Idk. Hoping there will be research soon.
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
It comes down to “adding code” that attempts to, or seems to, achieve something.
I used one such in the past, written in PHP by someone who was keen on proper typography.
They've equally filled the web and articles for the last years.
At least, AI images feel more on point.
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.)
Both are bad. Just use text.
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.
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?
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.
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.
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.
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 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.
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.
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
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.
> 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.
> 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.
Further, those abstractions that you just threw away to roll your own encryption were written by actually talented people who actually took the time to think about their implementation before shitting it out into the world. The implementations you'll find in libaries will be superior in every way to the trash your LLM will pump out.
So, TL;DR: congratulations, you've just announced to the world that you're proud that you're a bad coder and that the software you're being paid to produce is getting worse.
Totally unrelated video I watched recently: https://www.youtube.com/watch?v=7pqF90rstZQ
(for those with the kind of attention span that causes them to use LLMs, maybe just skip ahead to the "this is what you sound like" section, circa 17 mins)