Advertisement
Advertisement
⚡ Community Insights
Discussion Sentiment
59% Positive
Analyzed from 4901 words in the discussion.
Trending Topics
#code#don#llm#more#where#using#problems#llms#cars#car
Discussion Sentiment
Analyzed from 4901 words in the discussion.
Trending Topics
Discussion (105 Comments)Read Original on HackerNews
> I then tried using one of the AI tools to analyze my code in a project and a few other small tasks before it all came to an awkward halt. The system informed me that I had just run out of credits and I would need to provide a credit card to purchase more tokens I wanted to keep going.
> So you must believe me that the idea of paying a service in perpetuity so I could think just seemed so laughably absurd and horrific that I didn’t even bother giving them my card. I closed the laptop. I uninstalled the IDE and went back to using Emacs even.
I wholly support their personal choice. I am tired of articles from people who haven't used LLMs preaching about how it's all vibecoding, though.
Acting like LLM use is (EDIT: I meant is not) a spectrum between doing everything manually or handing control over to the LLM and vibecoding everything is a tired strawman argument.
This is still missing the point that LLM use isn't a binary choice between YOLO vibecoding or complete abstinence from LLM use.
I just don't know how many people have an overly negative opinion on AI assisted coding because they've just used the poor versions of these products given out for cheap/free. A similar critique is basing one's opinion on AI based on summary that Google provides for free in their search.
Like you said, the models available on free trials are usually toys compared to what developers use. Even Opus and GPT-5.5 are available on $20/month plans and you can buy a single month to try it out. The way they write about paying for a tool seeming "absurd and horrific" says it all about the level of actual research that went into their understanding. It's entirely based on news headlines.
Many many many of the problems with current gen ai will go away.
It's a fallacy because the reality is often not that…in fact sometimes the reality ends up worse. (See "enshittification", a process whereby technology gets worse over time, not better.)
But isn't the strawman here was that it wasn't a spectrum. That they couldn't just use it some, but all or nothing.
When I run out of tokens, I pay for extra. It doesn't feel good, but I do it because I didn't write the codebase - the drug dealer did. Just one more "fix" and the code should be good to ship. Oh no, out of tokens again? Just one more "fix", and another.
And the code that the AI writes is sprawling and almost incomprehensibly complicated. Overly complicated. It's like a tweaker wrote it, on methamphetamine.
I can make this comparison because many years ago I once had an ex that put methamphetamine (I didn't realize they had an addiction) in one of my vitamin capsules "as a joke", and I was up for 36 hours straight writing convoluted code, and then writing voluminous notes about the code I had yet to write. I had never done that drug before, or since (why they are an ex). I don't even drink. After that episode I re-read what I had written and it was quite scattershot.
And now I get the same exact feeling when using AI to write code, or have it write tickets, or plan out something, etc.
I use these tools daily, and it's like putting a drug dealer between me and the code. Sure it writes a lot more code than I could write without it, but at what cost? I really don't like where this is headed. And I don't think most software developers using AI realize what is happening.
.... are you saying that you can't just open up a file "written" by spicy autocomplete and add/change parts of it?
For code I wrote, it's typically faster for me to modify it than to get the AI to do it.
Managing agents is a lot like managing children. They will outsmart you 99% of the time. If your agentic environment isn't built for good sandboxing, you won't succeed.
If you build an environment that can represent mutual cooperation (e.g. helping an agent by doing the tool calls yourself if it was stuck) then it's soooo much better than just trying to rephrase the ruleset of the engagement.
Don't optimize for retries. Optimize for sandboxing and strong agent to agent communication that you can also observe, modify, and summarize.
- They're too expensive
- My buddy's 1995 Accord breaks down a lot
- Walking is healthier, plus you can stop and smell the roses
- I enjoy caring for my horse
- Sometimes you can get stuck in traffic
Fine if that's the way you want and can afford to live your life. But it is an exotic luxury belief. For those of us who are participating in the economy for real, the preference to not drive cars is not realistic.
If you can walk to your office and the temperature is always between 50 and 70 degrees F you would probably think cars are crazy, too.
Which, funnily enough, proves the point even further. Some people get so comfortable in their bubble that they become unable to even comprehend why other people make other choices in other situations.
So, the vast majority of people living in cities?
> the temperature is always between 50 and 70 degrees F
More like between -15°C and 35°C (though the upper range does depend on humidity).
I almost always walk to the office. The temperature range is a lot bigger (freezing in winter to uncomfortably hot in summer), and it's like 3 km, which many people wouldn't dream of walking. When my work was farther I used to cycle.
Most people can easily get to work without a car. Just depends on goals and motivations. Car is definitely the laziest way.
This is statistically very false.
It does a good job of proving my point that people within this bubble have a hard time understanding what the rest of the world is like.
EDIT: Oh, you are talking just about the US. Then your comparison doesn't make any sense because LLMs are available worldwide.
For those of us lucky enough to have the choice, the best bet is to sit it out for a year or so until it all comes crashing down, then re-engage with what's left of the software industry.
- They dangerous both to me as a driver, my passengers, and other road users, including pedestrians and bicyclists.
- They ruin cities which constantly have to accommodate ever increasing number of cars by destroying previously walkable neighborhoods to make room for roads and parking.
- They destroy our climate
- They are loud.
- Busses are nicer and I can read a book while riding the bus.
EDIT: To clarify on the public transit usage. The data is by-modal. Lower income levels are by far more likely to use road based public transit (such as busses), but high earners are more likely to live near a rail station and use rail based transit: https://cepr.org/voxeu/columns/public-transit-access-and-inc...
Fixed that for you.
* https://repo.autonoma.ca/repo/treetrek/tree/HEAD/render/rule... - syntax highlighting for 40 languages and file formats in ~10 minutes
* https://shufflenblues.com/expenses/ - real-time expenses progress updates with payment vendor API in ~30 minutes
* https://repo.autonoma.ca/repo/treetrek/tree/HEAD/git - real-time, cache-free raw Git reader implementation with cloning in ~5 days
* https://repo.autonoma.ca/repo/notanexus - PDFjs integration in ~3 days
However, these are likely not the "hard" problems you've mentioned. I feel like I can architect solutions at a higher-level now, without having to be completely caught up in many technical nuances. I'd rather not learn the extensive PDFjs API, for example, because it would take weeks of effort to understand.
Your syntax highlighting is very basic as well. Just ask LLMs to provide tests where it would fail to render correctly.
The first thing that comes to mind after looking at it: print("# not a comment")
Dependency-free, performance, FORTRAN, and it would take me more than ten minutes to find and integrate a highlighter that works across all of my code bases.
I searched for PHP-based Git libraries. All of them either invoked "git" using a system call or offered write abilities to the repo. I wanted a pure PHP solution that did not write to any files or invoke executable files (for security purposes). Maybe I didn't search long enough; at some point it becomes faster to tell the LLM what's wanted than to find a solution that fits.
> print("# not a comment")
Works correctly?
https://i.ibb.co/chgVkTz4/not-a-comment.png
And regarding the gym, sure, you might enjoy lifting dumbbells and solving puzzles to sharpen your brain. But that is not what engineers are hired for; they are hired to deliver a system using the best tools available. You can choose to farm by hand while the industry moves to using tractors, but sooner or later, you will be left behind.
And lastly, moving higher in abstraction allows us to tackle even more complex problems—I'd argue much more complex than the narrow puzzles we were facing before. Part of the resistance is simply an avoidance of facing higher-level complexity once the lower tier is automated.
I've been in the field for 20 years, and I do think the situation is analogous. We might not like it or we might deny it, but the fact is, LLMs do automate the mechanical part of thinking. Some people might not accept that, but that is the reality given my subjective experience and the experience of many others who are using the tool.
There are hard problems at every level of abstraction. TAGE predictor optimization up to handling data-center failover.
I don't really have a challenge for people like the OP, I get it. I too dragged my feet, even mourned the death of a type of work I had grown fond of. Then I got over it and realized I might prefer the romance of riding a horse into town, but I also like that there's semi-trucks delivering fresh produce to my grocery store year round. The leverage available right now is frankly insane. The one thing an "old dev" [as he self-labeled] can be sure of is that the younger generations will not share these hang-ups to the same degree and it's those people who will inherit the burden of maintaining and furthering the digital world.
virtually everyone building a startup uses these tools to automate a huge chunk of the work. you have no real chance otherwise.
my advice is to get on the cheap claude code sub and to check out the superpowers plugin.
the idea of the plugin is to help you create a “design” plan which promotoes to an “implementation” plan.
the former is a high level architecture decisions which YOU TAKE and use the agent to browse different approaches that might work.
the latter is the concrete steps the agent will take after you click execute. every file change documented, every migration required, etc. very easy to oversee it
just my 2 cents (as someone working on startups)
Huh? Do you mean people "building a startup" without any experience at all? The list of tasks in the comment you're replying to are well within the skillset of any software engineer that has that old job description classic "3 to 5 years experience".
Not to be rude, but if the code is the hard part for anyone, they never had a chance at the rest of it either.
I'll never understand why folks on here look at this tech in such black and white. Of course it has no concept of logic or understanding. But if you do, you can bring out whatever is in your head out into code at 10x the speed.
When building startups you need to move quickly because usually you have quite a bit of competition. If you think you can build a startup by being slow and thorough and take years to produce an MVP go for it :) i don't think you will succeed
I know its not rational, but it would be pretty darn terrible in my brain to pay for an IDE. Even more unimaginable to me to pay $100 a month for something...
All to say, "cheapskate"-ness from TFA really resonated with me, I don't see the sentiment around a lot.
The internet changed all that, because every net connected platform was built on platforms that cost nothing to learn or work in. I wrote my first HTML in freaking MS-DOS Edit. I learned JavaScript by copying code from View Source into an unregistered copy of BBEdit.
Do I have dev software I pay for? Of course! But not many (even fewer today than before everything switched to subscription pricing), and every one can be replaced with something that costs nothing.
The top thread on this post is people discrediting the author because they only used cheaper/free models, as if we should just accept and embrace the commoditization and enshitification of software development, even while it actively deskills us and creates codebases that cannot be maintained without the services that built them.
That's because AI wrote it to deter other AI from taking it's job
I mean, I get it, there's different kinds of people out there with different motivations, goals, spare time, etc.
But there's also a process of product design that I think the author is overlooking.
Lately I've been working and iterating on a number of DSLs, projects that might be a total waste of my time because they end up being poorly conceived or not very useful compared to a general purpose language!
I'm also working on a video game that is basically Magic: The Gathering meets StarCraft with Civilization style hex-grid conflict. It could be a total bust and entirely no fun to play (it's hard to tell if it's fun by itself because I enjoy working on the game while testing out the play patterns). It would suck to spend a couple of years on this if it's no good.
I very much enjoy the process of trying to figure out the best syntax and semantics for a new DSL or the process of iterating on gameplay elements when working on a game. The destination is also less important. I don't really expect anyone to use my DSLs or play my video games. I'm ultimately doing it for my own enjoyment.
Saying this, I am interested in the overall architecture and I've definitely learned from my mistakes, especially with creating DSLs. Like, having a TypeScript language server with a Rust runtime has some issues. It's kind of better to build the language server into the runtime so you're not maintaining multiple parsers, and depending on the language features, an additional pseudo-runtime in the language server.
Context engineering is allowing me to do things I've always wanted to do but don't have the time/energy. I'm writing in C++, assembly, Rust, Go. I'm fucking with boot loaders and all kinds of things.
It's brought me a far greater understanding of how cryptography, GPUs, CUDA, Apple Metal - all topics I have a vague interest in but have no time to work on.
The current raft of LLM models are genius children. It's like that 15 year old at college. But I have 30 years of experience and a genius child is pure power in my hands.
And it's a genius child that never gets tired. For a few hundred bucks a month I can have 3 geniuses working on my ideas through the night. Last night they wrote 20 different research theses on a topic and benchmarked them all. Then combined them into a best of breed algorithm better than anything that has been done before. It's an amazing world we live in.
I don't write this to throw mud at OP - they are entitled to their opinion. Merely to point out the contrast.
Since this blogsite has a .is domain I must assume they mean Egilsstaðir a lovely city with a population of around 2500 people.
I get the conversational aspect and value of it I just dont get people saying “I dont code anymore I manage agents” - besides obviously people selling AI
I can pick and choose which parts of the problem deserve my attention and which can be done by the LLM with me just keeping an eye on it while I mostly work on something else. I don't have metrics but I feel like I am doing higher leverage work with less friction.
Setting up the systems around the LLM itself is fun too. Hacking on harnesses and trying to improve the UX or the metrics is fun. Playing with different workflow topologies across agents is fun. Diving deep into context strategies, memory systems, prompting is fun. Trying to marry ideas from the past with what LLMs enable now is fun.
I don't see how this is soulless or unbearable but granted I'm not at a place that is demanding I maximize throughput. That would suck.
- pi-agent or code your own harness, it is not hard.
- codex for now allows you to use it in any harness but if they ever don't open models are catching up
- Reject low quality PRs the same as before. If management doesn't enable to you do that then that's a different problem.
most of the other things you mention feel tedious to me. i like diving into the code, understanding how to solve a problem, figuring out how to make the code structure look elegant and readable. find and comment on a clever short-cut, even if it means that i may not be clever enough to debug it later.
i don't get any of that with LLM generated code. i'd spend more time to clean up and fix the LLM code than i would writing it from scratch. to use LLM code efficiently, i'd have to give up all that. but i don't want to do that.
using LLM to code feels like gambling. every time i put in a prompt, its like rolling the dice. am i going to get a useful solution this time? and then reroll until i get a useful result rather than building up the application one step at a time.
I personally draw the line at plugins that try to set up entire workflows and take the human completely out of the loop. Those are next to useless imo for an engineer who knows what they are doing and are exactly how you end up with crappy code/products.
But to give my thoughts to your points I guess I just don't really care that I can't teach LLMs? It doesn't bother me because I do also still teach people, it's not one or the other.
On what you like about coding. I like that too, I still do it where I want to or where it is needed.
I agree with you on what parts are enjoyable but I guess I don't feel that I'm giving them up? I get to pick the problems I work on that way now. The only disagreement there is around 'clever' shortcuts. I get pleasure out of making things debuggable and traceable for humans.
I wish my odds at gambling were as high as they are with LLM generated code lol.
I do run into the whole 'this session was a waste, need to restart', but like once in a blue moon? Not nearly enough to turn me off from using LLMs daily.
On the teaching point again, my learnings around coding standards, architecture -> problem mapping, how to debug, are applied at the system prompt level and around a few key skill files, so when I say "implement ..." or "I'm seeing this behavior, where in the codebase is the most likely root cause? Why?" It does so close to how I would've done it.
I cannot speak for people who are using these things raw in the harnesses provided by the companies, or god forbid in the browser but you can definitely increase the odds of a good roll enough to be productive by changing the environment around the llm and to me that is the opposite of feeling helpless when it comes to LLMs.
I feel enabled to get more done, at my standards, on my time.
"woodwork is just procrastination that gets in the way of implementing your ideas."
We can find plenty of others, but my main point is that industrializing a process doesn't make it "procrastinating". There are plenty of jobs that are done by machines but are still practiced by humans for multiple reasons. If we think of coding as a means to create, then we have plenty of examples of good reasons to have both the industrialized process and the 'handmade' one.
"Thinking is just procrastination that gets in the way of your opinions"? ;-)
probably that nothing really matters, so I guess do whatever makes you happy. If programming makes you happy then you should do it.
I find programming Advent of Code fun so I do it, but I don't find writing yet another web interface fun, so I have Claude do it.