The author likes writing code by hand, and yet still uses LLMs? Am I alone in being the only one who doesn't like LLMs and at the same time, not use LLMs? Or am I the one taking crazy pills?
You're not alone. I'm puzzled by the rabid uptake of LLMs, given their semi-shitty output. Developers are all supposed to ride herd on LLMs, correcting and guiding their output, but mainly fixing it. If there's any aspect of a programming job that sucks, it's porting or fixing someone else's bullshit code. And that's what the current zeitgeist wants. I simply do not understand the enthusiasm.
The goal on the part of the C-levels pushing this stuff seems to be mostly to depress programmer wages and lower their bargaining power by reducing them to deskilled, interchangeable middle-managers. It doesn't need to work as advertised to achieve these goals, as long as it shapes widespread perception.
Plenty of useful idiots are willing to ignore the long term consequences for their industry- not to mention plenty of other negative consequences for the web, society, and the planet- for short-term gains and personal convenience. No surprise that quite a few previously respectable developers who have jumped on this bandwagon are at stages in their career where they can expect not to experience any of the direct harms caused by their choices. Classic ladder-kicking.
I've recently started using LLMs on a limited basis. The biggest win I get from using them is to get me un-stuck when I've run into a roadblock. Admittedly this happens more with hobby projects where I like to use less popular languages/libs that I'm not fully familiar with.
For example, recently I've been toying with MRuby + Crystal and had a blocker involving casting Ruby data types to Crystal. I worked on it off and on over the course of a month (reading a lot of code and docs) and made little progress. The LLM was able to solve it and a few other issues within a couple of hours. The code it produced was heavily edited afterwards, the value was the discovery of the correct API calls that I had missed.
Another nice use I've found is for generating FFI bindings for Crystal. CrystalLib already exists for this purpose, but it's somewhat limited. The LLM fills in the gaps pretty nicely, and it saved me some time.
I've read plenty of others posting on how they write most of their code via LLM, and I don't see how that would be useful for myself. It takes Claude quite a while to formulate a response, more time than it takes to just type the response by hand. That should change eventually but it could be years before it's fast enough for me. I have yet to try the whole "run several instances of a LLM in parallel and a bunch of things then review the results", but that seems like a lot of context switching, and potentially not worthwhile. Typically I find coding to be the least difficult and least time consuming part of my job.
Overall an LLM is just another tool, one that will probably provide more value over time as its refined.
I'm sure everyone is tired of hearing from me on this topic by now, but you surely are not alone.
I think it's a bit rare not to have a greedy employer trying to force your choice.
I've also been working on the same project continuously for five years. At that point LLMs offer you less than nothing at all. You have enough time and energy to write the code. I've coded the same system by hand not just once but 10 or 15 times.
Over that time scale the real challenge is learning: using your own past mistakes as fuel for future growth. Only a human can harness this rocketship growth strategy because only we, by learning, grow in such a way that we can keep growing and growing, each time surpassing our previous best heights and continuously breaking free of our previous most constricting limitations.
You're definitely not alone.
I still foundamentally dislike LLMs and agentic coding, and I still don't use any of this, despite everyone pushing for me to adopt it — at work, a rising amount of people on my open source projects, the damn twitter.
I sometimes throw one of my mixed Rust/Python OSS projects at them, asking to implement a feature, mainly to check with the status of those models, and to laugh at the results.
I am happier when I bike than when I drive, but I still drive on long distances.
Most of the time writing code is just a way to reach a goal and it makes sense to pick the tool that gets you there the fastest, even if it is less pleasurable.
It also makes sense to sometimes do the pleasurable thing, especially since it keeps your brain fit like biking does your body.
While I might agree in a broader sense, I don't really in the specifics of the context.
First, I consider the whole thing as a marathon rathe than a sprint. If I get faster today, but tremendously slower in a month from now when I need to retouch the same code, then the whole productivity boost isn't there. I build things in future perspective, not to close the ticket at the end of the day.
Secondly, I picked the job because of the pleasure. If I have to spend the vast majority of the time of my life being miserable, I'd just pick a different job/career/whatever. There's no reason here that can convince me I have to be less happy doing my job.
spc476 | 23 hours ago
The author likes writing code by hand, and yet still uses LLMs? Am I alone in being the only one who doesn't like LLMs and at the same time, not use LLMs? Or am I the one taking crazy pills?
David_Gerard | 21 hours ago
it's critihype, e.g.
this is yet another promotion for AI coding bots slightly pretending not to be one.
bediger4000 | 20 hours ago
You're not alone. I'm puzzled by the rabid uptake of LLMs, given their semi-shitty output. Developers are all supposed to ride herd on LLMs, correcting and guiding their output, but mainly fixing it. If there's any aspect of a programming job that sucks, it's porting or fixing someone else's bullshit code. And that's what the current zeitgeist wants. I simply do not understand the enthusiasm.
Internet_Janitor | 20 hours ago
The goal on the part of the C-levels pushing this stuff seems to be mostly to depress programmer wages and lower their bargaining power by reducing them to deskilled, interchangeable middle-managers. It doesn't need to work as advertised to achieve these goals, as long as it shapes widespread perception.
Plenty of useful idiots are willing to ignore the long term consequences for their industry- not to mention plenty of other negative consequences for the web, society, and the planet- for short-term gains and personal convenience. No surprise that quite a few previously respectable developers who have jumped on this bandwagon are at stages in their career where they can expect not to experience any of the direct harms caused by their choices. Classic ladder-kicking.
k749gtnc9l3w | 13 hours ago
Debugging alien logic is fun, typing boilerplate is boring.
wmoxam | 7 hours ago
I've recently started using LLMs on a limited basis. The biggest win I get from using them is to get me un-stuck when I've run into a roadblock. Admittedly this happens more with hobby projects where I like to use less popular languages/libs that I'm not fully familiar with.
For example, recently I've been toying with MRuby + Crystal and had a blocker involving casting Ruby data types to Crystal. I worked on it off and on over the course of a month (reading a lot of code and docs) and made little progress. The LLM was able to solve it and a few other issues within a couple of hours. The code it produced was heavily edited afterwards, the value was the discovery of the correct API calls that I had missed.
Another nice use I've found is for generating FFI bindings for Crystal. CrystalLib already exists for this purpose, but it's somewhat limited. The LLM fills in the gaps pretty nicely, and it saved me some time.
I've read plenty of others posting on how they write most of their code via LLM, and I don't see how that would be useful for myself. It takes Claude quite a while to formulate a response, more time than it takes to just type the response by hand. That should change eventually but it could be years before it's fast enough for me. I have yet to try the whole "run several instances of a LLM in parallel and a bunch of things then review the results", but that seems like a lot of context switching, and potentially not worthwhile. Typically I find coding to be the least difficult and least time consuming part of my job.
Overall an LLM is just another tool, one that will probably provide more value over time as its refined.
conartist6 | 9 hours ago
I'm sure everyone is tired of hearing from me on this topic by now, but you surely are not alone.
I think it's a bit rare not to have a greedy employer trying to force your choice.
I've also been working on the same project continuously for five years. At that point LLMs offer you less than nothing at all. You have enough time and energy to write the code. I've coded the same system by hand not just once but 10 or 15 times.
Over that time scale the real challenge is learning: using your own past mistakes as fuel for future growth. Only a human can harness this rocketship growth strategy because only we, by learning, grow in such a way that we can keep growing and growing, each time surpassing our previous best heights and continuously breaking free of our previous most constricting limitations.
gi0baro | 13 hours ago
You're definitely not alone. I still foundamentally dislike LLMs and agentic coding, and I still don't use any of this, despite everyone pushing for me to adopt it — at work, a rising amount of people on my open source projects, the damn twitter. I sometimes throw one of my mixed Rust/Python OSS projects at them, asking to implement a feature, mainly to check with the status of those models, and to laugh at the results.
tentacloids | 12 hours ago
Let me guess, you don't drunk drive to work either?
dsh | 4 hours ago
Wow there is literally a dril post for everything, isn't there?
catwell | 13 hours ago
I am happier when I bike than when I drive, but I still drive on long distances.
Most of the time writing code is just a way to reach a goal and it makes sense to pick the tool that gets you there the fastest, even if it is less pleasurable.
It also makes sense to sometimes do the pleasurable thing, especially since it keeps your brain fit like biking does your body.
gi0baro | 12 hours ago
While I might agree in a broader sense, I don't really in the specifics of the context.
First, I consider the whole thing as a marathon rathe than a sprint. If I get faster today, but tremendously slower in a month from now when I need to retouch the same code, then the whole productivity boost isn't there. I build things in future perspective, not to close the ticket at the end of the day.
Secondly, I picked the job because of the pleasure. If I have to spend the vast majority of the time of my life being miserable, I'd just pick a different job/career/whatever. There's no reason here that can convince me I have to be less happy doing my job.