Cursor 3.0 is officially the new King (My AI coding setup revealed)

追加: 2026-05-14

[00:00:00]So, Cursor just did something. Now, if you're like me and you already use Cursor a ton, I have some really good news for you because things are about to get a whole lot better. Today, not only am I going to share my entire Cursor workflow and setup that lets me blaze through all of my projects in parallel, the plugins I use, everything, but I'm also going to be diving into Cursor's latest releases, Cursor 3 and Composer 2, and why I personally think you should be using both of these right now. And I also just want to take a second to say thank you to all of you. I already use Cursor every single day in my workflow.

[00:00:27]But because of this channel and what we built here together at the Papa Fam, the Cursor team actually reached out and is sponsoring this video. So, a big thank you to you guys watching and a huge thank you to the Cursor team. Your tech is awesome. So, what's new? First up, we have Cursor 3. It's Cursor's entirely new interface. It's been completely written from the ground up and it's written in Rust for big performance gains. Now, you can quite literally kick off an agent from anywhere. So, any device, any interface, any model. Most of the time for me, I start my work at my desk. And what I love about Cursor 3 is that they've redesigned the entire flow. So you can now switch seamlessly from desk to cloud or vice versa. And then we have Composer 2. It's Cursor's next frontier level coding model. And it's much smarter than before. And it's incredibly fast, specifically built for coding. It actually scores on par with other leading models out there on Cursor Bench, Terminal Bench, and a bunch of other tests. And quite honestly, one of the biggest reasons why I think you guys will love this model is because it's honestly super cheap. So compared to other models like GPT 5.5 and Opus 47 which are both $5 per million input, Composer 2 is just 50 per million input.

[00:01:28]And the same applies for output tokens.

[00:01:29]$2.50 per million output on Composer 2 compared to around $30 for GPT55 and Opus 47. And it gets better. There's also a fast variant with the same intelligence. And no, it doesn't cost a crazy amount either. It's only $1.50 per million input and $7.50 50 per million output, which is still significantly lower in cost than GPT or Opus. And here's something that you may have not known that I thought was pretty cool.

[00:01:52]The more that people actually use the model, the more Composer 2 is improving in real time. Cursor says it self-improves every 5 hours. Okay, but enough talking. Let's hop into the juicy part, my workflow. I'm going to break down my entire setup so you can literally copy and paste it and start blazing through your own coding projects the same way I do in parallel. And while we're at it, we're going to jump through some of the cool new things that Cursor 3 has to offer. Let's dive in. So, we're back at the desk. The first thing I want you to do is go ahead and head over to Cursor. So, if you haven't got Cursor, head to the first link in the description. You'll be able to grab it through that link, download it, and then you'll come to the following page typically. Okay. So, to go ahead and get started, the first thing you want to do is head over to the agents window. Now, you can either click this, and this is actually the whole new cursor 3 interface I'm talking about because chances are you're typically used to seeing this kind of interface. We're going to go to the new one. So, command shiftp is another way you can do here and we can click switch to agents window or new agents window if it's not open already. I've already had it open before. So we can click that and now we're here. Okay. So now you can see this is the new cursor interface. And what I love about this is everything is designed around making it simpler and easier to go ahead and use agents. So in this case these are all individual sessions. And you can see at the top these are basically the different projects that I'm working in. So if we go over here, I like to group by repo, but typically it'll be in by default it'll be workspace. So if we click on new workspace, you can actually open up a new folder. And this folder right now is one of the episodes on the live stream series that I'm working on. You can check it out on the channel for yourself. But it's super super simple to have parallel agents running and see what each of them are doing because that is what used to be pretty tricky to navigate if you had loads of different agents working at once. Now, regardless if they're on the cloud or if they're locally, we can go ahead and work with them. So, I want to show this out in practice and show you some of the bits that I like and the changes that they made. So, here we have an Indeed clone.

[00:03:43]I'm going to go ahead and add something like a FAQ section to the landing page.

[00:03:48]So, we can either type it in or we can simply dictate it. Add a new frequently asked questions section on the landing page. Go ahead and run it in the browser when you're done. So, before we actually run that, we can do two things. We can select the branch we want to run on the also do we want it to run locally? Do we want it in a new work tree? And work trees are incredible. If you don't know what they are, effectively it's project copies similar to how we have git branches that work. Effectively, it's a git branch on your local machine. So that way if you had 10 agents running in different git work trees, there's never a chance that they can overlap and edit the same file by mistake. Whereas if you did it on this Mac, that can happen. So you want to make sure if you are running loads of agents in parallel, start using work trees. They're a lifesaver. And then you have cloud. Cloud will naturally be in isolated environments.

[00:04:34]You won't have the worrying about clashing uh and sort of agents overlapping on the same file. But it's worth knowing that this is something to factor in. In this case, we're going to do a new work tree. And what I like about this is that you don't have to second guess. Everything is kind of in your face. Maybe you don't know what plan mode is. In this case, they've got a button plan a new idea, right? So it puts it into plan mode. You can do shift tab to jump through the different modes, including multitask, which is a really cool mode. We'll jump through in a sec.

[00:05:00]They actually just released this in their April 24th change log. So, it's very, very recent. Uh, we can talk about that afterwards as well, but you can jump through everything, change your model. In this case, I'm going to be using the new composer to fast, right, which is the fast version of it. So, we're going to go into plan mode, and then we're going to go ahead and kick this one off. Now, you can again run it in the cloud. We're going to run it locally. Okay. Okay. So, while that's running off, I'm going to add a new agent in the cloud. So, I'm going to go ahead and trigger this one in the cloud.

[00:05:26]This time, I'm going to say, I want you to add in a new testimonial section in the landing page. Test your work. Send screenshots. Go ahead and get it built.

[00:05:34]Right. And then my accent always seems to get a bit caught up. So, in this case, yeah, built, right? And then we're going to change this model to composer 2. Okay. And then we're just going to shoot this one to go ahead and start working. So, I'm not going to do a plan mode here. I'm just going to go ahead and execute that one. Now, what I love about this, as you can see, this is the same repo, right? But it's got a clear distinction. And this one is in the cloud and this one is locally. So what I actually like to do is go down here and click on the repo. Right? So group by repo. And you can see now that it's way clearer to see. Right? So both were clear before, but I personally like this view. So I can see my individual project. I can say this is a cloud branch or session. Right? So a cloud agent is building here. And then this is on a work tree. And then you'll notice whenever you need to get involved, it will have a yellow dot. So in this case, I click on the yellow dot. And of course, I need to approve the plan. So at this point you can come in here and you'll notice that on the top right we get all of the tooling now right so if you're wondering before where can you see the files inside of cursor 3 now you can see it's secondary so effectively we can still access and deep dive when we need to but we the main focus is shifting towards sessions and interacting with agents as opposed to coding ourselves. So while that's done let's go into the the plan and here you can make changes go ahead and add to chat and basically go back and forth but in this case I'm going to trust that it did a good job. We can click build right or you can maybe build it in the cloud or in parallel. Let's say for example most of the time I start my work at my desk. But if I was for example you know I wanted this to get done and it needed to deploy ASAP then for example I could go ahead and do build in cloud and then from my phone I can simply go to cursor cloud agents on the web and I can just check in and carry on from there. Right?

[00:07:14]So it's pretty cool how we can do this and then you can also build it in parallel which I love. This is a new multitask mode. So we can check that out in a moment. But in this case, we're going to build this one locally. So I'm going to go ahead and build this one locally. And then we're going to do one more, which is going to be for the following. So we're going to go ahead and say for each section, I want you to add in nice icons to represent each section. So I'm going to go ahead and click okay for that. And then for this one, I want to try out that extra step.

[00:07:40]Right? So I'm going to do a plan new idea. And in this case, nice icons for each section. and for each section in the landing page in the landing page and then I'm going to keep it on composer 2 and shoot right so that's in the cloud now this is really cool I've got three different agents running really easy to see what's running where this is running locally on a work tree and if I hover over it I can actually see the actual file directory over here I can see that these two are running in the cloud super easy super nice and then the cool thing is I can jump between projects now so I can jump between you know a different project I'm working on and this is exactly where they're going with this new interface and this new way of working with agents. And this one came back to us and you can see it's now done with its work and it's waiting for me and it was already on the page which is why it's blue. Let's go ahead and execute that build. And in the meantime, while these three agents are working, I'm going to dive into my setup. So for my setup, I'm going to go to settings and I'll show you our plugins. But you can actually get to this point a lot easier just by simply going to the marketplace. So cursor in my opinion has the best selection of plugins that I have personally seen. And I know this because I have installed my favorite companies like Clerk, ComX, a few of these guys all inside using plugins. And I can't find these plugins elsewhere. So it's only in my opinion, I've only found them inside of Cursor Marketplace. So it's definitely something that I've noticed. Now once the marketplace opens, you'll get access to a bunch of different plugins. So in this case, let's say, for example, I wanted to install one. We go to Clerk and I've already installed it. But you can see it pulls in an MCP, skills, and a few others, right? And it's not just MCPS and skills that it installs as part of the plugins. If we go to one of my favorite plugins, superpowers, this one actually pulls in, for example, a bunch of skills, sub agents, commands, and hooks as well. And this really upgrades your your workflow. So when you do this, you can either choose to install it for the project that you're working on or as a user in itself. So in this case, if I went to install, let's say for example, I wanted to install work OS and I click add to cursor. You can see I can either add it to my personal or I can add it to individual projects. And the cool part about that is that that that's going to actually get applied to the agents, meaning the cloud agents if I added it to the project that they're working in or it'll be to myself, including all the agents I work on or my local stuff if it's on my own profile. So, I'll show you the ones that I've installed. So, if we go to settings and we go down to plugins, you can see I've got all of the following here. So, in this case, you can go ahead and pause the video at any point in time, see for yourself. Uh, but the main ones I would focus on are the ones that are not imported. Right? So, if I was, you know, if you were asking my opinion about which ones are most helpful, this is the text stacks that I tend to work with most. And you can see here that they just honestly massively help me out a ton in particular.

[00:10:19]Superpowers. This one is really, really a killer skill. And what I do is I use it for code reviews mostly. So, once a agent is done, I will ask it to use the superpowers uh plugin and skill and basically go ahead and execute a code review. And it's it's incredible. You should give it a try. It's really really good. You can access cloud agents, work tree settings, everything else from here as well. But even for skills, it's not really worth showing you my skills because I'd recommend if you were to start from scratch, I'd probably just use the plugins instead because you're going to get all the skills anyway as part of it. And then for tools and MCPS, cursor makes it super simple. You can see for yourself, you've got your your MCP servers here. But the most important thing that I like is that it's just simple and clear to understand. It's not just about editing and you know, an MCP config file in this case. If something needs authenticating, it will say it needs authenticating and then you can just click connect and it will start the Oorth flow. So, I really like this and I think it's really simple and in my opinion it's just it's just sweet, right? For appearance, if you guys are going crazy right now cuz you see me using light mode. I have it on system, but you might prefer cursor dark and so forth. So, you can feel free to mess around that yourself. So, back to our agents that are running now. Besides obviously being able to see them easily, uh, as we can see on the left hand side, one thing I really like is you can have agent tabs. So, if you go to the top right and you click split, you can actually go ahead and have different agent tabs running. So, this is pretty cool now because you can just go ahead and grab an agent session and go ahead and pop it over here, right? And in that case, you can see I can have splits, I can have quarters and so forth. If I remove this split now, I can have both up and running on the side. And it just makes it a lot clearer because sometimes you might have felt like, you know, I I can't see everything that I wanted to see. Well, in this case, you can actually go ahead and do it like that, right? So, in this case, it's pretty cool. You can see like plus one and that is the two that are running up here and it groups them together. So these two if I click on this it will load like so whereas this one will load its own window but this is something which I definitely use dayto-day. So definitely take advantage of agent tabs if you haven't if you didn't know about those let me know because I actually found out quite late about that and it was definitely welcomed. Okay so once an agent's finished working another thing you'll notice as well is that this has been created into a draft PR. All right.

[00:12:27]Now, you see on the left hand side, the actual icon changes. So, when you look at that, you can I definitely recommend getting used to seeing what the icons are. So, in this case, because we have a draft PR, if I click it now, it'll take me to that PR. And we can pretty much go ahead and review the PR. So, in this case, I'm going to mark this one as ready to convert it. And now you can see that is a open PR now on the left. So, once the work's done, we've got two options here. We can either go ahead and move to local, right? Or in this case, cuz we have a PR open, we can squash and merge it. Okay. So for this one, for example, I'm happy with it. I'm going to go ahead and squash and merge. So let's go ahead and merge this one in right now. You can see it's merged. And then I can come over here and simply just archive. Or they have the uh keyboard shortcuts that pop off. So in this case, command shift E. Yep. And that's archived. Right. And the thing I like as well is you'd be surprised how many times I've accidentally archived something. So you can also undo it, which is pretty handy. Okay. Okay. So now a demo that I wanted to show you is if we have something on a work, but we want to test it out, right? In this case, I've got a couple of changes here.

[00:13:24]But I would rather see this before I'm happy about, you know, committing and pushing or simply moving it to my local branch. In this case, what we could do is we could simply go ahead and run in the browser. We go ahead and just simply run in the browser. While that happens, I actually want to mention one little cool thing that I saw they added down here. You can see your usage, which is really easy to see, and also the context. If we click this, this is actually a new little popup. And it's really nice because if I find I'm burning through a hell of a lot of to tokens at one point, now I can see the breakdown of where those tokens are basically stacking up. This allows you to be a lot more efficient, a lot more optimal, and it's just overall really nice touch. So if you ever feel like you're burning through a hell of a lot of tokens, I definitely recommend checking out the little pop-up. And as you can see, it popped up on the screen on the right hand side. So in this case, it opened up in the browser. So what we can do now is actually mess around with it. So we can see the FAQs should be added in. And here we can see. Perfect.

[00:14:17]Right now, this is brings me on to the next point, the design mode. So, I'm going to press command B to hide the sidebar. Now, let's say I want to make a bunch of changes in rapid succession based off things I see on the screen.

[00:14:28]Right? Typically, we'd have to maybe screenshot, paste it in. It can get a bit tedious. Cursor 3 allows us to do this a lot more efficiently. So, if I click on design mode now, I can actually go ahead and see I can toggle the specific things. So what I actually want to do now is maybe make the first one, you know, extra highlighted. So maybe that's like the most important question which people ask. And our color scheme is green. So I'm going to say change the background of the first frequently asked question to green, right? And I'm going to go ahead and submit that. And that will actually go ahead and get an agent, you know, making that change. But what's even cooler is the new multitask that I mentioned that they released at the end of April. So this one is actually really nice because if you press shift tab, you can actually go into multitask mode and I can actually do a bunch of things now.

[00:15:13]So let's say for example when it's open I want this little segment. So when it's expanded I want this segment to have a slightly you know indented UI.

[00:15:26]So in this case I can go ahead and spin off this agent. And now if I exit this I can go up over here and let's say for example I want this posting. and go and say I want this to post. And I let's say maybe the same for this. I want this to post. And now you can see I'm stacking up different tasks that I want to do.

[00:15:44]And even for example here, right? Let's just go ahead and say make this red, right? So we've got three tasks. Now typically what would happen is you'd be blocked by your primary agent, right? So the primary agent would block everything else. But with the new multitask mode, it actually is really really nice. So you can see now it's spinning up a independent agent every single time and now it's not blocked by the first. So you can actually see look it's actually spinning up a background agent to take these tasks from us. So this is a game changer in my opinion because now this is not the same as having several separate sessions but in the same session now when it gets the resources to do so it will go ahead and spin up a sub aent and as you can see look it did it there right? So in this case, it will spin up the sub agents and it will go ahead and do it. So that way you're not blocked by your primary agent, right?

[00:16:34]And it will go ahead and carry on implementing the changes that we asked it for. And this is a really cool way that I've now started to work a lot more. So I actually find that I'll go ahead and have the app open on the right and then as I'm using it, click on design mode and I'll just simply pop out exactly the elements I want and just explain the changes. And previously it did get a little frustrating that without multitask mode because I'd have to wait for one after another to go ahead and complete. However, in this case, you can see I can have several sub aents running at the same time, right?

[00:17:04]And this is really, really nice to go ahead and see. And if you click on this working over here, you can actually get a list of all of the actual different agents working. One of my favorite features that Cursor 3 have added. And when they finish their work, you can see they basically report back to the main session. So here we can see this is what I asked it. This is the actual agent's finished work. And if we click open, this is kind of the report that they give back to us. So here's what I here's what was implemented. It head and shows it goes ahead and shows the changes. And there we have it, right? And that the red heading change landed successfully.

[00:17:38]So if we go ahead and scroll down now, you can see there we have it. The changes have been made and say, yep, the agents update is in and verified. And now you can see it's very subtle, but there is a posting happening here. And what's really cool about this is that basically your sub agents effectively report their work to the main agent in the thread and then will effectively go ahead and basically merge the changes for you. Another really cool feature that they actually added which I used in one of my recent projects was actually multi-root workspaces. So effectively imagine you have a front end and a back end and to be honest with you they actually have a really good video on the website that shows it off. So when you click on that dropdown, you can actually see for yourself that they have if you click on set up workspace, you can select not just one repo or you know folder where your project lives, but two. Sometimes we might have like an expo app for example for our native app and then maybe like an X.js website with the you know backend side of things, right? So we might have our back end running over there and our native app running over here. In that case, we might want to give our agent both of those contexts. So that way it can understand exactly what changes need to be made if it's not in a monor repo.

[00:18:44]This is exactly how you can do that by using the multi-root workspaces and it allows a single session or a single agent to basically target reusable workspaces made up of the different folders. So I thought this was a really cool thing and it does it it does nice little bit of separation between the two as well. So definitely give this a try for yourself. I did use it recently and I'll be honest, it was pretty cool. So, of course, our work tree is now done.

[00:19:07]How do we merge these changes in? So, we can go ahead and actually get kind of cracking with everything else. So, I showed you previously how to merge in a cloud agent, but in this case, we have a work tree local, you know, agent working here. How do I go ahead and proceed now to get this in this code inside? And this is what I like about uh cursor 3.

[00:19:24]They've made things very, very simple, right? You just got buttons that you click. You can review, right? Very simple. You've got a terminal if you want to go ahead and inspect things. Um, so in this case, the main thing that we want to look at is really the either move to local button or create PR. So in a work tree, you don't always have to create a PR, right? Sometimes you just want to pull those changes. So these, you know, 90 changes and three deductions, we just want to basically shift that to our local branch. In that case, we can just go ahead and click on move to local. And by clicking that, it will literally shift that into our main.

[00:19:55]And then effectively, we can go ahead and retire this agent and continue on with another task or new session. And the same applies if it's on the cloud.

[00:20:03]So in this case you can see on the cloud doesn't really make a difference. They made things very very simple. You can see we can just go ahead and click on move to local over here. And also you can see how if we have got something like a merge conflict. We can simply go ahead and click on merge conflicts. It will go ahead and trigger the option for the agent to go ahead and fix those merge conflicts on its own. And things like this are just very intuitive because for example without that I would have to open up the PR see that there's a merge conflict and then get the agent or myself to go ahead and manually resolve those conflicts. Uh most of the time I'd get an agent to do that but I typically my old workflow would have been I would just ask the agent and those things stack up when you're working across tons of different sessions. But now having a simple button where you can just click it it resolves the merge conflicts and it pushes super nice to have. And there we have it. It fixed the merge conflict. And one thing that is really cool as well, like I said before, let's say, for example, this work was still ongoing and I kind of needed to run out the office right now.

[00:20:57]I could go ahead and do for/cloud and I can basically switch this to cloud. And this in this case, we can carry on from elsewhere. And a nice little example of this is where we had a cloud agent running previously. If I go over to the uh actual web, so in this case, I've actually got it open over here. So this is over at cursor.com/agents.

[00:21:16]And then here you can see I've got all of my cloud agents running. So, this is a really, really nice example. And as you can see, if we scroll down, it tried to actually take a screenshot, but it found it didn't have the environment variables. So, I'd have to go ahead and configure this agent so it would have those environment variables cuz that's something you would have to do. Every project that you run, you typically have to set up the environment variables. You can do all of that inside of the cloud agent settings per individual repo or project. And then it will literally take screenshots, even take videos. And I actually have a little example of where I did it. So, over on the website, they actually got a really cool example, and I've done this a few times, actually, where you can actually just ask it to test on a cloud agent, and it'll spin up a video. And to be honest, I don't know how they do it with the smoothness because this is similar to how I will do my recordings, but you'll actually get a video on your session of your agent spinning up a computer session, going ahead and actually testing it, clicking it, navigating through things, silky smooth as well. And you can test out flows like this and it basically will get attached to your session over here.

[00:22:19]And I thought that was really, really nice. But as we said earlier, it's super seamless to go ahead and switch between regardless if I'm on the local, if I'm on the cloud, or if I'm away from my desk and I simply want to just carry on with an agent. I can do that using the web cloud agents. But an honest breakdown as to how I use my setup every day. GPT 5.5 gives me reliable results every single time. So I'll use that for specific tasks where I need that kind of approach. I tend to get very good creative results from Opus 4.7 and honestly for a lot of the other things I have started using Composer 2 a hell of a lot more especially when I go ahead and trigger cloud agents to work in the in the background and as I've gotten used to using cursor 3 and this style of working more I've definitely found that my productivity has gone way like through the roof in terms of the features that I can go ahead and get built and trust me guys like it's so easy once you understand how to do this like I showed you today whether regardless if you're on cloud cloud, local, and even then when it comes to finishing the work and actually merging that work back in. Cursor 3 has made it really easy for you to not have to be technical, you just got buttons that will pretty much pop up when you need them. For example, in the beginning, a plan mode button makes perfect sense.

[00:23:29]And then when the work is finished, what the next appropriate steps are, whether you're on a work tree on the cloud, it simply just gives you options. You don't have to know what to effectively do. You can just click on those things. It's basically simplified how to work with agents at scale across different projects. And for me, that's a personal game changer, especially when you're juggling quite a lot at the same time.

[00:23:49]This is how my day-to-day now looks when I'm coding. I'm pretty much using this interface every single day. That's why it was so cool when they reached out to go ahead and do a video with us. And if you're wondering, you know, can I still get my hands into the code and everything like I showed you before, all the tools are still there. So, for example, if I were to jump into a file, you can still go ahead and do everything the same way. Jump in and do a deep dive when you need to. It's just that the majority of the responsibility now is shifting towards sessions as opposed to manually coding. And I would definitely recommend just dialing in on some of your practices for example using agents MD files correctly, using skills correctly, understanding how to use plugins, MCPS, all of those things. And I promise once you get those things under your belt, using like this become an absolute game changer because you can really excel. Trust me guys, I've been doing this for nearly 20 years now and this has been the new way that I like to work. But give it a try for yourself.

[00:24:39]Use the first link in the description and let me know in the comments if your setup looks similar to mine in terms of plugins or if you have something different which I should know about. So there you have it guys. That is my cursor custom setup. Feel free to copy it and let me know down below in the comments how your cursor setup looks like. Is it similar to mine? Is it different? I want to know down below.

[00:24:56]And while you're at it, give Cursor 3 and Composer 2 a try. And be sure to use the first link in the description. It is an affiliate link and it helps this channel out a ton by telling the guys over at Cursor that you came from this channel. So, I appreciate that. And as always, guys, don't forget to like and subscribe to the channel if you enjoyed this content. And let me know what you want to see next on the channel. As always, it's your boy Sunny, and I will see you in the next one. Peace.

[00:25:28]I know I love you.