Should you switch from VSCode to Cursor?
Should you switch from VSCode to Cursor?
Introduction
As developers, we're constantly searching for tools to enhance our productivity and make coding more enjoyable. I have been using Visual Studio Code (VSCode) for over six years, it has been the go-to integrated development environment (IDE) for almost all the developers I have worked with.
In 2023, Cursor (designed by the Anysphere research lab) raised $8M from OpenAI and $11M in total funding as they delivered the message:
In the next few years, we'd like to build a code editor that is more helpful, delightful, and fun than the world has ever seen.
In 2024, many developers began testing or fully transitioning to using Cursor as their primary IDE of choice.
The purpose of this article is to give you a first-person perspective on my experience using Cursor and how I compare it to VSCode. Specifically, I will be discussing how each IDE performs for Data Science when using their respective AI assistance capabilities.
What is Cursor?
Originally forked from Microsoft's VSCode GitHub repository, the Anysphere team took advantage of its open-source codebase and decided it would be a great foundation for developing Cursor.
Cursor represents a bridge between current development practices and the future of collaborative human-AI programming. Cursor's mission is to develop an IDE heavily integrated with artificial intelligence (AI) capable of being an order of magnitude more effective than any one programmer.
By preserving VSCode's familiar interface and comprehensive extension ecosystem, Cursor lowers developers' adoption barriers, making the transition to an AI-powered IDE feel like a natural, seamless upgrade rather than a disruptive technological change.
My Previous VSCode Experience
I have been using VSCode for over 6 years and was a beta tester for GitHub Copilot which I have been paying for since its release.
From an interface perspective, VSCode is great. It is customisable, but even its default layout provides you with everything you need when working on a project (I can understand why Anysphere kept it this way). VSCode also has a great extension ecosystem, again, Anysphere acknowledged this and decided to keep this as is when developing Cursor.
On October 29th 2024, Claude 3.5 Sonnet was made available on GitHub Copilot. Before this, your only LLM option was OpenAI models due to the partnership between OpenAI and Microsoft. This wasn't terrible, but seeing alternative LLMs perform better on coding-related benchmarks made it frustrating as they were not available on GitHub Copilot.
As a Data Scientist, most of my exploratory work is completed using Jupyter Notebooks. I found VSCode (and GitHub Copilot) to work really well with Jupyter Notebooks, as shown below, each time a new cell is created you have the option to prompt GitHub Copilot for assistance.
VSCode also enables you to chat with GitHub Copilot about your project. Using shortcuts such as /explain, /fix, and /tests, you can prompt GitHub Copilot to assist at a vscode, workspace, or terminal level. Using the "#" symbol, you can prompt GitHub Copilot to look specifically at your code selection, codebase, editor, or file.
All of these shortcuts make using GitHub Copilot with Jupyter Notebooks (and alternative file types) in VSCode really good. The introduction of non-OpenAI LLMs is a great addition although still in its infancy.
For the purpose of this article, I am highlighting high-level points from my experience with VSCode and GitHub Copilot. Feel free to message me via social media if you have more specific questions you would like to ask.
What Made Me Switch?
Judging by "My Previous VSCode Experience", one might ask why I decided to download Cursor as my VSCode experience does not sound too bad.
The primary reasoning for this was due to the growing number of developer advocates I was speaking to who had made the switch themselves. Also around this time, Aman Sanger, Arvid Lunnemark, Michael Truell, and Sualeh Asif appeared on the Lex Fridman Podcast to discuss Cursor in detail. I listened for the entire 2 hours and 29 minutes and liked what I heard.
Finally, Cursor is free to install and comes with a free tier that allows you to get a feel for what it is capable of. I had to explore Cursor for myself after probably pushing it to one side for far too long.
My Experience Using Cursor
My first impression of Cursor was that I was extremely glad they kept the VSCode interface whilst also prompting you to download all of your VSCode extensions straight away. Before writing any code (or prompting an LLM), you feel right at home.
When opening Cursor for the first time, you are presented with a welcome checklist that appears above the explore sidebar.
The purpose of the checklist is to introduce you to some of Cursor's primary features, this was a nice touch as it introduced you to the shortcuts needed to perform each task using the IDE.
After onboarding (downloading extensions, etc.) I decided to import a side project I had been working on recently. It was at this point I realised Cursor does not work as well with Jupyter Notebooks as VSCode.
As shown above, there is no "Generate" tab available in Cursor. You can still use Ctrl/⌘ + K to prompt an LLM but the experience is not as good as VSCode.
Once you have your code snippet, you can use Ctrl/⌘ + L to chat with an LLM, unfortunately, when using Jupyter Notebooks in Cursor the apply feature does not work when accepting LLM-recommended code snippets, I found this to be very disappointing.
At this point, I decided to create a new Python project. I used .py files only and picked a topic that I had limited experience in and would likely take me several hours to develop an MVP without using AI assistance. I decided to select a game development project created using pygame.
Cursor has a feature called "Composer" which allows developers to generate an entire project structure by simply describing their project concept in a prompt. Instead of manually setting up files, configuring environments, and writing boilerplate code, developers can leverage AI to streamline the initial project setup.
By implementing Composer, you can specify key project information such as:
- Project concept or idea
- Preferred programming languages
- Testing requirements
- Desired documentation (like README.md)
- Virtual environment configuration (requirements.txt)
I wanted to test this feature by applying it to my pygame project idea. The initial prompt I used was:
Create a tetris game using pygame, add tests, and add all the instructions in a README.md to get the game running. Create the correct folder structure, splitting our src, tests etc, and create all the python scripts following the project structure
With only minor changes required, the results were outstanding! The project folder structure was perfect, tests were created successfully, and a very well-documented README.md file was generated explaining to the user how to initiate a game of Tetris.
All of this was achieved within minutes with some additional time taken to fix minor bugs and redesign the layout of the Tetris board.
For more information on the features Cursor offers, see here.
Conclusion
After using Cursor for the past two weeks, I have decided to use it for all of my future projects.
The experience I had using Cursor's Composer feature was incomparable to anything else I have experienced in my professional career as a Data Scientist. I see this as a game changer and wouldn't be surprised if there is a team at Microsoft working on their own version to add to GitHub Copilots offerings.
With that said, I am not completely turning my back on Vscode. Within recent weeks, there have been some updates to GitHub Copilot showcasing an intent to keep up with competition.
As Cursor is a fork of VSCode, moving back and forth between IDEs will not require any additional changes to your codebase or IDE configuration. My decision between which tool to use will be heavily decided by what features they offer, how those features help me with the task at hand, and finally, the price of accessing these tools.
Just a final point on price, GitHub Copilot currently costs 10 dollars per month, half the price of Cursor′s Pro tier which costs 20 dollars per month. Each offering does not require an agreed contractual period, should you choose to, you could alternate month-by-month between the two tools at no extra cost to yourself.
Disclaimer: All opinions shared in this article are my own based on my personal experience, I have no affiliation with Cursor. Furthermore, unless stated otherwise, all images included in this article are owned by the author.
If you enjoyed reading this article, please follow me on Medium, X, and GitHub for similar content relating to Data Science, Artificial Intelligence, and Engineering.
Happy learning!
