Transcription
What's up everyone? How's it going? Welcome back with me, Nathan.
So, a few days ago, Cursor just released two big updates to their IDE. And in this video, I'm going to walk you through this exciting changes.
All right, so recently, Cursor just released version 2.0 as well as their own AI coding model called Composer. In Cursor 2.0, there is a new interface that you can use to work with many agents in parallel. We'll cover the ID update in a moment, but for now, let's take a look at the composer model.
According to the team, this is a frontier model that is four times faster than similar intelligent models. The model is designed for low latency agent decoding and cursor, and it can complete most turns in under 30 seconds. Early testers reported that the ability to iterate quickly with the model was delightful and very much trustworthy for multi-step coding tasks. Composer was trained with a set of powerful tools, including one for semantic search, making it much better at understanding and working in large code bases.
Now, here's a demo of composer in action. In this example, it was given a task to investigate an issue where users were seeing panic attacks related to the event bus in version 1.9 release. This issue seems to be taken from this tail scale GitHub repository. So, the composer model takes on the task. It will fetch the context from the codebase reviewing the even bus code as well as understanding the patterns. It will also check on previous comets to see if there are any related changes and find possible solutions. Once the agent understands the context, it will work on planning a solution, creating a step-by-step list of to-dos for a structured process. After that, it will execute the tasks one at a time, analyzing the panic issue, implementing the fix and verifying that the fix is handling all possible scenarios. All of this is completed under 30 seconds from fetching the context to execution and then to testing. Composer can handle the entire process seamlessly, ensuring that the fix is properly implemented before reporting the summary back to you.
Next, let's talk about the benchmark score. On the screen here, you can see that the composer model is designed for software engineering intelligence and speed. On the benchmarks, the model achieves quality coding results with generation speed about four times faster than similar models. While the intelligence score is a bit behind the top tier models like Sonet 4.5 or Gemini 2.5, it makes up for that with impressive speed and most likely a lower cost.
Under the hood, composer is a mixture of experts model that's optimized for supporting long context generation and understanding. It is specialized for software engineering through reinforcement learning in a diverse range of development environments. In addition, the model is also trained to be a helpful assistant by minimizing unnecessary responses and claims made without evidence. They also find that during reinforcement learning, the model learns useful behaviors on its own, like performing complex searches, fixing llinter errors, and writing and executing unit tasks.
And that will be all for composer. Next, we will explore the upgraded cursor IDE.
Now, before we jump into the exciting part, don't forget to subscribe to my YouTube channel so you won't miss new videos that can really help you level up your skills. Also, don't forget to click on the bell to get notified whenever a new video comes out. Thanks a lot for supporting the channel. You really help me make useful videos just like this one.
All right, the newly upgraded cursor now has a multi- aent interface that's designed around agents instead of files. It supports running multiple agents in parallel by making use of G work trees or remote machines. Each agent is working in an isolated environment, so they won't interfere with each other's work. When you encounter a difficult task, you can have multiple models attempt the same problem so that you can get the best result. And when you want to dive deep into the code yourself, you still have access to the classic IDE layout with a single click.
This upgrade also improves the classic IDE, making it much easier to quickly review the changes an agent has made and dive deeper into the code when you need to. For example, you can easily run the same prompt across multiple models to compare their outputs and combine results for maximum intelligence and accuracy. You can see on the screen here that a prom was sent into the set 4.5 GBD5 fast as well as the composer one model. And all three of the models are working on tackling this test.
And then here's the new built-in browser. Uh this is where agents can now run and test their code directly making the iteration process super fast, smarter and fully autonomous. In this case, you can test an entire website automatically. It can handle everything from interacting with the application to validating results and it is definitely going to save massive amount of time and effort and all of these is being done directly within the browser interface in this upgraded cursor IDE.
And then uh this is a small release, but you can now turn speech into code with the voice mode. So you can provide the prompt by speaking instead of typing if that's what you prefer.
With the 2.0 release, there's also an improved code review process as cursor now has a unified interface where you can view all changes made by the agent across multiple files. That means you no longer have to jump between individual files to review the code.
And then there is sandbox terminals which is generally available for MacOSS. Basically agent commands now run in a secure sandbox by default and shell commands that are not in the allow list will automatically run in this sandbox. It will have read and write access to your workspace but not internet access.
For teams, there is a new team command feature where you can define custom commands and rules for your team in the cursor dashboard. These commands are automatically applied to all team members without needing to store the files locally. This is very convenient for team members.
Next, we have improved performance as cursor now leverages language server protocol or LSP for language specific features like goto definition hover tool tips diagnostics and much more. They also have drastically improved the performance of loading and using lsps for all languages. This is particularly noticeable when working with agents and viewing divs. Python and TypeScript LSPs are now faster by default for large projects with higher memory limits. The LSPs are also dynamically configured based on the amount of RAM you have available. Finally, a number of memory leaks have been fixed improving overall memory usage.
Next, we have plan mode in the background. So, this feature allows you to create your plan with one model and then build it with another. You can choose to build a plan in the foreground or background or use parallel agents to generate multiple plans that you can review. This is definitely a great new feature that will help you efficiently build as well as prototype with multiple agents.
And then there is also sharable team comments. Uh this is where you can share custom rules, comments, and prompts with your entire team. You can also create deep links through the cursor docs for easy access and distribution.
If you like this video and would love to support the channel, you can consider donating to my channel through the super thanks option below. Or you can consider joining my YouTube membership where you can use this channel's emojis, get early access to new videos, plus a lot more.
And that will be all for this Cursor 2.0 update. They also have smaller bug fixes and patches applied on this update. And if you want to see all of the changes, you can visit the change log provided by Cursor. I will leave a link to this in the description below.
Overall, the update looks solid, especially the new agent interface, which allows you to work with up to eight agents at the same time. If you want to try out this free update for yourself, then you can simply head over to Cursor's website at cursor.com, which I'll leave a link in the description as well, and just download the IDE, register for a free account, and you're good to go.
And that brings us to the end of this video. So, what do you think about Cursor 2.0? I have to say it seems really interesting and I'm going to try it out for a few of my projects. I hope you all enjoy today's video and get some value out of it. Let me know your thoughts in the comments below. I'll join the conversation and reply as often as I can.
If you're new to the channel, my name is Nathan and I help you build profitable apps and projects using AI and other tools. Make sure to subscribe if that's something you find useful. Don't forget to like this video, turn on the notification bell, all that good stuff as it really helps the channel to grow. With that being said, thanks so much for watching until the end. I hope you have a great day and I'll see you in other videos. Bye-bye. [Music]