Windsurf at the moment. It now can run multiple "flows" in parallel, so I can set one cascade off to look into a bug somewhere while another cascade implements a feature elswhere in the code base. The LLMs spit out their tokens in the background, I drop in eventually to reveiew and accept or ask for further changes.
Until you change model in one of the tabs and all other tabs (and editor instances!) get model changed, stop what they're doing, lose context etc. There is also a bug where if you have two editors working on two codebases they get lost and start working on same thing, I suppose there is some kind of a background workspace that gets mixed up.
