@ba
I recently switched back to using cursor after swearing by claude code for months
Using a combo of
- cursor's new composer model
- subagent git worktrees
- voice dictation with wispr flow
I'm now able to work way faster with multiple agents at once without any noticeable difference in code quality. It's crazy how much of a productivity unlock this has been.
Also important to note that all agent code is sandboxed by default and has a strict allowlist on both native + mcp tools to prevent any unwanted db writes or pushes to git. I now just let it run wild knowing it can't break everything on me.