Agentic development changes with Cursor 2

- Cursor arrived in the development scene as a disruptor, an IDE built from the ground up for AI-powered coding. But over time, token usage and pricing made the platform feel a bit constrained. However, with version two, they've made some significant changes that focus on parallel agentic workflows. Let's take a look. Now, the first thing you'll notice is that it doesn't look that much different than the previous Cursor, and that's because it now comes in two different modes, the regular looking editor mode that you're used to and a new agents mode, right up here on the left-hand side. This is a different looking layout with a chatbot focus. This is designed to look like a vibe coding application. It lets you run through multiple agents, even in parallel, to take care of things. One of the great features is the new model called Composer. So if you want to switch to it, you can click right here to get out of Auto mode, and then you can pick this new Composer 1 model. Notice that I also have the ability to use multiple models if I want to. However, you do have to have an Ultra account to use that, and that's $200 a month. But let's just choose Composer 1. I'm going to say let's add a light/dark mode toggle with icons. Let it read the user's computer settings and store state in local storage. This is going to start a new agent with the task that you've given it. Now, I often edit these videos for time, but right now, as it is working, I'm not speeding this up at all. And hopefully you can appreciate how fast the model works. This is another one of their huge innovations, an ultra-fast model that's escapable as other cutting-edge models, but four times faster. They trained the model on real-world software engineering challenges and large code bases with access to tools that make it a lot more efficient. Now, as I scroll back through this, I can see that it's made a bunch of changes. You can also hit this Review button and it takes you through all the files in a single view. Another innovation is the ability to run a browser that collaborates with your development and prompting environment. It sounds like an obvious thing and it's a feature that I've often seen implemented in many vibe coding platforms. Let's go ahead and ask it to run that for us. I'll say let's fire up our development server and open up the web preview for me. And it opened us up in a new window. Now, you can also open it up from right here and you can pull it up in a new tab. I'm always running my machine on dark mode, so it did pick that up automatically. And I'm going to try light mode and then do a refresh to make sure that it's using local storage to remember the state. That's perfect. I like dark mode, so I'm going to switch back. You can also toggle the Chrome Development Tools. If you want to take a look at something like the Elements tab, take a look at Lighthouse or anything else. Finally, let's check out how Cursor now lets you run multiple agents in parallel. It's using a feature from Git called worktrees in a clever way to make sure different models can execute code with separate copies of the project. It's like running a contest to see which model can solve a problem the best way, and then you get to choose which one you like the best. I'm going to fire up a new agent and under this dropdown, I'm going to choose use multiple models. Now, I want to see how different models will try a challenge here. So I've got Composer 1, Sonnet 4.5, GPT-4 Codex and Grok Code. You can choose up to eight different concurrent models at one time, and I can even give it some additional context here. So I'm going to use this select element tool to choose one of my items, and I can even add a screenshot directly from the browser. I'm going to say I want you to add icons to these divs at the top of the page under the headline. To do this, Cursor is using Git worktree, which creates a copy of the current website and runs it independently. So while it is trying out these different models to run through the code, each model is working on its own environment. Let's go ahead and send. And you can monitor each of the runs as they work through the problem. Looks like Grok errored out, it didn't like images, so I'm going to have it do GPT-4.1. And we'll see how the other ones are doing. I'm going to ask it to go ahead and let me preview this and we'll take a look at this one in the new tab. That looks pretty good. This one's not quite as good. It didn't get the alignment right. And notice that sometimes when you ask it to do things, it will take a screenshot by itself to try to get better context on what it's doing, right? That's looking pretty good. So I'm going to go ahead and create a pull request here so this connects directly into my GitHub project and I'll go ahead and merge this. And now my project is done. Once again, Cursor is taking the lead in agentic development with these new features, the agent layout that brings some vibe coding approaches to professional development, a new ultra-fast model that's extremely capable, an integrated browser that feels like it's part of the development experience and multiple parallel agents for professional development.