Agree and disagree. Yes, the bottleneck is going to still be human review. But with only a single Claude Code (or Cursor), I spent a lot of time (a couple minu

Agree and disagree. Yes, the bottleneck is going to still be human review. But with only a single Claude Code (or Cursor), I spent a lot of time (a couple minutes at a time) waiting for the agent to run through its loop completing what I asked it to do. In this time I found I was naturally switching context, but not always productively (checking slack, email, whatever). I called it "agent lag" time. Now with these worktrees tools, I can easily bounce between tasks I have the agents working on during these natural lulls. Overall I'm able to get more done and spend my time more efficiently. Also, Cursor has had the option to run multiple agents for a while (in tabs), but I always found it cumbersome because all the agents are operating on the same git checkout, so you have to take special care for them not to conflict with each other. Worktrees completely solve that problem.