[OLD] GitHub Copilot v Cursor?

Discussion
85

Could you elaborate please, like is cursor more context aware at the entire codebase level (like x proposed modification will affect y number of files in the codebase?) compared to Github Copilot?

87

I think Copilot can only go so far being a VSCode extension. Cursor controls the entire IDE and can branch off into its own IDE. Im not particularly experienced in how it works for extensions and forks of VSCode etcā€¦ but thats my thoughts?

88

MS own the VSCode editor and copilot. Copilot is quite heavily integrated into the editor at this point. Cursor have an advantage here over all other AI extensions, not Copilot.

1 Like
89

Yeah I donā€™t really get this either yet. I havenā€™t seen Cursor do anything radically different than any extension. It seems like it could have easily been an extension.

90

I think @claidler perfectly put it. we want to push the boundary of whats possible. and how fast you can write code. to do that we have to make good models but also a different user experience.

2 Likes
91

I think Continue extension is closest to Cursor at least superficially. The worry here might be that instead of developing/maintaining/supporting an extension to VSCode you have to provide the same to the whole IDE fork with expectations that the same 3rd party extensions work with Cursor as with VSCode.

Also, people expect that Cursor keeps up with the VSCode when useful stuff is released upstream. IMO that has worked out pretty well so far, but how much of your efforts need to be dedicated to the merging process is unknown to us. In other words, does it cause periodical disturbances that halts the other development etc.

Of course you know the tradeoff the best and what is your vision of Cursor becoming in the future and I trust that, but still wanted to express my thoughts on this topic.

92

Iā€™d been using Copilot for almost a year before switching to Cursor. Almost each time I tried to use inline edits with Copilot it messed up the whole code when applying the changes. The primary reason I switched to Cursor was that it was able to apply its suggestions correctly. Iā€™m wondering if Copilot has improved since then.

Also Copilotā€™s inline suggestions are only able to add code to the right of your cursorā€™s position (correct me if it has changed since then), but with Cursorā€™s Copilot++ you are able to apply edits backwards/replace the whole line before and after your cursorā€™s position + it allows you to jump to another line with Tab and apply the same change there if it detects itā€™s required.

2 Likes
94

I agree with this sentiment :slight_smile: part of our work is making sure that its not only nice UX but the features actually work.

95

I prefer Cursor.

  1. Better experience.
  2. Choice of LLM models
  3. Flexible symbol functionality
  4. Interpreter Mode
  5. Vision support (multimodal)
  6. /edit command
  7. Terminal āŒ˜+K
  8. Auto Debug / Fix with AI
  9. Rules for AI
  10. .cursorrules
  11. Non-programming related questions support
6 Likes
96

As our company was evaluating Enterprise-level subscriptions for AI code assistants, I had the opportunity to test both GitHub Copilot and Cursor extensively. Here are my findings and recommendations:

Key Differences

  1. AI Model: Cursor leverages the more advanced GPT-4 model, while Copilot uses an earlier version of GPT-3 (Codex). This gives Cursor an edge in code understanding and generation quality.
  2. Codebase Context: Cursor has superior codebase context awareness. It can quickly analyze multiple full files and retrieve the most relevant code snippets from the entire codebase. In contrast, Copilot only sees the code currently visible in your editor window.
  3. IDE Integration: Cursor is a fork of VS Code with deep AI integration, offering features like auto-debugging and AI-powered error fixing. Copilot is available as an extension for various IDEs but lacks some of these advanced IDE-level integrations.
  4. Chat Interface: Both tools offer chat interfaces, but Cursorā€™s chat is more intuitive and supports richer interactions like clicking on file names to navigate to the code.

Strengths of Cursor

  • Leverages the more capable GPT-4 model for better code quality
  • Excels at understanding and generating code based on the full codebase context
  • Offers unique IDE-level features like auto-debugging and AI-assisted error fixing
  • Provides a more intuitive and feature-rich chat interface

Strengths of Copilot

  • Backed by Microsoft and GitHub, ensuring long-term support and integration
  • Offers quick, inline code completions that run continuously, which Cursor currently lacks
  • Has a larger user base and more feedback data to improve its models

Recommendation

For our enterprise use case, I recommend adopting Cursor due to its superior AI model, better codebase understanding, and advanced IDE features. However, Copilotā€™s inline completions are also valuable for productivity.Ideally, using both tools together yields the best results - Copilot for quick code completions and Cursor with GPT-4 for more complex tasks that require full codebase context. You can refer to my detailed comparison published on our companyā€™s LinkedIn page: Comparison of using coding assistants with Microsoft Copilot | Kanaka Software posted on the topic | LinkedIn

6 Likes
97

You can try Codeium for that (free) and itā€™s better than Github Copilot at it. I use Codeium + Cursor

2 Likes
98

Cursor also has its own version of inline completions. Weā€™ve found it to be much better than copilot!

5 Likes