Collaboration
Colleagues, not chatbots.
An agent in Stateroom is not a window you negotiate with. It holds a cursor of its own, reads what it is scoped to see, and writes into the page beside yours — pausing, resuming, and taking direction the way a person would.
01 / Presence
You always know who is writing.
Every author in the room — human or otherwise — appears the same way: a named cursor moving through live text. Agents do not get a special mode or a special box. They get a seat.
Their replies stream into the document as ordinary paragraphs. While the words are still arriving you can already select them, question them, comment on them, or delete them — because streaming text is just text.
02 / Direction
Steering is a gesture, not a command line.
- Place its attention.
Select a region and the agent reads that region — your brief stays deliberate rather than everything-it-can-reach.
- Choose where it writes.
Its next contribution lands wherever its anchor sits, and that anchor slides gracefully when your own edits move the ground under it.
- Pause like you'd pause a person.
Suspended work holds its place. Resume, and it picks up exactly where things stand now — including whatever you changed meanwhile.
03 / Discipline
Everything replayable, so behavior is provable.
Each decision an agent makes is captured whole: what it saw, what it answered, which path it took. Replay the same history and you get the same work — byte for byte. That discipline is why concurrent editing here can be tested to destruction instead of hoped for.
One honest note: while you type faster than a colleague reads, its next sentence can momentarily trail your latest edit. The document merges everyone's strokes perfectly; meaning catches up on the next read. It is the same grace period two fast humans give each other — kept small, and visible when it happens.