As a developer, I’ve been loving the quickness that cursor adds to certain workflows, but my trust in AI is also not super high. Many times there are VSCode shortcuts that do a ton of what AI would do just by plugging in a value for it, except it reliably works across files, and has many less edge cases for me to be concerned about as the person reviewing the changes. A classic example would be renaming a symbol. If cursor prompted me if I wanted to rename something to something else, and I said yes, then I would feel much more confident knowing that the refactor wasn’t some LLM guess but rather just a reliable LSP based command that the LLM prompted for.
Would love to see prompts that pop up and say:
“Would you like me to rename sectionShowHeader to showHeader across all files?”,
perhaps even a different color, or maybe providing a list of the VSCode commands it intends to run could be a very useful interface for telling the developer what type of editing it intends to do.
Here is a loom video of the use-case that gave me the idea:
And also here’s a little drawing for a potential UI:
I could imagine this being useful for things like: renames, imports, extracting files etc… and could theoretically be a good way for plugin interop that doesn’t require cursor specific plugins, but rather VSCode commands that cursor can interact with.
3 Likes
Furthermore, I think that if something like this were to be implemented, it would also be super super cool to have just different types of Tab Prompts in general.
For example, right now tab prompts are all just text completion, but I could imagine tab prompts could be:
- completion prompts
- code action prompts such as described above
- interaction prompts, maybe a prompt could ask the user a clarifying question
- special prompts, crazy way down the line example, but imagine you could have some wild interactive things that pop up, such as a visual editor when you go to style a div that poops out tailwind code and shows you what your building or something like that.
The value proposition here is that as developers we rightfully distrust AI and have to be very guarded around accepting changes. Different types of interactions give us different types of confidence though. I imagine that developing a mental mapping of maybe color coded prompts and how much I trust the different types allows me to breeze over the types of refactoring I do trust and slow down and be careful on the types of refactoring that I don’t.
Nice video! Explains the idea well.
What if in addition, AI were integrated into VSCode’s existing Search & Replace UI? In addition to case-sensitive, word-match, and regex, what if there were AI options? I’d like that because I could verify the changes before they’re applied.
1 Like
That would be very cool, I know I would use it all the time, but to me seems like a totally different feature @mosofsky you should make a feature request for that, I’d upvote it in a heartbeat
