I agree that Cursor is quite good! You raise a great point about needing multi-agents that can work independently. I've been wanting exactly the same thing.. imagine having agents tackling all those failed unit tests simultaneously instead of the sequential fixing we're stuck with now.