Agents
The agent is just another member.
It connects the way a person does — same membership, same permissions, same presence. Which sounds like a small architectural choice until you watch one work.
Every step is visible
Tool calls, edits, and reasoning stream into the room as they happen, not as a summary afterward.
When an agent gets something wrong, you can find the exact step where it went wrong. That's the difference between trusting a system and hoping about one.
You can grab the wheel
A running job is live state in the room. Interrupt it, redirect it at minute six — instead of reading the postmortem at minute twenty.
Long runs are where agents drift, and the fix is being able to correct one mid-flight rather than starting it over.
No special bot lane
Agents drive rooms and activities through a bridge, under the same membership and permission checks as a person.
An agent can't see a room it isn't in. There is no elevated integration path around the rules everyone else follows.
It sees what you see
Open a file and the agent has it too, with the schema and the surrounding conversation.
It edits in place
Changes land in the document on your screen, live, instead of arriving as a block of text.
It knows when to speak
Agents answer when they have something to add and stay quiet when they don’t.