How do you juggle two live conversations? Every extra one costs a terminal tab

[AMrainman]

I keep hitting the same friction: a long turn is running in one conversation, I want to re-read a previous conversation, answer the agent there, and then come back. In pi that means either killing the running turn (any switch tears the session down through teardownCurrent()), or opening another pi in a second terminal tab.

I have been doing the second, and the tab cost is worse than it sounds:

• Screen space is zero-sum. Two 50/50 panes already truncate diffs and long tool output, and in a quake-style half-screen terminal a split leaves a handful of usable lines.
• Panes do not reflow. A conversation that suddenly dumps 500 lines cannot borrow room from an idle one, so I keep rebuilding layouts.
• Tabs hide state. No way to see which conversation is streaming, waiting on a tool approval, or finished.

I filed #11829 (detach instead of teardown: stop rendering a session, keep its runtime and turn alive, replay buffered events on re-attach) and left a note on #4846. I am not expecting a core change soon, so I would rather learn how everyone else handles this:

• do you run one pi per project, or one per conversation?
• tmux / Windows Terminal layouts? Named sessions, fixed pane grids?
• do you push the parallel work into subagents instead of parallel conversations?
• anyone driving --mode rpc workers from their own multiplexer?

Curious what actually holds up over a full day.