I don't write code anymore – I sculpt it
jerpint.io41 points by jerpint 7 hours ago
41 points by jerpint 7 hours ago
Why is it that AI glazing blog posts like this inevitably use a shitty AI generated image? I am absolutely sick to bits of seeing them
The text reads suspiciously like AI too. I don't X -- I Y. Short paragraphs, lots of em-dashes.
It does look like it passed through an LLM, but you're wrong about em-dashes. There are no em-dashes in it, just hyphens.
The title alone was enough for me to suspect exactly what I'd see. A post praising AI, written by AI.
i was always curious how these em dashes were written? like in my keyboard i dont see the symbol, is there any shortcut?
On macOS, option plus hyphen gives an em-dash. Also if you're on macOS you should try some other keyboard keys too with option.
Some markdown-like static blog generators just replace -- with emdash.
I used one such in the past, written in PHP by someone who was keen on proper typography.
Why are you sick of AI images, but you're not about stock images ?
They've equally filled the web and articles for the last years.
At least, AI images feel more on point.
It’s an indicator of extremely low effort. I don’t think fundamentally it’s people not liking the picture, or liking it less than a stock picture (even if the all look the same) it’s just a tell that someone is being super lazy. Same as people don’t like seeing AI posts on forums like HN. It’s less a judgement about the content of the post and more about the way the poster is interacting. Everyone can use AI themselves if they wanted to see throwaway AI output so it feels condescending when it’s presented to us.
Maybe because OP likes images that were actually taken and stock images are fine that way. Different people can have different preferences.
My problem with stock photos is that unless it is clearly a stock image, it can mislead. For example, when looking up information on the solar panels that Jimmy Carter had installed on the White House, the article had a stock photo of a modern photo-voltaic panel. So I was initially confused, as what was described was a solar collector that concentrates the sun's heat to provide hot water for the kitchen. Which makes a lot more sense of why it wasn't re-installed during some roof renovations under Regan (the initial purpose of it was more to inspire the nation, that wouldn't be served much by re-installing it after it had been removed for roof work as the original moment had passed).
I don't like AI images or stock images. If a picture is not needed (or helpful; sometimes it is not needed but can be helpful) to describe it which is specific to that article, add it, but stock images and AI images are not helpful, and can sometimes be deceptive in some contexts (although so can pictures specific for the article, if the pictures are badly made).
Often, the text will be good enough, or better (since then you do not need to download the picture, it does not take up space on the screen (or on a paper if printed out),e tc.)
They're both pointless filler, but because stock photos are from the real world, they don't suffer from "detail collapse" - sure, there's not much meaning to be found, but the detail stays coherent as you keep looking at them for longer so your brain quickly realizes they can be safely skimmed.
Your visual system looks for detail, it's consistent with your world model but empty of meaning, it doesn't trigger your attention network, you can ignore it and keep going.
AI images do suffer from detail collapse - the longer you look at them, the less sense it makes. Look at the image in the linked post - the sculpture makes no sense, the code inscribed on it looks like real code - oh wait no it's not, the characters quickly blur into garbled scratches, and oh ? there's a smaller block of even more "meaningless" code nested into the first?
Your visual system looks for detail, it's inconsistent with your world model, so it triggers your attention network in search of the meaning behind it (things that are not consistent with our internal world model are of utmost importance to our attention), but there's none to be found. You've just wasted mental cycles on slop, whether you consciously realize it or not.
In either case shitty MSPaint drawings would be much better as they would express personality, and there'd only be the level of detail that their author judges useful, but that would require actual effort.
> Why are you sick of AI images, but you're not about stock images ?
Both are bad. Just use text.
AI images tend to have uncanny-valley details about them that make your brain go "brotha, eugh!" in a very subtle way, perhaps just below the threshold of conscious awareness. Stock photography is not nearly so unsettling, so your brain just skims over it.
Honest question. For the purpose of adding a throw-away visual to help set the tone of an article, is there really any difference between shitty AI slop vs shitty Photoshop slop? Either way it is obvious that this isn't a real studio photo, and that wouldn't be expected. I guess the better (or more traditional) article picture would be a hand illustrated piece that an in-house artist would throw together in maybe an hour or two, depending on the detail and desired quality.
I think you're better off with no image if the only option is slop. It detracts from the content
I call it “AI blindness”, people genuinely don’t see what’s bad about AI output, it happens across code, text, and images. Same as the title itself being X not Y AI slop.
I love this! Just sculpting the wrinkles right out of my brains!
I know the feeling. I still wonder though, am I faster? Do I understand what I’ve made as well as I used to? What have I learned? How did the experience benefit me? What value did I gain?
Idk. Hoping there will be research soon.
Do your own n=1 research! Unfortunately the only companies that would fund this would be AI labs. They aren't going to report that it sucks
One cool way I saw someone do it was to estimate a task and then randomize whether they'd use AI or hand code it
Is this post "sculpted" too? It certainly looks suspiciously so. It feels disrespectful to expect people to read something you could not be bothered to write yourself.
I wrote something very similar back in October! The "sculpting" metaphor really stuck for me.
> I would never have written my own sorting algorithm to sort a list in the past. I would instead rely on abstractions left for me by those with more experience.
And by doing so, never gain experience of your own. It is a truly alien mindset to me to take pride in never understanding what code does, to be comfortable relying entirely on magic words provided to you by others. For me, the simple existence of a bug in the magic word wasting hours of my time, a failure that is not my fault but a failure for which I bear the negative consequences of, is infuriating enough that I am always compelled to have more understanding and eventually more control of everything I write.
People get really defensive when I point out that they are proud of the things that they don’t know. Willful ignorance and anti-intellectualism is a cancer in anything it infiltrates and SWE is no different.
There was a snarky comment earlier saying, try implementing something that's too hard for you, THEN try going back to writing code by hand - implying agents are indispensable.
And I thought to myself that the most memorable parts of my career, the most rewarding parts, were when I took some task that was too hard for me, then, with great effort, became a developer for whom that task was no longer too hard for me.
All the times I told someone I could do something I didn't know how to do but believed was possible then had to figure it out filled me with existential dread. There was a good feeling after I succeeded, but I get a good feeling from just building something big and functional with AI, as long as I feel like I was the driving force behind the ideas in the thing, so maybe I don't need the stress?
Why do you sound so petty? Do you understand every single layer of the stack below you? Do you understand how your code gets translated down to C, assembly, machine code? How all that becomes electrical signals on a PCB? How the material properties work at the physical level and how its manufactured? Because if you don't, aren't you, myself and basically everyone relying on the 'magic' words of others?
If you're trying to accomplish X, how much does all this matter directly related to X? There's always infinite number of things to learn about and depth in every dimension. We explore depth where we are each personally curious or where its necessary.
I liked the post. It was short and concise and presents a unique way to view this new coding paradigm.
"If I have seen further than others, it is only because I was standing on the shoulders of giants," Isaac Newton
I dont think parent comment was being petty. Not understanding what agents write, especially in larger and/or more complex code bases is a valid thing to be concerned about. It's not really about understanding assembly or machine code, and I think more about really understanding your code and how things fit together. That way when something goes wrong you have an idea of how to fix it.
"I dont write code anymore I just review what Claude Code writes" is, unfortunately, not a unique way to view coding with agents. About a million people have discussed it like this in the past year or so.
I don't understand every single layer of the stack below me, but I consider it a weakness, not a strength.
The comparison is false anyway, even if one just understands a high level language, there is no magic, you’re working in a closed system where you can understand and explain what your code is doing. There may be edge case, performance things, etc where understanding at a lower level is important but you can work completely in the abstraction.
Vibe coding, certainly as an “abstraction”, is nothing like that, you’re hoping you can get across the behavior you want, and have no visibility into what choices are actually being made with regard to how it’s specified. It’s the same as being a manager that can’t code, you’re entirely at the mercy of the people doing the actual work. That’s not true when you code without knowing semiconductor physics.
I suspect Newton is rolling in his grave at your words. Standing on the shoulders of giants also means respecting the giants. That means dedicating time to learning what they have to teach. Not treating them as a black box with no credit or consent, in the name of your own "glory".
> Do you understand how your code gets translated down to C, assembly, machine code? How all that becomes electrical signals on a PCB? How the material properties work at the physical level and how its manufactured?
Yes, yes, yes, yes, and not really but I would absolutely learn about it if I had any ability to control it. As it is, knowing more about material properties would not allow me to improve my programs, and I do not have a hardware foundry so it would not allow me to improve the hardware either. If I happened upon 1 trillion dollars or so, though, I would certainly invest time in learning all of the technical details of hardware manufacturing with the goal of creating a better hardware stack than the ones that we currently have available to us.
I didn't understand nearly as much from the start of my programming journey, of course. But as others mentioned, that was a weakness, not a strength. The lack of understanding bothered me, and I was able to learn all of the things that I did because I was continuously pushing myself to learn and understand more. If I had settled for never trying to understand, to just accept my current knowledge as the limits of my knowledge forevermore, then obviously I never would have been able to. Over months and years, I was able to continuously expand my knowledge little by little until eventually knowing a lot, rather than stagnating in the same place forever. There are still things I don't understand, of course, and my learning journey continues even today.
> Do you understand how your code gets translated down to C, assembly, machine code?
> How all that becomes electrical signals on a PCB? How the material properties work at the physical level and how its manufactured?
> "If I have seen further than others, it is only because I was standing on the shoulders of giants," Isaac Newton
There is difference in taking pride of not knowing things and considering it as regrettable necessity.
Did the word "Refactoring" become uncool when Design Patterns did the same?
I don’t understand the kind of mentality that takes to put out genuine slop like this. Use the LLM or the agents or the fancy way to call a markdown file a skill to do your job; but stop pretending that it makes you in any way exceptional.
The whole post says nothing, seems AI generated itself and on top of that, adds nothing of value - simply exists to increase the entropy in the world.
People who were very bad 0.01x devs (or not devs at all) that now feel competent because they can produce something when previously they could produce nothing.
I, on the other hand strongly resonate with this post. I am not a programmer by profession, but a programmer by need - I write tools to get stuff done, and this is exactly the missing link. I am able to craft algorithms, architectures and focus on the UI, and llms (mostly Claude) allow me to write the low-level that needs to be written that I inspect. What used to take days takes hours and allows us to solve issue we were previously not able to in our org.
Theres a lot more AI hype vendors than AI produced software anyone is actually using it seems.
This piece starts off making it sound like the computer is pretty much doing all the work, while the human maybe weighs in on a matter of taste once in a while, if they like, but by the end, the list of what the LLM can actually do is really short. Implementing a sorting algorithm for you, perhaps, but not necessarily one without “egregious flaws,” and really you should still use a library for that. Replacing high-quality libraries of mature software, that have tests, etc, is obviously one of the poorer uses of vibe-slop coding.
It comes down to “adding code” that attempts to, or seems to, achieve something.
I feel most like I'm sculpting code when I'm working in Lisp. I can sort of feel the shape and contour of the procedure I'm creating as I work. Even when writing in something like C, I feel the feeling of "filling out a tax form". Making sure all the things in Section 1 boxes A through E are correct, because Section 6 box D depends on them being so.
Coding with AI assistance feels like what it is: outsourcing. Letting an accountant take care of the tax forms, except the accountant is Wheatley from Portal 2: a chatty, subtly below-the-threshold-of-competence robot.
While I experience a very disturbing nausea, with just a healthy bit of existential dread, whenever I read whatever these AI Bros(TM) manage to... ahem, "sculpt", you actually managed to write something not immediately recognizeable as slop. Kudos.
Points deducted for the wholly unnecessary image. Text can, after all, stand on its own.
You are, however, strongly influenced by the writings of the tool covered by your musings. Might I suggest to perhaps not indulge in such excessive hyphenation?
ai slop justifications will become the norm.
> I think about what we are trying to implement, and if this iteration brings us closer to it.
> Where things go, how pieces fit, reusable patterns - this is more question of subjective taste and big-picture thinking.
I do that, too, only I call it coding, and it doesn’t require me to rewrite a bunch of badly written slop first.