I left Cursor for VSCode
— but soon returned
July 2025
Cursor is still (as of July 2025) superior to VSCode for AI-assisted coding. Compare the videos below to see how autocomplete is so much nicer in Cursor.
Context
I'm a software engineer with 10 YoE, and I've been happily using Cursor since Fall 2024, but given the recent pricing outrage I decided it was time to give VSCode another chance.
I switched to VSCode Insiders (as it supposedly has better AI support) + Traycer for a week - which was long enough to realise that Cursor is still ahead of the game!
Why Cursor is better
Chat-wise VSCode is fine, but what killed me is how far behind autocomplete is. And autocomplete is where AI in coding shines in my opinion, as it really boost productivity and avoids the small tedious tasks that adds up.
- Copilot in VSCode did not work well when editing across files (it feels like it does not have context of changes made between different files),
- the suggestions were more often incorrect (despite clear types from TypeScript) and
- in general it felt much more sluggish and slow when coming from Cursor.
And here's the proof
This small and somewhat trivial refactor clearly shows the difference. With Cursor the same task took me half the time compared to VSCode.
Setup:
- I renamed a function and changed the signature slightly in one file.
- In another file that imports the function, I update the usage.
Cursor
- It took me 30 seconds to complete the task
- It instantly knew what the new function was named and suggested replacing it
- It instantly knew that the types had changed and suggested updating usage to match the new type
Cursor in action
VSCode/Copilot
- It took me 67 seconds to complete the task
- VSCode did not suggest renaming the function or import
- VSCode insisted on using the outdated function signature
- I had to manually click on a line before autocomplete started showing suggestions
VSCode in action
So in conclusion, I'm back to Cursor.
I have dabbled with Gemini CLI and Claude Code, but I prefer writing code (with assistance) over writing prompts (and I rarely get great results for complex tasks anyway, but I guess that's because I'm not using that one magical prompt that makes everything great).
Tell me why I'm wrong
There's probably a hundred reasons why you believe this is all my own fault, how I missed something obvious or why I should buy your course on prompt engineering. Except for the last one, I'd genuinely like to hear them! Ping me on BlueSky and tell me if you have a different experience.