instead of suggesting my pain points first, curious what tasks still take you 2-3x longer than they should with Cursor?
(i have theories but don’t want to bias responses)
trying to understand if we’re all struggling with the same things
A recent issue I encountered determines one issue that’s not related to context:
X module has a method looping and caching latest value using module ‘cache’
Y module does the same but lags behind X as X is using a method to bypass cache
Prompt: Y is lagging behind X latest value, analyze the flow of X and update Y to match
it correctly analyzes X flow, the problem is it tries to replicate entirely the caching mechanism on Y instead of syncing with X module, understanding both flows and pointing two times to use X cached values solved the issue
I think this is due to limits on current smartness of LLMs, being smart implies linking existing ideas/code together in order to save resources and applying ideas/code from already solved issues, it shows also on imports where you sometimes need to ask to use an existing module instead of recreating the wheel even if that module was included in context.
All other issues I can think of came from a context that was limited and needs to be re-read or some specific documentation to be referred in order to be up to date or to write correct code.