Cursor: 3 Million Lines Later, Here’s the Big Takeaway

So here we have an article by cursor. Uh this is where they made this project to to create a browser completely from scratch. It it's a quite difficult task to create a browser from zero. So they they did it uh and they gave Chpt 5.2 a whole week to run uninterrupted to create this and they wrote this article about it and it's pretty interesting. And you see here how it um starts to render HTML and you see it doesn't really work at the start but as it iterates it gets better and better and they they say it kind of works. It is not perfect. It has lots of bugs apparently and it's it doesn't support all the CSS and everything that the modern browsers have but it's a very interesting experiment that they did and by the way uh Chad GBG wrote like 3 million lines of code uh just earlier you said it's it doesn't matter like you're not judged by the code you write because you know the code is written by these AI tools and you can just you know make three million lines of code and throw it away next day. That's quite literally what we're doing nowadays. So yeah, this is a very interesting thing that is happening and it is very related to the other video uh the Ralph Wigum one. I don't know. So you you uh looked at the article and you said this is they they use a different a different method. So what they are doing they have multiple um agents doing different tasks right yeah they don't mention Ralph at all in this I think this is more of an exercise of and I believe everyone should do this is how far can they push how far can they push AI what can it do for us like I don't know I didn't read or I don't know what the goal of this was but what what do they say are they are they actually trying to build a browser yeah this is they they're trying. No, it's not about I I believe from what I understand it was just to see what they can do. It it seems like uh cursor is trying they're trying to do multiple like they're trying different ways of creating um software every week. And here they uh what's next is interesting. Multi- aent uh coordination remains a hard problem. Our current system works but we are nowhere near optimal. Yeah. Planners should wake up when their tasks complete to um to plan the next step. Agents occasionally run for far too long. We still need periodic fresh starts to combat drift and tunnel vision. These are like we see these like we try to fix something and it the the the model gets to into this tunnel vision. It's really difficult to get them out. Uh so you have to start and it is hard to run them and give them a let's say broad or like general task and and expect them to create something of quality and of detail. they are not there to do such a thing yet because of these issues but we are getting there we are like trying different things Ralph William one of them and multiple agents uh also is another way of doing it uh but the core question can we scale autonomous coding by throwing more agents at a problem let me actually zoom in um at a problem has uh at at a problem has a more optimistic answer than we expect uh expected. Um hundreds of agents can work together on a single codebase for weeks making real progress on ambitious projects. The techniques we are developing here will eventually inform cursors agent uh capabilities. If you are interested in working on the hardest problems in AIS software development um yeah they are trying to hire. So this is interesting to me that they are trying something and then they're publishing their results. That's it. So I think like to to drag Ralph back into this, I think they've encountered all of the things that Ralph uh solves, right? And what these guys cursor are I mean I know they are they've built the basically the best AI uh IDE you can get but they are working within the shackles of what AI can do. I got to phrase this in the right way. They're working underneath AI whereas I feel like the Ralph thing works above AI if that makes sense. like they are, oh, we have to context is built up, so we need to think about how we do that. It's like, well, what if context isn't a thing? Like, it's almost like it's it's so outside the box. It rewrites all of these are just not issues. And and there was a second article that they wrote about the orchestration issue. You know, you don't need agents running side by side and how do you fix that? It's like, well, what if you what if it doesn't matter or what if there what if you run linearly? Well, it's not about yeah parallelism parallelism uh which is again locked into the idea of you have six developers working on six features. Well, if you have one developer fully autonomous working on all of the features one after the next, you know. Yeah. Well, time is the the concern here. I believe like you prefer time is not an issue now because you can you can build features in minutes as opposed to days, you know. Yeah, but if everybody can do it, the whole industry is changing. And now the the Exactly. Yeah. You like smaller teams can do bigger stuff. Uh there was this idea that we will have like billion dollar companies this year uh made by one person, individuals. Yeah. Yeah. Individuals will be able to create like huge companies because you know this new paradigm. Mhm. Mhm. For sure. You can you can have a lot of Mr. M6. Lot of Ralphs. Lot of Ralphs riding. Yeah. Who w who wins? Mr. M6 or R? Let us know in the comments. Who's a What's a better an analogy or character mascot? Yeah, I I think Mr. Music is Yeah, he he's funny. This is part of a larger conversation on my show, Command AI, which we stream live every single week. We discuss the news and all things related to AI in the world of design and web.