Cursor follow-ups now queue to the next tool call instead of interrupting mid-action.

A message you send while the agent is working waits for the next tool call rather than cutting off the current one. (Cursor Changelog) Small change, real effect: correcting an agent no longer costs you a half-finished edit or an aborted test run, so course-correcting early stops being a gamble. I interrupt agents constantly and about a third of the time I regret it.