Summiz Post

VS Code Is Fighting Back (RIP Cursor?)

Theo - t3․gg

Thumbnail image for VS Code Is Fighting Back (RIP Cursor?)
Post

VS Code is stepping up its AI game to rival Cursor, but still falls short in context and type safety, leaving users with mixed feelings.


You can also read:


The AI editor wars are heating up, with Cursor, a VS Code fork, leading the charge in AI editing. VS Code is stepping up its game, incorporating features similar to Cursor's in a recent update. Cod out is a notable AI tool for code review, helping with bug detection and summarizing pull requests.

Cursor aims to enhance AI editing significantly with UX improvements that couldn't be implemented in VS Code due to extension limitations. Its creators forked VS Code, made their own editor, and charged for it, resulting in a great editor experience.

VS Code's recent update copies many of Cursor's features, including attaching extra context to chat, generating tests using Co-pilot, and enhancing chat history with AI-generated names. Users can now pick a file and send it as context when asking Co-pilot for assistance, and generate tests for code by clicking on a function.

Experimental features in VS Code allow users to start typing in the editor and use the current line's content as a prompt for inline chat. Co-pilot can even detect when a user is prompting instead of writing code and automatically start an inline chat.

Let an AI help an AI. It's not going to be super slow and inaccurate. Not code rabbit AI; they can cut you a few times in half and find a whole bunch of bugs you might have missed. It can summarize PRs, draw diagrams, and even suggest changes that you can apply with one click.

Challenges with AI-Assisted Coding and Testing

The speaker expresses skepticism about AI-generated tests, arguing that human-written tests are more valuable because they assert human expectations. An example of AI-generated test code shows a test that mocks various functions and checks for specific calls, highlighting the verbosity of AI-generated tests.

Setting up tests presents challenges, starting with the need to export functions for external use. The speaker expresses frustration with Co-pilot's inability to handle certain imports and aliases, leading them to rely on Cursor for better results. Despite some trust issues, Cursor is preferred for fixing these problems.

Co-pilot struggles with context and type safety in code generation. It fails to generate the desired code, often ignoring context and producing incorrect functions. The speaker attempts to make Co-pilot download an image instead of a zip file, but it struggles with type safety and unchecked index access.

The introduction of code generation instructions in Co-pilot offers a glimmer of hope. The speaker sets instructions to ensure type safety, but Co-pilot continues to disappoint, ignoring the provided context and producing the same type errors. Despite these setbacks, the speaker remains curious about the potential of these tools, even if they currently fall short of expectations.

Creative Uses and Useful Features

The speaker discusses the fun of changing settings on a friend's VS Code to make it respond in Pig Latin or use inverse casing. While these features are amusing, they don't seem to have much practical use.

A useful link about improving test generation using GitHub Co-pilot is highlighted. The speaker emphasizes the ability to generate tests either by using the "generate tests" option or SL commands in the inline chat. They find the idea of SL commands intriguing, especially for triggering specific behaviors.

The potential of using Slash commands in inline chat for specific actions is exciting. Despite improvements, VS Code still lags behind Cursor, indicating more work is needed. The speaker finds more value in Cursor compared to VS Code's features.

Investment Biases and Final Thoughts

The speaker candidly discusses their investments in Microsoft and other companies, acknowledging potential biases. They emphasize that their investments are bets on potential success and that their opinions are not swayed by these investments. They believe the likelihood of their comments affecting the success of these investments is minimal.

Despite some impressive changes in VS Code, the speaker feels it still falls short compared to Cursor. They conclude that while VS Code's features are interesting, they will continue using Cursor for now. They invite the audience to share their thoughts on whether these updates are enough to switch from Cursor or if they remain skeptical about AI Dev tools.

Conclusion

Despite VS Code's updates, it still trails behind Cursor, with the speaker finding more value in Cursor's capabilities.


Want to get your own summary?