Cursor 2.0 is Here - Anyone Can Code Now!?
Let's learn how to use Cursor AI📅 2025-10-29
Transcript
[00:00] Cursor 2.0 just got launched. Can anyone code? Now, in this video, I'm going to lay out everything that's new in the new Cursor 2.0. They say Cursor V. And I'll be honest with you'all, Cursor has hit a home run here. Let's check it out. Welcome back, y'all. Let's check out how we utilize Cursor 2.0 as we got some pretty cool updates here. Right off the bat, I'm going to enter in my test app here. This is a little React app we created. If you don't even know what a React app is, or what's a test app, what are we doing here, Corbin? Come over to my GitHub, find my repository called test app. You can download it. You can fork it. You can clone it into your own
[00:31] git repo. But essentially, this is just like our little nice little sandbox application. We're going to be able to do a bunch of tests with and mess around with GitHub in the description down below. So, first things first, because we're all visual learners. I want to see something visual. Corbin, we're going to run this app first so we can see it at local host. Run the app. Boom. You already noticed right off the bat that chat looks a little bit different. So, we're going to see how we can leverage those features. And actually, let's use one of the first new major features here. Cursor as of recently added the ability to do browser, but they gave it a massive upgrade. If I come over here to the globe, you can notice that we can
[01:02] now open it in a browser tab within the IDE and also Google Chrome like we were able to do before. But I really like this browser tab, y'all. If I click this behind me is going to be the tab that cursor AI is going to be able to see vividly when doing development in the front end. So, for example, we're going to do the prompt run the app at localhost 3000 and navigate to it in browser. Hit enter. In the past, this would open up an incognito browser in Chrome, which is fine. It worked. It was beautiful. But this is way better, y'all. So, we're gonna let this run real quick. And boom. We have localhost 3000
[01:33] in our IDE. Now, why do we care, Corbin? Why is this a big deal? Cuz watch this. If I come up here to this tool called select element, I can select any part here and change it. So, for example, I go to the title and I could be like change to democratize developing. Hit enter and cursor has context that I am referencing this exact part. Now for something simple like a header one right there. Some of y'all are like okay who cares? That's super simple. But when your application gets extremely complex, we've only had the ability to either do vision context, eg screenshot this, or
[02:05] alternatively put text input here. What I really like about this feature here is that with really complex frontends, I have the very easy way of identifying very specific parts of the front end. So therefore, that is dub number one for the 2.0 cursor. Now, of course, because it's in this kind of environment, you might be like, Corbin, how do I get console logs here? They put it right here. So I click this. This is like the Chrome UI. Come over to console. And this is how you're going to be able to see all of your console logs here. Little debug messages, a little bug being like, hey, something's up with the code type of messages. Here is a couple shortands that could help you out in
[02:35] chat as well. Because I notice sometimes it'll say like, "Hey, I'm looking at it." But it's not really looking at it. First, you can use this thing to take a screenshot. these three ellipses, take a screenshot, take a full page screenshot, and this will add it to the underlying chat so it can essentially know what you're talking about. Another thing you can do is quite literally tell it in the prompt to look at the page. So, a prompt like navigate to localhost 3000 or navigate to localhost, whatever your port number is, using the terminology of navigate, it is receptive and it will actually navigate to it. Nice. I also did another video on this topic when it
[03:06] comes to browser and its power when it comes to taking other people's code, specifically front-end code. If you want to check that out, go and just check out my channel here. It's literally called how to steal code with cursor AI or something like that. All right, let's check out another feature just to lay it out of all the different features here. We have a new aenic coding model, which we'll check out. Agent layout, which will be next here, browser, which I just showed you, parallel agents. This is interesting. Multi- aents, a bunch of different models, and in editor, bugbot. First, let's check out this new agent mode. I come up here to switch to agents. So, what cursor AI did here is
[03:36] they made it so you don't even have to look at the repository code. You essentially just talk to your app. Now, for example, if I'm like, you know what, I want to talk to my app, but also have my UI here. Look at this double whammy here. I can talk, have my UI, see what's happening here, and just go crazy. Also, I have all of my pass agent task here as well. If some of y'all are oldfashioned like me, I don't particularly love agent mode because of the fact that I do kind of want to see my code in my repository, but this is very much a UI that is very vibe codify friendly. So, in this new
[04:07] editing mode, let's do another task here. We might as well choose the cursor model that's new. So, I'm going to click this. It's called composer one alpha. And let's try something. Let's do the simple prompt of make my app a lot more fun. Add confetti on the plus and minus button right above me. Plus two, minus two, everything like that. Composer one alpha selected. Hit enter. Your next question might be like, Corbin, is composer one the next best model? To put simply, when it comes to choosing the right model for coding, it is extremely circumstantial on what you are building. Therefore, my opinion on this topic
[04:37] changes a lot. When I first started coding with AI, I was absolutely in love with chat GBT. But Claude sonnet is really pushing numbers. Now, when it comes to this new model with cursor composer, the way you gauge whether or not this is the best model for you is use it. And if you notice the code outputs are like nixel good kaput or just trash, don't use it. For my specific use case, I still kind of like claude a little bit. But who knows, maybe they will improve this model a lot more in the long term. To be fair though, I still need to keep testing with it to get a better idea.
[05:07] Democratize is developing. Oh, I like this. It's at plus two minus two. Confetti. Confetti. Confetti. Confetti. Confetti. Wait, what did I say? Confetti. Put me on the counter. Put me on the counter. All right, let's check out another new thing in Cursor 2.0. And we'll stay in agent mode because, you know, I'm not an editor anymore. So, another thing that I thought was cool is multiple models. And what this allows us to do is before on auto we let cursor AI choose the best model for the use case but you might be like cursor I know better. Now you can
[05:38] do use multiple models here and maybe you're like I like composer I like GBT5 I absolutely hate cheetah don't choose cheetah let's add 4.5 sonnet and now it will dillydally between these three models. I like this. This is a very nice granularized way of handling this comparative to auto which is like which one are you using right now? We'll keep it at auto here so I can show you the next major thing y'all. cycle agent count. This is crazy. So, let's do two. What this is going to essentially do is you are going to put in a task and then it's going to give us two different
[06:08] types of outputs here. So, you can go all the way up to four if you want to do that. Let's just see an example real quick. So, let's make our nice little front end here a little bit more interactive. So, when I do minus and plus, we have a nice little bar chart. Add a bar chart for the number on the counter. Hit enter. And look at this, y'all. This is crazy. So notice that we have two separate chats essentially going off. But in reality, this is two different agents that are like, "Hey, let's go. It's time to go. It's flow mode." And they are firing off on that initial prompt here. What you'll also
[06:39] notice is that while this is incurring, I hit new agent here. And I can shoot off another task. What's happening, Corbin? Where where's the older task? Don't worry, it's right here in progress. They've categorized it for us. From here, I can shoot off another task or alternatively wait for this task to be done. to be honest with you, unless the other task is a completely separate part of the code, eg you're working on something on the front end, but then you want to shoot off another task for the back end, I would use it. But if it's the exact same repository or if you're using a poly repo where if the other task has to do, let's just say this,
[07:10] okay, you are dealing with monetization of Stripe. You run one of these agents and it's shooting off and you have like four different versions of that. I wouldn't start a new agent on the topic of monetization of Stripe cuz the code can get conflicting. It can get very confusing. Basically, if you ran an agent to solve how to upload a profile image for the user, I wouldn't create a whole separate agent to run that same task. But a little nuance. That's why we use the cycle agent count method here. We got two outputs. You'll know it's done when it says add a bar chart. It's in the review process here. And we can
[07:42] see two different versions of it. The first major thing you should really look at when it comes to multiple versions of an agent running this kind of flow is how many lines of code were added and removed. Plus 33 means 33 new lines of code were added and plus 92 means 92 new lines of code were added. I want to see the code I can simply come down here to review and you will see each one right here. What is a line of code Corbin? 28 29 30. These are lines of code and the reverse is true as well. So if I come over here it says minus two. So two lines of code are removed. Now what do
[08:12] you need to know as a software engineer when it comes to what that means for your app? You know that old saying more money more problems? That could be a little bit similar here with more lines of code, more problems. The code could be perfect. Those 93 lines could be absolutely perfect. But what you are doing is you're introducing 93 new lines of code into your software. Where in the beginning of your app, that's not a big deal cuz obviously you're pumping out code. But in the more maturity stage of an app, 93 new lines could be a red flag. But to be honest with you, like 93 is not that much. I'm thinking more along the lines of maybe 200, 300 new
[08:44] lines. That's where it gets a little bit like what's incurring here. But keep that in mind. Now, let's compare the two. So, I got the first one here, auto apply. So, when I ran that, it actually broke. Keep that in mind. This is why we typically do PRs, and I'll show you what that means pretty soon here. But the idea is that it had to change some code here, so it actually would run. Here is auto one. We go. I mean, the visualization's kind of trash, [laughter] but obviously, we could have been more specific with our prompts, but this is its visualization, right? Value one minus 4. And it's like a nice little
[09:15] blue bar. So, I'm like, you know what? This is trash. We can check out auto 2. I don't know why I'm calling them auto one, auto two. That's essentially my way of saying the two different agents that ran here. To do that, I'm going to say apply all. And then we here we go with auto two. Obviously, not as good either. Now, to be honest with you all, when it comes to front-end development and the multiple workflows, I wouldn't necessarily suggest it. I think just doing one shots and one goes is probably the best method. Let me show you a safer way to see front-end changes than what I just showed you. I just wanted to brute force it so you get an idea of what that even means by variation. So coming back to agent mode here, I want you to come
[09:46] up here to create PR. What this will do is will automatically create a pull request and therefore we can create this pull request and see if we would want to merge it into our main branch. Like I've already said thousands of times on this channel, if you don't know what GitHub is or don't know how to leverage it, I'll leave that 46-minute tutorial in the description down below again so you can understand why we even merge. What's a PR? What is GitHub? All that stuff is explained in that 46-minute tutorial. Let's see the next cursor AI feature. There's two other things here that I want to kind of point out in today's video, and that's going to be the audio one. So, voice input. So, for example, I
[10:17] need you to make my UI way better. Way better. Make it look like perfect Apple UI. Absolutely perfect. Stop recording. Enter. And some of y'all like this. All right. Some of y'all been tweeting at me like, I love just the voice input. I'm like, okay, are we calling this yap coding? We're yap coding now. And with that, we'll see what it does. It's Apple, so you already know they had to go with that white light UI theme with the bright blue W or L. Let me know
[10:48] in the comment. There are other features found in the Cursor 2.0 update, but those are probably the biggest ones that stood out to me. One little quick tip when it comes to when I would use a parallel workflow or maybe a 4x, 3x or 2x in the sense of I put in the prompt and get three different outputs, two different outputs, four different outputs. I would suggest using that kind of workflow for more back-end heavy logic. Eg what is the best way for me to upload a PDF see all the answers for it see all the code for it and then you can kind of decipher that way which would probably be the most efficient way to do that because one very interesting thing
[11:18] about coding is that you could in theory arrive at the same end value so maybe your value is analyzing a PDF but the underlying pipeline to compress the data of that PDF what is the most efficient cost-effective way to do that at scale that's a real thing you could have five different ways to compress a PDF's data Uh, but four out of the five is like a C++, but one of them is an A+. Therefore, we could use that multi- chat feature. Maybe not multi-hat or whatever. Multimodel feature. Okay. Two,
[11:50] three, four. Zia. If you want to see what's new and cool on this channel in AI and all these different tools, make sure to check me out on X description down below. But as you already know these style videos, I'll see you in the next. I am an absolute love of cursor 2.0, know, but specifically the browser feature because now on front end development is so easy. You can quite literally point at anything in the front end and be like, you know what, change that. I need a different color. Wait, make it bigger. Make that text bigger. I like the text bigger. Put on the top type of video. Put me on the counter. Put me on the counter.