flo merian

Experimenting with Cursor

At @Bucket, we're always looking for products that can boost productivity and inspire us to build better software. We recently experimented with @Cursor to replace @Github Copilot.


Key takeaways

  • Tab completion: Cursor includes a helpful tab completion feature, suggesting refactorings and improvements as you go — huge time saver!

  • Chat: the chat/composer is better than Copilot as it can modify across files in the codebase. However, we still needed to correct the code most of the time :/

  • Full context: Cursor has full context, so it provides more helpful advice and changes

  • Generating comments: it makes it easier to add code comments — they need clarifications from time to time tho

  • Working with existing codebases: Cursor seems better at writing code from scratch than contributing to large existing codebases

  • Hallucinations: as far as we can tell, it had fewer hallucinations compared to alternatives

  • Prompting strategies: being somewhat specific in prompts by referencing classes and functions seems to work well — We're currently experimenting with `.cursorrules` files to improve prompt quality

How about you? What's your experience with Cursor as a team?

Add a comment

Replies

Best
Ambika Vaish

Really appreciate this breakdown; especially the bit about Cursor being better at writing from scratch vs working with legacy code. We've felt something similar on our end too.

The ".cursorrules" experiment sounds interesting would love to hear how that plays out over time.

Curious: how are you deciding when to trust the suggestions vs rechecking everything manually?

flo merian

@ambika_vaish good question, Ambika! i'd mostly use @Cursor for quick code generation and refactoring. for quality code generation, some team members at @Bucket switch to @JetBrains from time to time.

Dima Havryliuk

Hey Flo, I see you already mentioned the Composer, so you should be familiar with how cursor inserts the generated code. But if you haven't seen/trusted the Agent mode in the latest Cursor version - you should definitely check it out.


If you switch from composer to Agent, Cursor will auto-update all the files with the generated code and apply the changes. This is super efficient. Also you can connect the new Model Context Protocols in Cursor settings and it will be able to call a lot of different tools like pushing to github, have more advanced browser capabilities, etc.


Also, you should try asking Cursor to do more comprehensive tasks like creating the tables, and API gateways for the functions it just generated. It now can write cloudformation scripts, which will setup the backend for you entirely. Blew my mind when I first used it.

flo merian

@dima_havryliuk wow! thanks for the suggestions, Dima! will definitely give it a spin.


re: MCP. absolutely! fun fact: we started experimenting with @Cursor the moment we decided to build an MCP server for @Bucket #dogfooding. really looking forward to the opportunities it opens up here.

Jake Crump

This is a really cool write-up on trying out Cursor for the first time. I'm curious, you mentioned that this was an experiment to see if you wanted to move off of Copilot. Would you say that after this, you would want to switch fully over to Cursor?

flo merian

@jakecrump definitely. what started as an experiment — exploring @Cursor's features to build an MCP server — ended up being a love story.


we now advocate to go on the Business plan. looking at you, @roncohen 😹

Jake Crump

@roncohen  @fmerian Very cool! Another one to maybe check out is @Windsurf I know our engineering team loves it and moved over after using @Cursor for a bit.

flo merian

@roncohen  @jakecrump oh yes! @steveb recently suggested we give @Windsurf a spin!

did the eng team face any issues migrating from one editor to another?

steve beyatte

Good takeways. Were you writing code the old fashioned way before trying Cursor? Do you think you'll use Cursor in your dev workflow moving forward? Check out @Augment Code and @Windsurf for two other options that I like better than Cursor.

flo merian

@steveb thanks, Steve! we were using @Github Copilot before experimenting with @Cursor and yes — we're seriously considering upgrading to a Business account!


re: @Augment Code and @Windsurf — what make them better alternatives from your perspective?

Rajiv Ayyangar

@steveb  @fmerian  @ken_miller4 Might be able to speak to augment. From what I hear, windsurf and augment are better with large codebases and have a better agentic mode.

flo merian

@steveb  @ken_miller4  @rajiv_ayyangar interesting! thanks for the feedback, I just dropped a message to the team for further experimentation.


any experience with @Jolt AI? the product description suggests it's "built for 100k to multi-million line codebases."

Gabe Perez

Man, we need more Key Takeaway threads here for Product Comparisons. This was great - I've always been curious to use @Github Copilot but am stuck on Cursor.

I'm curious to learn more about your prompting strategies. I've been avoiding using cursorurles (idk why) and have been treating Cursor like a junior engineer and myself as a PM and it seems to be going well.

and UGH on chat - I totally feel this. What I try to do now is only modify cross-code if it's written in the same language or very simple. As soon as the code gets more complex, I try to do only one file at a time and a function or two at the time. not perfect but helps.

Ashish Kumar

The generative features are nice, but when it comes to collaborating on existing projects, I agree that it still needs a little more polish