← Back to blogYouTube Video

Published January 21, 2026

The Vibe Coding Era is here (and it's here to stay!)

Can't play the video or having issues? Here's the direct link.

Summary

The vibe coding era is here to stay, and any developer still writing every line of code by hand is going to get left behind. I break down why Karpathy's original 'vibe coding' advice — ignore the diffs, paste errors blindly, work around bugs — is terrible for real projects, and how elite engineers like DHH and Linus Torvalds now use agentic tools like Anti-Gravity and Claude Code to build real things (the Omarchy Linux distro, Torvalds' audio visualizer). My take is that this should be called supervised collaboration: treat the AI like a junior programmer you guide, architect the system yourself, and review every output. I explain why Opus 4.5, Gemini 3 and GPT 5.2 changed my mind, and how to start — pick any agentic tool, throw serious work at it, and when the output is bad, ask how to prompt better instead of writing AI off.

Key takeaways

  • Pick any agentic coding tool — Anti-Gravity, Cursor, Claude Code or OpenCode — and throw serious work at it with a latest model like Opus 4.5, Gemini 3 or GPT 5.2.
  • Never blindly accept diffs or paste errors without understanding them — the original 'vibe coding' advice is terrible for real projects.
  • Treat AI like a junior programmer: guide it, architect the system yourself, and review every output — that's what DHH calls supervised collaboration.
  • When the AI produces bad code, ask how you can guide or prompt it better instead of concluding AI is useless — that's the new skill to develop.
  • Tune out the Twitter hype but don't tune out the utility — the genie is out of the bottle and this isn't a bubble that's going to pop.
  • Expect AI to write most of the code in the near future; your job shifts from writing code to guiding, architecting and reviewing.
  • AI agents can now search docs, run shell commands and test code — let them do that work instead of doing it yourself.

Description

If y'all are still writing code by hand, you're wasting your time. I say this as a software engineer with 6 years working in this industry. Even I'm impressed with vibe coding now, especially with Opus 4.5. And it's not just me. People like Linus Torwalds, DHH, and others are also using AI to write so much more code. And in this video, I'm gonna convince you to dive into vibe coding just like I have.

Transcript

Oh boy, I thought we were still at least two or three years away from me making this video, but the hype is real and the vibe coding era is here, and it's here to stay. Now, if you're a developer who still is not using AI code and AI tools to speed up your workflow, you're gonna get left behind. I hate to say this because I'm also one who's always written every line of code manually in every project I've built. But the reality is these models have gotten way too good, and if you're not using them right now in your work, you're being naive. Simple as that. So, in this video, I'm going to give you my perspective on AI coding, not vibe coding. Again, vibe coding is just a click-baity title. What it should really be called is AI-assisted coding, and that's what I'm going to talk about in this video and give you my perspective on that as a real software engineer myself. And hopefully, by the end of this video, I will have convinced you to try out AI models in your own code because, as you're going to learn in this video, if you don't, if you're not using these models right now, you're being left behind. So, let's get started. I think the best way to get started with vibe coding is to first look at where it came from, which is this tweet from Andre Karpathy. I hope I pronounced his name right. He's the one who coined the term vibe coding and just came up with the whole concept. "There's a new kind of vibe coding I call vibe coding where you fully give in to the vibes, embrace exponentials and forget that the code even exists. This is a horrible piece of advice, by the way, but we'll move on. It's possible because the LLMs are getting too good. Also, I just talked to Composer with Super Whisper, so I barely even touched the keyboard. I asked for the dumbest things, like decrease the padding on the sidebar by half because I'm too lazy to find it. I accept all. I don't read the diffs anymore." This is again very horrible advice, and we're not gonna do that. But that's basically what vibe coding is. When I get error messages, I just copy-paste them in with no comment. No real software engineer is gonna do that, obviously. This is dumb, right? The code goes beyond my usual comprehension. I'd have to read it really for a while. Sometimes the LLMs can't fix a bug, so I just work around it or ask for random changes until it goes away. Yeah, this is, um, do I really need to tell you how bad advice this is? Like, do guys really need me to tell you how horrible this advice is? Anyway, that's vibe coding, and that is a horrible kind of coding for real projects. You do not just ignore errors and ignore the code that you're generating. You actually need to understand how it works. So, this is a horrible way to build projects with AI, right? I'm going to show you a better way, but before we do that, let's first look at a couple of really great elite software engineers who are actually using AI to speed up their workflow, and I think you're gonna like this. Here's a tweet from DHH. DHH is the creator of Ruby on Rails. I'm sure many of you have used that before, and also the Omarchy Linux distro. You probably know what that is, but if you don't, it's just a Linux distro that's really great, right? And here he says, "Make it work, AI. Make it fast, also AI. Make it beautiful, still human." Now, I'm starting to disagree with this because I can also make beautiful front-end UIs just with AI by just giving it a detailed style guide and asking it to follow those instructions and make the app look like that. So, I would call "Make It Beautiful" also AI. But the fact is, DHH, a very smart programmer, smarter than me, much smarter than me, also is using AI and also agrees that AI is going to speed up his workflow, and it is a valuable, fantastic tool to be using. Here's an article from DHH's blog where he's promoting AI agents. For those of you who don't know, this guy went on a Lex Fridman podcast and he basically just shit on AI coding and AI agents a lot over there, and yeah, he was not really fond of using AI for coding, right? But in this article, he promoting AI agents to do much more work for him. You can read the entire article over here, but the fact is even DHH is now using AI to write a lot of his code over here. And I really like how he defines it over here. It's not vibe coding, it's supervised collaboration because you're using the AI models to write the code, but you're also reviewing all of the code that they generate. You're treating them like a junior programmer that you're guiding, and you're reviewing all of their output. I much prefer using the term supervised collaboration over here instead of vibe coding. But the point is this guy is also using AI, and he's someone who did not like using AI to write code ever, right from the beginning. He was against it, but now the models have gotten so good that even DHH is on our side. And that's just DHH. I've got more for you. And this isn't all. Remember that Omarchy Linux distro that I was talking about that DHH built? Yeah, he used AI a lot to help him build it. Here's a video of that. So, listen to this. Games are mostly about making me better, making me smarter, making me faster at getting up to speed on new APIs, new techniques, about understanding more concepts because I asked the AI to explain it to me, about finding the correct leads on why a bug is the way it is. I think Umachi, for example, would not have existed without AI. I would not have had the patience to trawl through Linux forums to figure out how all the obscure error messages were to be decoded. No, no, no. AI was an enormous productivity boost by giving me a place to paste in those error messages and finding something better than a condescending Stack Overflow post that was three years out of date. The productivity. So what do you think about that? Right. And there was also this commit from Linus Torvalds. Most people know who that is, but if you don't, he's the creator of the Linux kernel. One of the smartest programmers that ever existed. Right? He is using Google Anti-Gravity, which is the same vibe coding, I don't even want to call it vibe coding, which is the same agentic coding IDE that I'm using. Torvalds is using the same thing to build his little audio visualizer tool thingy. And here he says, "Is this much better than I could have coded by hand? Sure is." I believe this is the project that he's talking about, Audio Noise, on his GitHub. If you want to check it out, you can go there. I'll put a link in the description. And as you can see, he says over here, "Also note that the Python visualizer tool has been basically written by white coding. I know more about analog filters, and that's not saying much, than I do about Python." So yeah. So you know what I think about this? When people like Linus Torvalds and DHH, really, really smart and accomplished software engineers who have previously spent a long time bashing AI and hating on it, are now starting to come around to it and using it in their workflow so much, so much that they're building entire projects and Linux distros out of it, I think that's something very interesting. I think there's something very interesting going on over there, and we should look into it. Right? Don't you think the same? So even after looking at all of those things, if you think that AI is not here to stay, my friend, I don't know what to tell you because, like, the models have gotten so good that even people like Linus Torvalds and DHH are using it. You don't have to listen to me. I'm a nobody, right? But these people have built entire operating systems by themselves. That is huge. And remember, I spent a long time hating on AI myself. The only reason I ever came around to vibe coding is because I tried out Opus 4, which is a Claude model. The moment I tried out Opus 4, I was just sold on AI and agentic coding as an idea. Before this, I did not like using AI to write code. I would only just go over to the websites, the Claude or ChatGPT website, tell them to generate some kind of code, and just copy that into my program. I would not use agentic tools. But ever since I tried out Opus 4.5, I've been blown away by how good it is. So if you were not using agentic coding tools before, I don't mind. I'm not going to judge you for that because I wasn't either. The models just were not that good, right? But if you're still not using agentic coding tools, especially when we have models like Opus 4.5 or Gemini 3 or GPT-5.2, if you're still not using these models to write your code and figuring out how to use these models to write good code and speed up your workflow, then you are going to get left behind. There's just no avoiding that. I keep saying that again and again, but it's the truth. You are going to get left behind. Now look, I get that. I get why you're not so interested in using AI to write code. A lot of it is just hype, right? A lot of it is just the Twitter people just hyping up the AI industry and these AI tools way too much, right? But the thing is, you don't need to listen to the hype, all right? Even without that hype, AI coding tools are still extremely useful. This is no longer just a bubble that you're waiting for it to pop. It's not gonna pop. It's not a bubble, trust me. Don't wait for the bubble to pop because you're gonna be waiting for a very long time, possibly the rest of your life, because this is here to stay, all right? The genie is out of the bottle, and he's not going back in again. If you are not a fan of the Twitter hype around this entire AI industry, tune it out, all right? Tune it out. The problem is most of you developers, you tune out the AI hype but also tune out the genuine value and utility of AI models. Don't do that. You gotta stop doing that. Tune out the hype. We don't need the hype anymore, but do stick with the utility and the usefulness of these AI coding tools. Here's how you get started. You literally just pick an agentic coding tool, like I use Google's Anti-Gravity. There's also other IDEs, like there's Cursor. You can also use terminal tools like Claude Code. A lot of people are enjoying Claude Code. You can use OpenCode. There's a ton of tools, right? Again, I'm personally using Anti-Gravity, but again, most people are using Cursor. There's a ton of people using Claude Code and OpenCode as well. It's really up to you. The point is not to pick the best tool over here. The point is to just pick an agentic coding tool and get your feet wet with using it. Literally just download one of these tools, pick a Claude or Gemini or whatever model you want to use, pick one of the latest models like Opus 4.5 or GPT-5.2, and start throwing some serious work at it, right? Start throwing some of your serious work at it. Test the limits of these models, see how capable they are, see how much you can use them to speed up your workflow. Anytime these models give you a bad output, your first, your immediate thought should not be like, look at this. Anytime these models give you a bad output, you should not think, "Oh, AI models just suck." That's not what you should think. What you should instead think is, "How can I guide the AI agent to write better code?" That's what your immediate thought should be, not that AI models are useless. They suck and I should not be using them at all. That's a dumb conclusion to drive. You should instead think, "How can I guide the AI to write better code? How can I prompt it better? What am I missing? What missing context did I not provide to the AI which led to this bad code?" This is a new skill that you need to develop. This is not vibe coding. This is agentic coding. And if you just always keep asking yourself this question, "How can I guide the AI agent to write better code?" every time you run into some sort of a roadblock or obstacle with using these models, you're going to go very far. You're going to learn how to use these tools much faster. And honestly, the way these models have improved so fast, I don't think you're even going to run into this issue very much because if you can just prompt halfway decent prompts, if you can just give the agent halfway decent prompts, it can do really well even with just that. Because one, the models have just gotten so much better than they used to be, and two, we've given them more tools. And when I say tools, I mean things like these things like being able to access your code base, things like being able to search the web for documentation, things like being able to test code, run shell commands. AI agents can now do all of these things. They can do all of the work that you throw at it. They can do all of the work that you were previously doing, and you just need to guide the agent to do this work for you. Like, just imagine, just imagine an AI agent that can literally search Google for documentation. Is that not crazy? Like, we as programmers used to spend hours doing this ourselves. If an agent can do that for us, is that not fucking amazing? So you need to be using these tools. They've just gotten so much better than they used to be. We're so far ahead from the Microsoft Copilot era that we started off from, and I don't even know, 2025, 2024, I don't even know, right? We're so far away from that, where the models and the tools are so much better than that. You need to be using these things because the reality is AI is going to write most of the code going forward. I don't think it's unfair to say that AI is going to write 90% of our code in the very near future. The models just keep getting better and better, and this is going to become a reality soon enough. And if AI is writing all of the code, what do we as programmers do then? We guide it. We're the human who's guiding the AI agent to write all the code. But actually writing the code manually, that is no longer our job. Our only job now is to guide the AI agent, architect the entire system, and review the outputs of the AI agent. And in that sense, we shouldn't be even calling this vibe coding anymore because that's a dumb term. Instead, what I much prefer is supervised collaboration, which is what DHH said in his blog post, right? I much prefer this term because it really captures the idea behind how we use AI agents instead of vibe coding, which is just dumb, when you let the agent take over your thinking, which is just dumb. Let's instead call it supervised collaboration, and let's start to use AI to write our code. This video is for all of the developers who have still not decided to use AI for your coding. I hope that after watching this video, I have changed your mind about these things and you are going to try these tools out. Look, nothing, there's no harm in trying, right? So download one of the tools that I mentioned over here and just try them out, right? Maybe you'll like it. Anyway, that's all for this video, guys. Thank you so much for watching, and like, comment, subscribe, share the video, and I'll see you in the next one. And of course, quick self-promotion plug for me. I've got my agency's website over here, and I build web and mobile apps for people and businesses. And if that's something you're interested in or if you know somebody who's interested in that, guys, refer them to me. Top link in the description. You can book a call with me, and we can talk about your project, and we can get started working on that. And if you like the work that I'm doing over here, you're going to enjoy working with me. So if you're interested in this and you have a project for me to build, let's talk. Top link in the description. Thank you for watching.

Share this article

Hassan Aziz pointing up

Dude, you’re at the bottom of my landing page.

Just book a call already if you’re that interested.

You scrolled all the way here.