Vibe coding is dead

Added: 2026-05-12

[00:00:00]Vibe coding is dead, I'm going to react to this video.

[00:00:04]Uh, I haven't watched it yet, but uh I have a funny feeling I understand what it's about. Let's keep going. I have to tell you a secret. While I enjoy the ready drama and the Twitter toxicity as much as anyone else, LinkedIn is by far my favorite app when I'm looking for real unhinged delusional content.

[00:00:16]>> Yeah. If you are not that familiar with it, LinkedIn is the place where the fake it till you make it business professionals come to share their pretentious opinions about absolutely everything. Yeah. But something strange happened in the past year or so. All your project managers, operations managers, accounting managers, task managers, and QA managers discovered ChatGPT. And they are using ChatGPT to generate low-effort articles on how ChatGPT is helping them 10x everything.

[00:00:36]Use AI. Absolutely. Use ChatGPT.

[00:00:39]You have ChatGPT, I love ChatGPT. I love ChatGPT is is fantastic. It's a not It's ChatGPT. Then they discovered Claude, and they generate more low-effort articles about how Claude is helping them 10x the ChatGPT 10x. These are the people who for the past year hyped up vibe coding up to the point where the Collins Dictionary had to declare it the word of the year in November.

[00:00:56]>> Word of the year. Somebody told me to stop using it. It's agentic AI. It's assisted AI. It's like, no, it's vibe coding. You don't know what the code's doing. You can look at it line by line.

[00:01:06]There's going to be a bunch of stuff in there that you don't understand. You can't fix. You don't know how it works.

[00:01:10]>> [laughter] >> Let's keep going. Vibe coding was supposed to change everything. It would finally deprecate those weird geeks with no social skills >> Yeah, deprecate those weird geeks like me.

[00:01:19]GO AHEAD, TRY.

[00:01:21]DO IT. HYPING JOBS. IT WOULD TURN all those managers into millionaire solopreneurs, and it would help companies cut human costs and increase their profits.

[00:01:28]>> Right.

[00:01:28]>> But now, just only a year after the idea of vibe coding was formalized, the very same people who were generating low-effort articles about how vibe coding would make them millionaire solopreneurs are now generating low-effort articles about how vibe coding is dead. Yeah. Don't get excited, though, because we are not yet abandoning all this nonsense. Okay.

[00:01:42]>> The next delusional iteration is to replace vibe coding with vibe engineering and agentic engineering.

[00:01:47]>> Sure.

[00:01:48]>> is that we need another 6 to 12 months to complete it.

[00:01:50]>> yeah, yeah, they keep saying this.

[00:01:52]6 to 12 months. They've been saying that for a few years now. Look, look, if they really had this disability, they would have already done it. They would have already done it. And we don't see it yet. We see half-assed stuff, some demos, crud apps. It's And people get complaining like, "Oh, I'm stuck. I got stuck now. I can't figure out how to fix this bug, and then I ran out of tokens, and then they they rate-limited me."

[00:02:17]Discard the vibe and agentic terminology, and the industry will settle on simply calling the process of building software engineering. Uh-huh.

[00:02:22]>> So, what is vibe and agentic engineering?

[00:02:24]>> Okay, here we It is, and I want you to brace yourself for this revelation, software engineering. It is just software engineering. Yeah. The kind that has been practiced since the 1960s by the weird geeks your operations manager was so excited to replace.

[00:02:34]Exactly. We had lots of basic applications. I I came in into the thing with basic. And there was a ton of apps, and they're all written by random people. And the the the the spaghetti logic and the lack of engineering practices and the inability to just understand what the code's doing, it was all there.

[00:02:53]It's just hacking it along. Bunch of That's where go-to was deprecated because it was using go-tos [laughter] all over the place and couldn't understand what the hell was going on.

[00:03:01]And that was the state of the art. It was like, "No, this is great, but power to the people." It's like, "Okay." Yeah, those those are not maintainable codebases. You can't fix them. You can't edit them. You might You might know how it works, but you give them a minute, you give it to somebody else, they'll be like, uh we got to start over, guys. So, let me walk you through the earth-shattering innovations of agentic engineering.

[00:03:19]First, you write a specification where you write down what the thing is supposed to do before you build it. Then you break the work into smaller, well-defined tasks. This is called task decomposition, and it is covered in the first 2 weeks of any computer science degree, and also in approximately every project management certification your manager already has framed on their wall. Then you thoroughly review the code before merging it. And if the review and prompt iteration are taking too long, you might as well write that piece of code yourself, just like in the old days when it was simply faster to fix a bug yourself than to delegate it to a less experienced colleague.

[00:03:46]>> That's what's up. Yes, we are basically slowly going back to building software like we did a couple of years ago, but we are having I mean, it's for proof of concept, for getting some of my sketching ideas out, getting some like overflow over general like concept, but if you're actually going to push the push it to production and you want to like fix it, update it, change it, modify it, and not get rate-limited, you're going to have to have a real engineer. it as a completely new thing.

[00:04:08]What actually happened here is straightforward. A lot of non-engineers got very excited that AI would let them skip the engineering This this is this is this is an exception. This is hype.

[00:04:18]Who paid you?

[00:04:20]Who's Who you working for?

[00:04:22]>> [laughter] >> part of software engineering. They tried it, and the software was bad. Now they are rebranding, doing it properly as a novel methodology, and giving it a new name so they can continue content about it. There is a lot of pushback in the software community when it comes to AI code generation, mostly because we all know this is not a magic bullet. And there is a huge difference between somebody with no technical knowledge vibe coding an app and an actual software All right. Very big difference.

[00:04:46]Huge difference. If you don't know how this stuff works, it's sort of like I don't even know what the I have comparison. You already know what developer generating some code with AI.

[00:04:53]The process might look the same on the surface, but the results will be vastly different. For the past year, a lot of people have tried to make it look like the outcome of vibe coding is the same regardless of the technical knowledge of the person doing the vibing. This is what got Who Who Who's paying them?

[00:05:07]Like with the SPLC stuff, like the the stuff coming out about that stuff, people doing all kinds of tricks behind the scenes.

[00:05:14]If it was really real, we'd be seeing new Facebook. We'd see a new We'd see a new Amazon pop up. We'd see all We'd see Amazon just use these tools internally to take over every other business. Come on.

[00:05:24]>> lot of non-technical people really excited, and a lot of technical people, me included, very frustrated. Another common LinkedIn article was generated around this idea of coding advancing from low-level assembly to higher-level languages and now to plain English.

[00:05:35]>> No.

[00:05:36]No.

[00:05:37]No, English is not good enough for that for doing for do for software development. No, it's not enough.

[00:05:44]There's too much tacit knowledge.

[00:05:45]There's too many details that have to be done correctly if you're going to provide value and to your to your customer your to provide value, you're going to have to do go down to the bottom of it.

[00:05:55]Except for basic crud apps, okay.

[00:05:58]So, if English can be used to build software, then everybody can do it, and we are all the protagonists in our own the social network movie.

[00:06:03]>> Did you like being nobody? Do you like being a joke? Do you want to go back to that?

[00:06:06]>> But what's funny is that plain English is now being quickly simplified to cut down on output tokens while maintaining full technical accuracy. Plugins like Caveman are extremely popular be- Caveman skill plugin code explain makes agent talk like Caveman cutting 75% of your code. [laughter] Caveman mode. Cuz it turns out that when you are paying per token, make button blue is more cost-effective than please kindly modify the visual appearance of the aforementioned interactive element to reflect the color of the ocean.

[00:06:32]>> Exactly. Just wait until these guys find out that it's even cheaper, easier, and probably even faster to simply open the IDE and add in the color blue CSS rule directly. But you got to know what that syntax is. And if we're colors, it's this it's minor. For the other CSS stuff, there's a lot going on there. And an AI assist will is helpful instead of looking up Stack Overflow and being abused by local websites, having to do therapy afterwards. Uh yes, this is Okay, but point taken. This is the ultimate irony of the vibe era. We were told AI would make software development more human, but instead it's making humans talk like Caveman just to keep the API bill under four figures.

[00:07:08][laughter] If you find yourself typing update oath, no fluff, just code, make no mistakes, you've just invented a really inefficient, expensive version of a programming language. So, here's my main issue with all this. Whether we like it or not, AI is here to stay. It will disrupt the industries, and it will displace some jobs. It has some very good use cases.

[00:07:23]>> yeah, yes, and it's already uh But they're being But it's being used as an excuse also to bring in the H-1B issue and all the offshoring, pushing this things offshore and keeping the prices the same, and and then getting all except for getting slopified. Why doesn't it work?

[00:07:38]Yeah, this is going to be the This is going to be a backlash of epic proportions of all the people that got fired that actually got capable of like competing directly against the the old companies they used to work at.

[00:07:49]>> And some decent use cases. Sure, the financials don't make any sense yet, and once we get over the customer acquisition phase, we'll have to pay much more for these services.

[00:07:56]>> Wow.

[00:07:57]>> But AI is happening, and we can see it across the board. However, this is not enough for some people. Having humans use AI as a tool while they keep their jobs is not cost-efficient, is not disrupting enough, and it doesn't get any real engagement. Boring.

[00:08:08]>> [screaming] >> So, we get TWEETS LIKE THIS ONE, [laughter] WHERE SOME ARMCHAIR expert knows for certain that the next Opus model is going to destroy all software companies.

[00:08:15]>> Right. They keep on saying that. It's been years now. They keep saying it. But then the next week comes, Opus 4.7 is released, and it immediately fails one of the most basic reasoning [laughter] tests.

[00:08:27]They keep on doing the rug pull. Like we Like Like in that old Peanuts cartoon.

[00:08:32]Peanuts cartoon.

[00:08:34]Like Lucy puts the the ball out, says, "No, no, I'm not going to take it away this time. You're really going to kick it, Charlie. You're going to kick it all the way INTO THE SPACE."

[00:08:42]>> [laughter] >> GOING TO KICK IT, BUT REALLY THE NEXT ONE, THE next one. And some of you will argue that this is cherry-picking, but then another day passes, and the overwhelming consensus on the Claude subreddit is that Opus 4.7 is a massive regression and a serious downgrade.

[00:08:56]>> had the 4.7 was because they wanted to to to save on compute costs. They're They're Cuz they're running out of money, and then the grift isn't working, and their their customer base isn't growing, and they haven't >> [laughter] >> This stuff is so crazy.

[00:09:09]>> from 4.6. Apparently, 4.7 is a massive >> still keep falling for it. It's like cuz they just want to keep the hope alive.

[00:09:15]This is going to be the next one.

[00:09:17]They're going to give me these superpowers that will make me take over the world. It's like, you really think that's going to happen? Come on, guys.

[00:09:22]Look, if they had these things, they would be taking over the world with them themselves instead of kind of selling them to you for a dollar a minute.

[00:09:29]>> [laughter] >> instructions, hallucinates, fabricates, web searches, invents code with random add-ons, adds responses with useless fillers, and burns through usage limits at an alarming rate for worse output. On the other hand, they managed to make it more human because 4.7 is constantly trying to call it a day or pick this up later.

[00:09:43]>> All right, we've laid the groundwork here. Let's circle BACK TO >> [laughter] >> THEY DIDN'T REALLY SAY THAT. THAT'S GREAT.

[00:09:49]>> There are only a few messages refusing to do the work. But that's all the the Yeah, but that's all the the training data, right? It's all the training data.

[00:09:55]It's all like, "Yep, okay, we're going to circle back and >> [laughter] >> Protect no. Industry speak, the corporate corporo speak.

[00:10:05]>> Which all my Jira tickets these days.

[00:10:06]So, by the looks of it, if I coding was the word of the year 2025, I expect plateauing to be the word of 2026.

[00:10:12]>> Plateauing, yes. We're at a plat- We're definitely at a plateau. And if when they bring you guys back in, when they when they rehire, you got to make sure it's like, "Hey, you They're going to promise you, "No, no, we're bringing you back. We need you. We need you to help of all the The moment they think that the signs are on the The writing's on the wall, guys.

[00:10:31]So, it's like this is the time to become entrepreneurial. Like the micro monopoly stuff, which I have a whole T-series on.

[00:10:38]This is the only answer to this stuff long-term. Short-term, you to get your skills, get the figure out what the businesses are and all that stuff, but like thinking that you're going to be being an employee with these kinds of characters in the the thing just You got to think about a way to get out of the entrepreneurial. I know, it's super scary and there's a lot of risk involved. I get it. I get it. But you got to find your niche. You got to find your slot. Use these tools to exploit that because the sooner as they get have a tool or as soon as they get they have a way to ship it offshore, sooner as they can bring an H1B into a to do your job for half the cost, they're going to. And seeing some people are so excited about writing Proven fact. coding plain English, here is your awesome trivia of the day.

[00:11:20]When Grace Hopper and her peers designed COBOL, which stands for Common Business Oriented Language, they planned for managers and non-programmers to be able to read and write code. Wrong. That was the dream. It's been the dream forever.

[00:11:30]With BASIC, they were saying anybody could do it. It's like anybody can do it, but should anybody do it?

[00:11:37]For proof of concept, great. For personal projects, awesome. But trying to put that stuff into production and have somebody else like on a team trying like figure out how the things are supposed to work together and all that stuff, you're going to have to have an engineer. You You have to have engineering practices cuz it just gets out of hand. It's not the creating of it. It's not the It's the understanding the code. It's the maintaining, sharing with other people, maintaining it, updating it, extending it, fixing bugs, finding out where bugs are at, monitoring the thing.

[00:12:06]That's the hard part. Not the generation of the code. That's That's kind of the easy part now.

[00:12:10]It's all this other stuff like compre- comprehending what the hell's going on.

[00:12:15]And using practices other people are familiar with and idioms other people not trying to roll your own thumb to roll your own roll your own roll roll your own whatever to sort out like algorithm is tight, dude.

[00:12:29]code in English-like syntax. And while COBOL became the backbone of global finance even to this day, it failed Hopper's specific goal of democratizing programming. Well, it did, but we found out that the programming is not that democratizable.

[00:12:43]For personal stuff, yes, but not for general purpose. It's just There's She took it She thought she could cuz she's wicked smart. She's not She's not dumb.

[00:12:51]And she thought everyone was at her level. It's like, "No, there's all this tacit knowledge that's just been baked in that you have to know about.

[00:12:58]Okay, let's keep going. coding non-technical managers to build complex systems, it just forced programmers to write extremely And made a huge mess that we can't fix, can't understand, can't extend, can't can't can't modify, can't figure out what's going wrong.

[00:13:10]verbose, rigid, and ultimately difficult to maintain code. What happened is that because COBOL was designed to be easy to read, early programmers tried to save memory by using two digits for years.

[00:13:19]This led to the Y2K bug, which cost >> That was not That was a different issue.

[00:13:24]They were trying to save memory cuz it was super expensive. Data would They thought they could get away with it. was billions of dollars >> to fix. And many COBOL programmers were brought back into the workforce because the business managers, who were supposed to be able to maintain the code, couldn't actually handle the technical debt.

[00:13:39]>> [laughter] >> That was a huge like Man, I lived through that. They There was They made such a big deal about the Y2K and it was just a whimper. But it was It was a big deal. There was a lot of problems and they They got lucky. They They were able to fix it in time. The conclusion is really straightforward. Vibe coding with non-technical people simply doesn't work. Right. As writing code becomes cheaper, everything that follows becomes more expensive. For proof of concept, for product market fit, to try and see if there's an a market someplace, great.

[00:14:07]Push to get it into production, real bad idea.

[00:14:10]>> Teams now deal with higher code churn, where large volumes of generated code are quickly rewritten, reverted, or patched. Developers increasingly act as reviewers, integrators, and debuggers, where the real cost lies in judgment rather than generation.

[00:14:20]>> Correct. It's always where it's been.

[00:14:22]It's always been the value is in the is in the is not in the generation. That's not the That's not the issue. We already figured out with the mythical man-month.

[00:14:30]That book made it very clear. Using lines of code as a metric is a bad metric. So, less time goes into Because people will game the metric. Generating the thing is not the metric.

[00:14:41]Fixing it, understanding what the hell's going on, up the main- maintaining it. Let's keep >> typing. While more effort is required for validation, debugging, and long-term maintenance, and these are the areas where technical skills really matter. If you like this video, you should consider joining our community where I'm posting Okay, you're going to join join his community, but uh huh.

[00:14:59]All right, that was a good rant. I feel better.

[00:15:02]Give me a like and subscribe. I'm going to leave the daily dose so you can give him a like and subscribe and I'll rant at you later.