Ok, I am a bad programmer. But AI can write better code than me faster than me. So what’s the problem here? I really struggle to understand this mentality when it comes to using AI for coding (which I’d argue it’s actually good at!)

I’m sorry but there’s no world in which you can convince me this individual has coded anything meaningful in the last 5 years, or that they have used any AI dev tool in the last 6 months. The “deskilling” part is legit enraging me

Stuff like this makes me feel crazy. I'm a software engineer with 15 years of exp and Claude code is insanely good. I use it every day at work because my company pays for it after it was made clear how valuable it is. It isn't magic, it is a tool that I've learned to use well

Exactly!

It's pure cope, or bait. I've been coding for ~30 years, I'm working on a popular app/game that we started in the before-times. It is seeing insane levels of progress these days.

Wow that’s awesome!

the best analogy i can make is that it's like asking AI to translate something into a language you don't speak then saying it did a good job if you speak the language you can fix it, it saved time. if you can't, you have no idea you're calling someone's mother a slur

The only explanation I can think of for ppl like in this video is that they used free models instead of say, Fable or Sol/Astra

True

im unsure what AI model he's using thats "bad" at coding? temuAI?

Right?

You are a bad programmer so any code that AI produces looks good to you.

Not any code! But code that works 😊

I'm a programmer by trade. It writes better code than me, but it is nowhere near as good at engineering.

Makes sense

I mean, the person in the video is clearly a terrible programmer. However they are correct that, for professional software engineering, AI is not remotely as good as a competent human.

I don't use it to generate code, there are way smarter use-cases for LLMs as a programmer, for one when I need to look at some source code that has zero fucking documentation or comments, LLMs are remarkably useful for analyzing it and breaking down what mystery functions do.

I'm convinced that the people who think AI is bad at everything stopped using it when Will Smith was eating spaghetti with his ear. It's ironic that they're saying ask it something you know on a deep level when they demonstrate they don't know what it's capabilities actually are anymore.

AI is only as good as far as you can validate the output. You can't discern good programming if you are not familiar with how to be a good programmer

It completely depends on the size, complexity, and target architecture of your codebase. Python scripts for x86 Windows? Yeah it’s got it. A large codebase on some weird flavor of embedded linux? It’s gonna have a harder time getting things right

Yup makes sense

Relying on AI for code you can't understand just replaces slow bad code with fast bad code you can't fix. Where do you draw the line between tool and crutch?

That’s the Q at hand I think

The hubris on that guy! At this point, at *any* skill level you have to be obstinate to not be able to improve *some* aspect of your programming work w AI. Legends like Carmack & Linus are enthusiastic users ffs

if you couldn't write the code the AI generated yourself, there's a good chance you don't understand the AI generated code this is very bad when it comes to maintaining a large project over time. it leads to less code reuse and more technical debt

oh yeah ai can make some pretty dumb mistakes especially if you can go into deep stuff where there is little training data but tbh also nothing you can't really fix with ai

Back in the late 80s/early 90s in UK Food Culture it was common to sneer at @DeliaOnline & many TV cooks for not being "chefs" nor cooking "properly", then @jamieoliver & @Nigella_Lawson (&c) arrived and taught a generation to love *food* — and the critics faded. Same here.

Why not go through the effort to learn how code works and build a skillset? What’s wrong with proficiency, curiosity, and knowledge? At least you can check your work if you use Claude.

Writing toy scripts at home where you don't care about quality is quite a bit different than building production software that not only needs to work, but also needs to be robust, maintainable, simple and doesn't make your colleagues pull their hair out when they have to debug it

For sure! That’s why I think blanket takes like this are silly

The problem is that programming is engineering. If you're not good at it, then you are by definition not capable of knowing whether the code AI has generated is good enough. It's the narcissism of being unable to tell the difference between good to you and objectively good.

Guys, I went in with preconceived notions and, would you believe it, I saw exactly what I wanted to see. I used Gemini 3 point whatever flash, the best model in the world, and it couldn’t even center my div. What we’re seeing is that we’re just better than people who use AI. Like, just use your noodle poodles.

"you're outing yourself as a bad programmer" is wild lol. every professional software engineer is using AI to help them with their work

i admire how some people just come on to the internet and lie for fun

Professionals view the art of coding as a craft. Agentic coding takes much of that away. And the workflow that replaces it requires a lot of investment to get the same quality of outputs.

Listen to how he(?) phrases it, almost apologizing for using AI. He doesn't give specifics, because then he could be rebutted. He's pandering to his anti-AI audience for likes, not objectively assessing AI's capabilities. I did software engineering for one of the top tech companies in the country, went to a top CS program, and AI is easily, indisputably smarter, faster, and better than I am at basically all things technical. It makes some mistakes, sure, does some things suboptimally, but so do I. Ask yourself this: how did AIs "bad at coding" manage a sophisticated (if largely harmless) hack of HuggingFace?

Idk maybe we shouldn’t use AI due to the damage to the environment? Or even the amount of water it uses?

Just cause it’s better than you doesn’t mean it’s good… you’re never gunna get better if you’re relying on AI to do everything for you

It’s gonna always be better than me bc I fundamentally don’t have an attention to detail. But I think that’s why it’s fine to use in low/no stakes projects!

Attention to detail is a skill than can be practiced and improved upon. again, you're never gunna get better if you're relying on LLMs to do everything for you.

What *is* the problem? Aren't *you* the bloody AI journalist?

I'm a full stack developer and engineer. AI coding capabilities are absolutely amazing, but high level of thinking reasoning skills are lacking. Anyone claiming that "it is horrible" with broad strokes just have no clue what they are talking about. There is some wisdom needed to understand AI capabilities, what it is good at, and how to use it. There is also skill involved in producing prompts and creating a good work framework around AI models. You don't get this from speeding 5 minutes to test it, especially when you are coming with a HUGE bias against it. If you have a very large software and/or if it is high stakes, then architect the skeleton and high level design WITH the assistance of AI. For most applications, AI is better than the majority of coders at the implementation level, but it is also amazing at brainstorming architectural solutions and UX/design work. At work, my productivity has increased by AT LEAST 20x, and the code that I've been shipping has been more reliable at the same time. Anyone claiming AI is bad has a serious skill issue and/or is in denial because they feel threatened by it.

Many possible interpretations here, but I think what they are saying is that most of software engineering is maintenance. In the absence of human-like intelligence, LLM'd software will hit a maintainability wall. People won't realize this until it's too late.

Before llms I was completely unable to code. Post llms I'm now able to code. I used to watch coding tutorials ,while I was able to copy what I saw, they didn't really help with understanding. With llms I ask them questions and they explain things to me in a way I understand.

It's a productivity tool period

this description of the model feels representative of the state of ai assisted development maybe 2 years ago