How many coding agents do you run in parallel - and what breaks first?
I'm Roman, maker of Wrangler. Before building it I was running five Claude Code / Codex sessions in parallel, and the thing that broke first wasn't the code - it was me: eleven tabs, an agent silently waiting 40 minutes for an approval, and no honest answer to "is this feature actually done?"
We launch here on Wednesday, and before that I'm genuinely curious where everyone else is on this curve:
How many agent sessions do you actually run at once?
How do you notice one is stuck waiting on you?
Do you trust the agent when it says "done" - or do you re-run the checks yourself?
I'll share what we learned building the control tower (worktree isolation, running the repo's own tests in the agent's worktree before anything goes green, an interrupt queue for blocked sessions) in the comments. War stories welcome - the messier the better.
Replies