Auto-generated transcript Before we start this video, I do feel very sick, so if I sound weird throughout this video, just bear with me. I'm, uh, I've got a little cold and a fever to deal with, but hopefully that doesn't bother most of you. Now, I want to talk about this tweet from this guy named Adam Wolf on Twitter, or X, whatever you want to call it. So this guy works at Cloud Code at Anthropic AI, and he tweeted all of this, which went really viral. which is why we're going to talk about it. So, this new model is something else. Since Sonic 4.5, I've been tracking how long I can get the agent to work autonomously. With Opus 4.5, this is starting to routinely stretch to 20 or 30 minutes. When I come back, the task is often done simply and idiomatically. Now, I don't deny that AI models are getting better, but this next tweet is where I really start to disagree with this guy. I believe this new model in Cloud Code is a glimpse of the future we're hurtling towards, maybe as soon as the first half of next year software engineering is done that is a very huge statement to make all right you don't just say stuff like that unless you've literally got something like i don't know agi or something in your data centers and i don't think entropic is even close to that soon we won't bother to check generated code for the same reasons we don't check compiler output this is such a dumb statement to make i can't believe someone working at entropic is making such a dumb statement. Saying something like this just makes me feel like you don't even know how to code, honestly. Because first of all, there are plenty of different cases where we actually do read compiler output. Like, I don't know if you've ever actually built a program before, Mr. Adam, but we do actually read compiler output in a lot of cases because there's a lot of useful information in there. So either the people at Anthropic aren't exactly all that good at coding, or they just say dumb stuff like this just for marketing, just for hyping up their products to people who have no idea how technology works. And maybe that works for them. Maybe that kind of marketing actually gets some results for them. But this is just a completely dumb statement to make. And I'll tell you why. I'll tell you why this is such a dumb statement to make. LLMs and AI tools are always non-deterministic, whereas compilers and programming languages are deterministic. There's a huge difference between a compiler's output and the output of an LLM. So if you're gonna say something like, we won't bother to check generated code for the same reasons we don't check compiler output, you are making a false statement over here. The output generated by an LLM is going to be random. It's going to be something you can't really predict. You can write the best prompt in the world, and even then the AI is going to give you a different output for that same prompt. You can test this out yourself. go over to chat gpt give it some prompt to generate some sort of output or response and it will give you a different response every single time it may sound like the same response but the words are going to be different the language is going to be different and so on and so forth right so llms and ai tools inherently have a randomness factor in them their output of these tools is always going to be random now we can still use it obviously Like for example if I tell AI to generate a function a code to calculate the hash of a particular file. If I say that, it's going to give me the correct code nine times out of ten. But that code will always be slightly different each time I ask it for it, right? And that makes LLMs and AI tools inherently unreliable. They literally are unreliable by design. Not by a flaw, but by design. Whereas compilers and programming languages are deterministic. Computers are deterministic. That means that no matter what happens, if I, let me move this down a bit, if I say something like, I don't know, print hello world, and then I run this code, it's always going to print hello world on my screen, on my terminal, right? It's never going to print anything else. It's never going to miss a character. It's never going to crash. It's never going to do any of those things, right? But if I say the same thing to AI, there's always a chance that it's going to give me a completely different output. And that's just how it works. That's just how LLMs and AI have been built. And so yeah, sure, maybe you can ignore compiler output each time you compile your application. If there's no errors, you can probably ignore your compiler output. But that's because you know that the compiler is going to do the exact same thing every single time you run it. You do not have that guarantee when you're using an LLM. So LLMs are just completely unreliable 100% and that's just how it is. So yeah sure maybe you can use LLMs and the AI generated code in your side projects, in your personal projects that nobody's gonna ever see and maybe you can just blindly trust the LLMs output in those cases but for real production grade apps you cannot just ask AI to write all the code and you don't even know what it does. you don't even try to review it and clean it up and all that. We're just never going to live in a world where that is possible. Not unless you fundamentally build something completely different from LLMs because LLMs are non-deterministic. They have a randomness factor that makes them completely unreliable in 100% of scenarios. There's always going to be a chance that the LLM is going to fuck it up and in order to make sure that your app works in production, you're going to have to review and fix all of the AI generated code. Now this isn't the first time Anthropix CEO has made wild and completely baseless predictions. This is a tweet from, what is it, March 11, 2025. And currently it's November. So how many months is that? Let's see, April, May, June, July, August, September, October, and then November. We're almost at the end of November. That is eight months. he says that in the next three to six months, AI is writing 90% of the code. And in 12 months, nearly all code may be generated by AI. Well, it's been eight months and I don't see AI generating 90% of the code, not even close. Let's see what he said. But now getting to kind of the job side of this, I do have a fair amount of concern about this. On one hand, I think comparative advantage is a very powerful tool. If I look at coding, programming, which is one area where AI is making the most progress what we are finding is we are not far from the world I think we be there in three to six months where AI is writing 90 of the code And then in 12 months we may be in a world where AI is writing essentially all of the code. But now getting to kind of- Right, so as you heard, AI is supposed to be writing 90% of the code in six months. Well, it's been eight months. It hasn't written not even close to 90% of the code. There are no major successful apps that have been completely vibe coded or AI coded. It's just not working. So yeah, either the CEO of Entropic doesn't even know how AI and coding and programming even works, or he just says these things just for marketing. It could be either of them, to be honest. Now, I do agree that AI does help to make coding a bit faster. There's definitely cases where you can just, instead of writing the code yourself, you can give AI the right prompt and it will generate a pretty decent and pretty usable bunch of code that you can just integrate into your application. There's definitely a ton of useful use cases for AI with regards to coding, but there will never be a time where you can just tell an LLM to build an amazing app that makes you $20,000 a month and you just don't even have to check it for errors. You don't even have to check it for issues and you can just ship that. There will never be a time where that is possible. It just won't. LLMs are inherently unreliable and random by design. Their output is non-deterministic and random. We've talked about this a lot in this video. Even if your developers are writing code faster thanks to AI, they still have to make sure that that code actually works. They still have to pay the technical debt of organizing and cleaning up that code and making sure it works. And this is something you cannot escape. Even before AI, when developers were writing all of their code by hand, we still have a ton of technical depth in every single production grade project, in every single enterprise project. So why do you think that if AI is writing more code, we're somehow going to have less technical depth? In fact, we're actually going to have more technical depth. And there's two reasons for that. First of all, you're actually writing code much faster, thanks to AI. Not that fast, but like faster than before, right? AI is a much more useful autocomplete than intelligence for instance so you're writing code much faster what does that mean more code more code means more technical depth and secondly you're not writing that code ai is that means you don't have the same depth of understanding of that code since you didn't you didn't write it you know how you know how you build up build a feature and then a week later you look at the code and you don't even recognize it and you have no idea what's going on yeah and that's code that you did write yourself. If AI is writing the code and you're just copy pasting it into your application, you have no idea how that works. You have no idea how the entire thing is set up. And that just means so much more technical debt. And that technical debt needs to be paid back. So yeah, even if AI helps you to write code faster, it's not going to help you build and ship apps faster. It's not going to help you build products faster. Because if you want to build a good product you need to make sure you paying back the technical debt as well It just That just how it works guys A lot of software today is becoming bloated buggy and downright painful to use Windows 11 literally uses React Native in their start menu which makes no fucking sense, by the way. Why are you using something like React Native in the start menu? This is just going to cause performance problems all the way. Instead of this, you should be using something native. And this isn't even the worst thing that Microsoft and Windows 11 have done so far. Microsoft literally wants to preload File Explorer to make it faster because it is extremely slow in Windows 11. You can read the whole tweet. I'm not going to bother reading the entire thing right now. But basically, their File Explorer is so fucking slow that they have to literally preload it, which means run the entire thing, start the entire thing and hold it in the computer memory just to make it seem like it's running much faster. Why do we need this? File Explorer is literally just a tree of files. Why do you need this? It is completely pointless. And it just goes to show you that most of the developers working in these companies either don't have any idea how to write good code or they're just using AI to help speed up the entire thing, which basically just means that they ship out terrible broken products like this one over here. Like it makes no sense for you to have to wait 10 minutes to open up an application that just lists a bunch of files in your system. It makes no sense at all. And I'm pretty sure that AI or vibe coding, I hate that term, by the way, I hate calling it vibe coding. But AI coding, I'm pretty sure that's one of the reasons that we're seeing this and shitification of software all over the world. Managers and CEOs want features and they want them right now. Their view, their non-technical view is that AI is supposed to make you 10 times faster, which means you should be able to ship 10 features for the time of one. And that's just not how it works at all. All this just leads to worse deadlines, lower budgets, and just an ultimately a terrible experience for your users. Heck, if Microsoft keeps fucking up Windows 11 so badly, even in the future, this year probably is going to be the year of the Linux desktop. 2026 might just be the year of the Linux desktop if Microsoft doesn't get their shit together. So, all that to say, AI is not going to replace us. AI is not going to replace developers at all. So, dear managers and CEOs, instead of trying to replace your workers, maybe you can try working together with them to create better products for your customers. AI can be a wonderful technology and it can help us build wonderful things, but only if we augment our developers and our workers with AI instead of trying to replace them with AI because because that is just not going to work. So hope this video gives you some insights, some new knowledge. That's it for this video. If you liked it, press like. If you want to comment your thoughts or what you think about this video and the all the stuff I said, make sure to leave a comment. I read all of my comments and I'll make sure to reply to you and maybe you can have maybe we can have a discussion on this and feel free to share this video as much as you want and make sure you subscribe I post a couple videos every week on topics just like this so if this is something you're interested in then make sure you subscribe and turn on the notifications that's it for today that's it for this video thank you for watching and I will see you in the next video peace