Transcription
If I look at coding, programming—which is one area where AI is making the most progress—um, what we are finding is we are not far from a world, I think we'll be there in 3 to 6 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. That is a stunning statement from Anthropic CEO Dario Amodei, and it actually might be more true than you think. I'm going to explain exactly why I think that might be true. It might be a bit of an exaggeration, a bit of an aggressive timeline, but I actually think we're a lot closer than most people realize.
So first, let's just start with Vibe coding. Vibe coding, as a concept, has taken over the Internet over the last few weeks. Now, as a reminder, what is Vibe coding? It allows basically anybody, even somebody who has never coded in their life, to write code, to iterate on code, all without having to write an actual single line of code. What you do is you use something like Windsurf or Cursor, or you go straight to a model, and you do copy-pasting, which is a little bit more tedious but still works. And you essentially tell the model what you want built. And so all of a sudden, there is this explosion of people who can build software, whether that's SaaS applications or games or web pages, portfolios—anything you want, you can build without actually having to know how to code. Keep that in mind; that is the initial trend. And so it's not just individual developers who are now using AI to code; it's everybody. It's the largest companies in the world.
Just a few months ago, we had this Forbes article: "AI writes over 25% of the code at Google." Google CEO Sundar Pichai made a significant revelation about the growing influence of artificial intelligence in software development during the company's third-quarter earnings call on Tuesday. According to him, AI is now responsible for generating over 25% of new code for Google's products, while human programmers oversee and manage these AI-generated contributions. And they are certainly not the only large tech company using AI to code now. They're probably not doing Vibe coding. Most likely, in a large company like that, where every line of code might be used by millions or even billions of people, they are most likely doing tab completion. So you start to write a piece of code, you hit Tab, and it completes the code for you. That would still, under these definitions, be considered AI-coded. And don't forget, just a couple of months ago, Mark Zuckerberg went on Joe Rogan's podcast and said almost the same thing: "AI agents are going to be deployed inside Meta as mid-level software engineers, probably by the end of this year. Let's watch, probably in 2025, we at Meta are going to have an AI that can effectively be a sort of mid-level engineer that can write code."
And so we have AI models that are incredibly capable of writing code. We have multiple code editors that allow you to do Vibe coding without writing a single line of code, and they're really capable. We have more and more the ability for this AI to control the dev environment, which again, just gives them more control, allows them to figure out issues and iterate on its own to solve bugs, solve errors. And we haven't even started talking about agents yet. Now, with agents, we're giving these core models that are already incredible at writing code so much more ability, whether that's memory, tool usage, frameworks around them to make them better, which include prompting strategies and a lot of other things. Just think about Manis. Manis is an agent. Manis has the ability to search the web for API documentation; it has the ability to spin up containerized environments, create folders, navigate through those folders, take notes, extract the notes, iterate on the notes, and it is really, really powerful. It can even run your code for you.
So do we really think, within 3 to 6 months, 90% of all code is going to be written by AI? Well, again, let's think about what that really means. So let's use the more conservative, quote-unquote conservative, timeline of 6 months. Now, we have all of the code being written by humans in the entire world, but the amount of code written by agents and AI in general is accelerating at a much faster pace. That is because humans are the bottleneck in writing code. We can only write code as fast as our brain can work; even with automated tools, it's still only as fast as we can think. But for agents, they can write a lot more code in a much shorter period of time, just because they are able to think a lot faster than us. So the total amount of code being written by AI is accelerating at a much faster clip than what is being written by humans. Plus, humans more and more are getting used to writing code with AI and Vibe coding. But it's not perfect; there are a lot of issues with the code that AI writes. It's not perfect; it also struggles to understand really large code bases. I've definitely run into issues as I'm Vibe coding where, once my project gets to a certain size, it really takes a lot more effort to make changes and to add features to it. But still, the projects that we're seeing on Reddit and X that were completely Vibe coded from scratch are incredible.
And so that original clip by Dario was part of a longer interview that he did, and I want to show one more clip which really speaks to not just coding but all thought work in general. And I think this stuff is coming a lot faster than people think. We're going to have to reimagine what society is like when AI can not only do things better than us but faster, cheaper, and really at higher volumes than we can ever imagine. We essentially have infinite intelligence on the horizon, coming very soon. Watch this clip: "We're going to have to look at what is technologically possible and say we need to think about usefulness and uselessness in a different way than we have before, right? Our current way of thinking has not been tenable. I don't know what the solution is, but it's it's it's got to be, it's got to be different than we're all useless, right? We're all useless is a is a nihilistic answer. Um, we're not going to get anywhere with that answer; we're going to have to come up with something else. So we are really going to have to think through things. I have said this for a while now. I think in the short term, we are going to have this explosion of developers; the amount of people that are going to be able to write software, build useful things in software for the world, is going to exponentially increase because we have all of these tools that are going to enable people who don't necessarily know how to write the nitty-gritty of the code to be able to build things. And then in the mid-run, we're going to have this transition period where all of a sudden AI is just able to write the code on its own, and less and less it's going to need human intervention to actually write useful things. And when when I see things like Manis, which you essentially give it a prompt and it goes off and does the entire thing over minutes or maybe even hours and then comes back with a finished product, that's kind of in the direction that I see it going in the midterm. Again, far from perfect right now, definitely makes a lot of errors, a lot of bugs along the way, and the final output is not always great, but it kind of shows what is possible. It is the seed of what is coming. And then in the long term, and this is just complete theory of what might happen, but I suspect code is no longer going to look like what we think of as code. The reason code looks like it does today, so mostly the English language and mostly looks like regular sentences, is because humans are really bad at coding; it needs to look that way for us to be able to read it. But in this future world where AI writes all the code, AI is reading and executing the code; code doesn't need to look like that. It could be completely different; it could be symbols; it could be sounds; it could be something that we can't even imagine. And then I think about that example when two AIs started using gibberish to talk to each other because it was so much more efficient. Now, gibberish is created by humans, created by ElevenLabs; it is decodable by humans, but it won't always be that way. There might be a future in which we can't understand their language, and we certainly can't understand the code. That's kind of already the case with how large language models work. We understand that you put an input in, then there's this black box in the middle, and then you get the output, and we understand it's predicting the next token, but we don't actually know why all of the nodes or the atoms, whatever you want to call them, inside the model are firing, why they fire when they fire; we don't really understand it. And so that is what I believe is going to happen in the long run with coding. We're just going to have this language that is uninterpretable by humans. Let me know what you think down below. If you enjoyed this video, please, please consider giving a like and subscribe, and I'll see you in the next one.