually want to do that work? I don’t. I spent years working as a freelancer and I cleaned up a lot of shitty code from other freelancers. Not really what I want to spend my 50s doing.
I greatly respect your opinions here but I really doubt that would ever happen.
It's already happening. My buddies are in the 'late bloom' phase of their careers and they are doing quite well as of late.
AI supported coding is like four wheel drive: it will get you stuck but in harder places. The people that use these tools to reach above the level of their actual understanding are creating some very expensive problems. If you're an expert level coder and you use AI to speed up the drudgework you can get good mileage out of them, but if you're a junior pretending to be a senior you're about to cost your employer a lot of $ hiring an actual senior.
One thing I’ve noticed is that some folks are over-confident about the benefits of LLM’s and seemingly gloss over the implicit costs.
And for good reason - the ill disciplined human body optimises for short term benefits. The disciplined body recognises the flaw in this and thinks much broader.
But wouldn’t the models get better at fixing complicated code eventually?
We don't know. We seem to be hitting diminishing returns, but we don't exactly know where it will stop
Is there a source for this? Scaling laws work and we have about 4 orders of magnitude in the exponential growth before we run into true bottlenecks
What I like to say is that writing software is getting so easy that I don't know how to do it anymore.
If anything I'd expect all these tools to be easier for new engineers to adopt, unburdened by how things were before.
> unburdened by how things were before.
What burden are you talking about? Using LLMs isn't that hard, we have done harder things before.
Sure, there will be people that refuses to "let go" and want to keep doing things the way the like them, but hey! I've been productive with vim (now neovim) for 25 years and I work with engineers that haven't mastered their IDEs at the same level. Not even close!
Sure, they have have never been "burdened" by knowing other editors before those IDEs existed, but claiming that I would have it harder to use any of those because I've mastered other tools before is ridiculous.
Not sure how to address this without just restating TFA. Not all change builds on existing knowledge, and sometimes it is so rapid that keeping up is difficult.
Absolutely agree.
I took this approach when the Kubernetes hype hit and it never limited my prospects.
This argument only makes any sense at all because the demand for software developers continually grew.
As long as more software developers are needed your logic obviously holds, it is irrelevant whether you are a master. There are enough jobs for "good enough". But what if "good enough" is no longer a viable economic niche? Since that niche is now entirely occupied by LLMs.
People did say that in the 90s. Hence the rush to put everything on the web, whether there was any real business case for it or not. And most of it went up in flames at the end of that decade.
Tell that to recruiters! If you're senior you're always expected to know everything.
I feel like many people in the comments aren't aware that Karpathy is an ML scientist for whom programming is a complementary skill, not a profession. The only reason he came up with "vibe coding" is because maximum complexity of his hobby projects made it seem believable. Maybe take his opinions about fate of programming with a grain of salt.
He is brilliant no doubt, but not in that field.
He's a pretty decent programmer.
It's interesting that some months ago when his nanochat project came out the HN Anti-AI crowd celebrated him saying "I tried to use claude/codex agents a few times but they just didn't work well enough at all and net unhelpful, possibly the repo is too far off the data distribution"
But now it is working for him he's suddenly not an expert...
What you’re calling the “crowd” was not the same people. Every time someone makes a claim like yours, I go and check and don’t see the same usernames in the conversation. “Different people have different opinions and different ways to express them” isn’t really an insight; it tells us nothing nor does it make anyone worthy of criticism.
You can’t, in an honest argument, lump different strangers into a group you invented to accuse them of duplicity or hypocrisy.
Having created 100 of nano-sized projects does not add up to having developed and maintained one large code base.
Coding agents are eating up programming from the lowest end, starting from pressing button on the keyboard to type the code in: completion was literally their first application. I don't think it will go all the way to the top, though, the essential part of the profession will remain until true AGI.
Metaphorically, think how integrated chips didn't replace electrical engineering, just changed which production tools and components engineers deal with and how.
Obviously we all are adapting to changes, but if he or someone are panicking about being behind, that can only be because they've never been in too deep.
> But now it is working for him he's suddenly not an expert...
Or maybe he didn't lie then but is lying now?
Calling him a liar seems fairly unnecessary? For one thing people's minds can change, or that can be talking in different contexts. Or - as in this case - new technology could have been deployed that changed the game.
Maybe that's true, but I will say that one of the reasons I recommend his Python ML videos to people is not just the ML content but also his Python is good and idiomatic. So I would not agree; I think his programming is a well practiced skill.
FWIW though I think his predicted worldview will render it very difficult to acquire this skill, as people grow reliant on gen AI for programming rudiments.
As far as "programming skill" goes, writing "good and idiomatic" Python is pretty bottom of the barrel. I don't think the GP is all that off, most people who are famous for some programming-adjacent skill (or even programming) aren't good at programming.
>As far as "programming skill" goes, writing "good and idiomatic" Python is pretty bottom of the barrel.
Complete bullshit. Beginning programmers writing good and idiomatic Python isn't "bottom of the barrel", or did you think I was recommending his videos to 20 year seasoned pros to improve their coding?
Some people on this site need to check their arrogance and humble themselves a bit before opening their mouths.
Exactly, I would put more weight on this if it were coming from someone who actually works as a regular programmer in the industry
As an Opus user, I genuinely don’t understand how someone can work for weeks or months without regularly opening an IDE. The output almost always fails.
I repeatedly rewrite prompts, restate the same constraints, and write detailed acceptance criteria, yet still end up with broken or non-functional code.its very frustrating to say the least Yesterday alone I spent about $200 on generations that now require significant manual rewrites just to make them work.
At that point, the gains are questionable. My biggest success is having the model take over the first Design in my app and I take it from there, but those hundred lines if not thousand lines of code it generates are so Messi, it's insanely painful to refactor the mess afterwards
I have a hell of a time just getting any LLM to write SQL queries that have things like window functions, aggregates and lateral left joins - even when shoving the entire database schema DDL into the context.
It's so frustrating, it regularly makes me want to just quit the profession. Which is why I still just write most code by hand.
I write a lot of SQL and I haven't had these issues for months, even with smaller models. Opus can one shot most of my queries faster than I could type them.
Instead of stuffing the context with DDL I suggest:
1. Reorganize your data warehouse. It needs to be easy to find the correct data. Make sure you use ELT clear layers, meaningful schemas, and have per-model documentation. This is a ton of work, but if done right the payoff is massive.
2. I built a tool for myself to pull our warehouse into a graph for fuzzy search+dependency chain analysis. In the spring I made an MCP server for it and Claude uses that tool incredibly well for almost all queries. I haven't actually used the GUI or scripts since I built the MCP.
Claude and Devstral are the best models I've used for SQL. I cannot get Gemini to write decent modern sql -- even the Gemini data science/engineer agents in Google Cloud. I occasionally try the paid models through the API and still haven't been impressed.
>> I write a lot of SQL and I haven't had these issues for months, even with smaller models. Opus can one shot most of my queries faster than I could type them.
Same. SOTA models crush every SQL question I give them.
I think this might be a big part of the problem with the conversation about AI right now. The models have become so much better in the last ~6 months in my experience and lots of people wrote them off 1-2 years ago after they couldn't do x and 'we've hit a wall' was being thrown around everywhere.
If you really know SQL, writing an SQL query basically just feels like writing a prompt for a database client anyway, except it does exactly what you ask for.
I have a running joke at work.
* LLMs are just matrix multiplication. * SQL is just algebra, which has matrix multiplication as part of it. * Therefore SQL is AI * Now who is ready to invest a billion dollars in our AI SaaS company?
Or it’s just that astronaut with a gun meme: “Wait AI is just SQL?….Alway has been.”
My trick is to explicitly roll play that we’re doing a spike. This gets all of the models to ignore all of the details they normally get hung up on. Once I have the basics in place, I can tell it to fix details.
It’s _always_ easier to add more code than it is to fix broken code.
Most people have not fully grasped how LLM's work and how to properly utilize agentic coding solutions. That is the reason for issues when it comes to vibe coders having low quality code. But that is not the limitation of technology but the user (at this stage). Basically think of it this way everyone is the grandma that has been handed a palm pilot to use to get things done. Grandma needs an iPhone not a palm pilot but the problem is that we are not in that territory yet. So now consider the people who were able to use the palm pilot very successfully and well, they were few and they were the exception, but they existed. Same here. I have been using coding agent for over 7 months now and have written zero lines of code, in fact I don't know how to code at all. But i have been able to architect very complex software projects from scratch. Text to speech , automated llm benchmarking systems for testing all possible llama.cpp sampling parameters and more, and now im building my own agentic framework from scratch. All of these things are possible and more without writing one line of code yourself. But it does require understanding how to use the technology well to get this done.
If you don't know how to code then you are not able to judge what your producing accurately.
here you go I open sourced one of the projects https://youtu.be/EyE5BrUut2o
All of the applications you mention could be scoped as beginner projects. I don't think they represent good proofs of capability.
Well why don't you look at it for yourself and tell me if this looks like a beginner project https://youtu.be/EyE5BrUut2o
Yes, this does look like a beginner project & exactly what i expected from someone who doesn't write code.
This is extremely simple software.
Claude is extremely verbose when it generates code, but this is something that should take a practicing software engineer an hour or so to write with a lot less code than Claude.
I like all the LLM coding tools, they're constantly getting better, but I remain convinced that all the people claiming massive productivity improvements are just not good software engineers.
I think the tools are finally at the point where they are generally a help, rather than a net waste of time for good engineers, but it's still marginal atm.
I hardly ever open an IDE anymore.
I use Claude Code and Cursor. What I do:
- use statically typed languages: TypeScript, Go, Rust, Python w/ types
- Setup linters. For TS I have a bunch of custom lint rules (authored by AI) for common feedback that I've given. (https://github.com/shepherdjerred/monorepo/tree/main/package...)
- For Cursor, lots of feedback on my desired style. https://github.com/shepherdjerred/scout-for-lol/tree/main/.c...
- Heavy usage of plan mode. Tell AI something like "make at least 20 searches to online documentation", support every claim with a reference, etc. Tell AI "make a task for every little thing you'll implement"
- Have the AI write tests, particularly the more expensive ones like integration and end-to-end, so you have an easy way to verify functionality.
- Setup Claude Code GHA to automatically review PRs. Give the review feedback to the agent that implemented it, either via copy-pasting or tell the agent "fetch review comments and fix them".
Some examples of what I've made:
- Many features for https://scout-for-lol.com/, a League of Legends bot for Discord
- A program to generate TypeScript types for Helm charts (https://github.com/shepherdjerred/homelab/tree/main/src/helm...)
- A program to summarize all of the dependency updates for my Homelab (https://github.com/shepherdjerred/homelab/tree/main/src/deps...)
- A program to manage multiple instances of CLI agents like Claude Code (https://github.com/shepherdjerred/monorepo/tree/main/package...)
- A Discord AI bot in the style of my friends (https://github.com/shepherdjerred/monorepo/tree/main/package...)
> make at least 20 searches to online documentation
Lol sometimes I have to spend two turns convincing Claude to use its goddamn search and look up the damn doc instead of trying to shoot from the hip for the fifth time. ChatGPT at least has forced search mode.
I've found that telling it to specifically do N searches works consistently. I do really wish Claude Code had a "deep research" mode similar to 'normal' Claude.
Thanks for sharing. So the dumb question - do you feel like Claude Code & Cursor have made you significantly more productive? You have an impressive list of personal projects, and I can see how a power user of AI tools can be very effective with green field projects. Does the productivity boost translate as well to your day job?
For personal projects, I have found it to be transformative. I've always struggled with perfection and doing the "boring parts". AI has allowed me to add lots of little nice-to-have features and focus less on the code.
I'm lucky enough that my workplace also uses Cursor + Claude Code, so my experience directly transfers. I most often use Cursor for day-to-day work. Claude has been great as a research assistant when analyzing how data flows between multiple repos. As an example I'm writing a design doc for a new feature and Claude has been helping me with the investigation. My workflow is more or less to say: "here are my repos, here is the DB schema, here are previous design docs, now how does system X work, what would happen if I did Y, etc."
AI is still fallible so you _do_ of course have to do lots of checking and validation which can be boring, but much easier if you add a prompt like "support every claim you make with a concrete reference".
When it comes to implementation, I generally give it smaller, more concrete pieces to work with. e.g. for a personal project I would say something like "here is everything I want to do, make a plan, do part 1, then do part 2, example: https://github.com/shepherdjerred/scout-for-lol/tree/227e784...)
At work, I tend to give it PR-sized units of work. e.g. something very well-scoped and defined. My workflow is: prompt, make a PR on GitHub, add comments on GitHub, tell Cursor "I left comments on your PR, address them", repeat. Essentially I treat AI as a coworker submitting code to me.
I don't really know that I can quantify the productive gain.. I can say that I am _much_ more motivated in the last few months because AI removes so much friction. I think it's backed up by my commit history since June/July which is when I started using Cursor heavily: https://github.com/shepherdjerred
Cursor is an IDE.
Oh to clarify I used to use Cursor but the last month or two I've used Claude Code almost exclusively. Mostly because it seems to be more generous with credits.
This is what an AGENTS.md - https://agents.md/ (or CLAUDE.md) file is for. Put common constraints to correct model mistakes/issues with respect to the codebase, e.g. in a “code style” section.
What does your software creation workflow look like? Do you have a design phase?
Why would you spend $200 a day on Opus if you can pay that for a month via the highest tier Claude Max subscription? Are you using the API in some special way?
At a guess an Enterprise API account. Pay per token but no limits.
It’s very easy to spend $100s per dev per day.
The $200/month plan doesn't have limits either - they have an overage fee you can pay now in Claude Code so once you've expended your rate limited token allowance you can keep on working and pay for the extra tokens out of an additional cash reserve you've set up.
> The $200/month plan doesn't have limits either... once you've expended your rate limited token allowance... pay for the extra tokens out of an additional cash reserve you've set up
You're absolutely right! Limited token allowance for $200/month is actually unlimited tokens when paying for extra from a cash reserve which is also unlimited, of course.
I think you may have misunderstood something here.
When paying for Claude Max even at $200/month there are limits - you have a limit to the number of tokens you can use per five hour period, and if you run out of that you may have to wait an hour for the reset.
You COULD instead use an API key and avoid that limit and reset, but that would end up costing you significantly more since the $200/month plan represents such a big discount on API costs.
As-of a few weeks ago there's a third option: pay for the $200/month plan but allow it to charge you extra for tokens when you reach those limits. That gives you the discount but means your work isn't interrupted.
Extra Usage for Paid Claude Plans: https://support.claude.com/en/articles/12429409-extra-usage-...
Thank you for the explanation, but I did fully understand that is what you were saying.
What I don't fully understand is how you can characterize that as "not limited" with a straight face; then again, I can't see your face so maybe you weren't straight faced as you wrote it in the first place.
Hopefully you could see my well meaning smile with the "absolutely right" opening, but apparently that's no longer common so I can understand your confusion as https://absolutelyright.lol/ indicates Opus 4.5 has had it RLHF'd away.
When I said "not limited" I meant "no longer limits your usage with a hard stop when you run out of tokens for a five hour period any more like it did until a few weeks ago".
That's why I said "not limited" as opposed to "unlimited" - a subtle difference in word choice, I'll give you that.
Oh, I wasn't arguing that it isn't "easy to spend $100s per dev per day". I was just asking what the use-case for that is.
I’ve had decent results from it. What programming language are you using?
Sometimes I have a similar file or related files. I copy their names and say use them as reference. Code quality improves by 10 times if you do so. Even providing a a example from framework's getting started works great too for new project.
Yeah the pain of cleaning up small mess is great too. I had some tests failing and type failing issues, I thought I will fix it later by only using AI prompt. As the size was growing, failing Typescript issues was growing too. At some point it was 5000+ type issues and countless number of failing unit tests. Then more and more. I tried to fix with AI, since it was not possible fixing old way. Then I discarded the whole project when it was around 500k lines of code.
Question: How many LoC do you let the AI write for each iteration? And do you review that? It sounds like you are letting it run off leash.
I had no idea how it would end up. It was first time using AI IDE. I had only used chatgpt.com and claude.ai for small changes before. I continued it for the experiment. I thought AI write too many tests, I will judge based on test passing. I agree, it was bad expectation + no experience with AI IDE + bad software engineering.
I am a software developer and mainly a programmer for decades now. I love programming. I love to be "once" with the computer. I will never give this joy up. If I need to sell shoes at daytime, I will program real computer programs in the evenings. If it won't be possible with modern machinery anymore, I will take my Commodore 64. I am a free man.
Edit: Corrected since/for. :-)
you mean "one" not "once" right?
If you have never experienced becoming once, try DMT or a large amount of acid on a quiet day.
The only time I've felt this much behind was in high school when everyone was talking about how much sex they were having.
AI code is the Canadian girlfriend of programming.
Not being from the US, it took me a moment to realise that "she's in Canada" is a nice excuse for why nobody has met her.
Time to shell out the $200 my friend.
You need a lot more money than $200 if your code base is more than 100,000 lines.
If only more people understood what quadratic attention means in the real world.
Are you saying the LLM costs grow quadratically in the size of the code base? The prompts are already highly subsidized, can't wait to see what happens when they charge the actual price to the consumer.
For now I'm gonna enjoy my VC subsidized burrito deliveries while I can.
The real question to ask is are you receiving the delivery or are you the courier training your replacement :)
> OpenAI's sales and marketing expenses increased to _$2 billion_ in the first half of 2025.
Looks like AI companies spend enough on marketing budgets to create the illusion that AI makes development better.
Let's wait one more year, and perhaps everyone who didn't fall victim to these "slimming pills” for developers' brains will be glad about the choice they made.
Well. I was a sceptic for a long time, but a friend recently convinced me to try Claude Code and showed me around. I revived an open source project I regularly get back to, code for a bit, have to wrestle with toil and dependency updates, and loose the joy before I really get a lot done, so I stop again.
With Claude, all it took to fix all of that drudge was a single sentence. In the last two weeks, I implemented several big features, fixed long standing issues and did migrations to new major versions of library dependencies that I wouldn’t have tackled at all on my own—I do this for fun after all, and updating Zod isn’t fun. Claude just does it for me, while I focus on high-level feature descriptions.
I’m still validating and tweaking my workflow, but if I can keep up that pace and transfer it to other projects, I just got several times more effective.
This sounds to me like a lack of resource management, as tasks that junior developers might perform don't match your skills, and are thus boring.
As a creator of an open-source platform myself, I find trusting a semi-random word generator in front of users unreliable.
Moreover, I believe it creates a bad habit. I've seen developers forget how to read documentation and instead trust AI, and of course, as a result AI makes mistakes that are hard to debug or provokes security issues that are easy to overlook.
I know this sounds like a luddite talking, but I'm still not convinced that AI in its current state can be reliable in any way. However, because of engineers like you, AI is learning to make better choices, and that might change in the future.
> a semi-random word generator
Calling tools like Claude Code a "semi-random word generator" is certainly a choice, and I suspect it won't age well.
> as tasks that junior developers might perform don't match your skills, and are thus boring.
Yeah this sounds interesting, and matches my experience a bit. I was trying out AI for the Christmas cuz people I know are talking about it. I asked it to implement something (refactoring for better performance) that I think should be simple, it did that and looks amazing, all tests passed too! When I look into the implementation, AI got the shape right, but the internals were more complicated than needed and were wrong. Nonetheless it got me started into fixing things, and it got fixed quite quickly.
The performance of the model in this case is not great, perhaps it is also because I am new to this and don't know how to prompt it properly. But at least it is interesting.
This sounds a lot like the classic "the way to get a good answer on the internet is to post a wrong answer first", but in reverse - the AI gives you a bad version which trolls you into digging in and giving the right answer :-)
<musing aloud mode>
I think AI coding should not be permitted in the first two years of training in CS. One should have to learn the basics of reading quality documentation, creating quality code and documentation, learning how the different pieces of software work together, and learning how to work with others.
LLMs are great for people with some idea of what they're doing, and need "someone else" to pair program with. I agree it will cripple the architectural thinking of new learners if they never learn how to think about code on their own.
That’s a totally fair take IMHO, and I’m very much conflicted on several ends on this topic—for example, would I want my juniors to use an agent? No; not even the mid levels, probably. As you say, it’s easy to form bad habits, and you need a good intuition for architecture and complexity, otherwise you end up with broken, unmaintainable messes. but if you have that, it’s like magic.
Let's wait one more year, and perhaps everyone who didn't fall victim to these "slimming pills” for developers' brains will be glad about the choice they made.
In that year, AI will get better. Will you?
AI is only getting better at consuming energy and wasting people's time communicating with this T9. However, if talented engineers continue to use it, it might eventually provide more accurate replies as a result.
Answering your question, no matter how much I personally degrade or improve, I will not be able to produce anything even remotely comparable in terms of negative impact that AI brings to humanity these days.
I see this logical pairing a lot.
1) AI is basically useless, a mere semi-random word generator. 2) And it is so powerful that it is going to hurt (or even destroy) humanity.
This is this is called "having your cake, and letting it eat you too".
There’s nothing incongruent about that pairing (though I also think you’re not being entirely fair in describing what your parent comment said). Atom bombs also fit: They are basically useless and they are so powerful that they can destroy humanity.
With LLMs, the destruction is less immediate and overt, but chatbots do provable harm to people, and can be manipulated to warp our sense of reality.
https://en.wikipedia.org/wiki/Chatbot_psychosis
People are having romantic relationships with their chatbots and committing suicide because of them. That is harm.
Atom bombs also fit: They are basically useless
Let's ask your friendly local Ukrainian refugee about that.
People are having romantic relationships with their chatbots and committing suicide because of them. That is harm.
So the only permissible technologies are those suitable for use by children and the mentally disturbed. I see.
> Let's ask your friendly local Ukrainian refugee about that.
You understand “basically useless” does not mean “entirely useless”, right? That’s why the word “basically” is there.
I know Ukrainian people. I know Ukrainian people who are in attacked cities right now. They are friendly, and all of them would understand my point.
> So the only permissible technologies are those suitable for use by children and the mentally disturbed. I see.
That is a bad faith argument. HN rules ask you to not do that and steel man. It is obvious that is not what I said, “permissible” isn’t part of the argument at all. And if you think one needs to be “mentally disturbed” to be affected, you are high on arrogance and low on empathy and information. There are numerous stories of sane people becoming affected.
https://archive.ph/2025.09.24-025805/https://www.nytimes.com...
Wait'll you hear about Dungeons & Dragons! As if backwards masking in rock and roll music weren't enough.
You're right, I don't have much empathy for bullshit pop-psych as an instrument of motivated reasoning. If ChatGPT can convince you to kill yourself, you weren't mentally healthy to begin with, and something else would have eventually had the same effect on you. Either that, or you were an unsupervised child, victimized not by a chatbot but by your parents. A tragedy either way, but good faith requires us to place the blame where it's actually due.
> Wait'll you hear about Dungeons & Dragons! As if backwards masking in rock and roll music weren't enough.
All ask you again to not engage in bad faith.
> If ChatGPT can convince you to kill yourself, you weren't mentally healthy to begin with, and something else would have eventually had the same effect on you.
That is false.
https://en.wikipedia.org/wiki/Suicide_barrier#Efficacy
> Research has shown suicidal thinking is often short-lived. Those who attempted suicide from the Golden Gate Bridge and were stopped in the process by a person did not go on to die by suicide by some other means. There are also a variety of examples that show restricting means of suicide have been associated with the overall reduction of it.