See agents think as they work
Every agent now shows its reasoning as it works and each step it takes, in Work, automations, Canvas and the Agents view, and a finished run folds away.
The current step, live
While an agent works you see what it is thinking and the step it is on, with the commands and reads that step ran. Earlier steps sit behind a disclosure, and when the run finishes every step folds into one line you can open again. Each command or read opens its exact call and output in the terminal pane.
The same view appears wherever an agent runs: a chat in Work, an automation's conversation, a Canvas session and each helper in the Agents view.
Helpers keep their own thread
When an agent hands work to a helper, the helper's thinking stays in the helper's own activity instead of mixing into the main answer. Reopening the app in the middle of a run shows each helper's current status, not the one it had when it started, and a run that ended earlier no longer shows steps still spinning.