One thread per piece of work.
Start a conversation from Chat for each job. Your agent replies in the same thread: it acknowledges longer jobs with its plan, posts at real milestones, and tells you what it did and how to try it when it's done.
While it's still busy after replying, the thread says so, with a link to watch live.
Decisions come with options to tap.
When your agent needs a decision it asks in the thread with a few answers to tap, its recommendation first. Type your own if none fit.
If you keep picking the same kind of answer, it stops asking and just does it.
Your message comes first.
Send a message any time. Your agent reads new messages between steps and answers you before carrying on, or changes course if you ask it to.
The Nudge bell decides how urgently a busy agent sees it: on, it gets your message before its next step; off, it reads it at its next inbox check, like email.
Watch every command as it runs.
Logs shows exactly what your agent is doing: commands, their output, files it writes and its replies. You can type to it there too, or press Stop to halt the current step.
The full history stays in the Logs, even after the agent has summarised older work to free up its memory.
Let it keep going for a while.
Turn on Work mode in the header to have your agent keep working on its own for a set time, say an hour, picking its next steps from the brief and notes. It stops when the time is up or when you message it.