Cloud agents
Sharing the agent, not the transcript
There are two different things people mean by sharing, and only one of them survives contact with a second person.
The short answer
Sharing an AI agent with teammates means giving them a way to assign it work and inspect its output without going through you. Copying a transcript does not do that. Polaris shares the worker itself: one roster row anyone can assign to, one queue that accepts work from any member, deliverables posted as comments with attached files, and an append-only record of who asked for what.
- What gets shared
- The worker, not the session
- Deliverable format
- Comment, file, or Docs page
- Re-brief method
- Comment on the task
Two definitions of sharing
Sharing a transcript
- A snapshot, already stale when pasted
- Your colleague cannot ask it a follow-up
- The brief that produced it is not attached
- Files arrive as uploads with no version history
- You remain the only route to the agent
Sharing the worker
- A roster entry any member can assign a task to
- A colleague comments and the worker wakes again
- The brief, checklist and comments live on the task
- Files carry versions, with the old one preserved
- The worker answers to the team, not to you
How a teammate uses a worker you configured
No handover, no credentials, no explanation of your prompt.
- 1
They open a task and pick the worker
The worker appears in the same owner picker as every human. Assignment writes a queue row through the same trigger, whoever did the assigning.
- 2
They write acceptance criteria
Checklist items on the task are handed to the worker as its acceptance criteria, so the brief is structured rather than folkloric.
- 3
They watch the session or walk away
The per-step event stream renders live for anyone with access to the task. Nobody needs to keep it open for the work to finish.
- 4
They read the delivery in place
The result is a comment authored by the worker, plus any files attached to the task and any pages drafted into the shared Docs tree.
- 5
They push back in the thread
A comment from anyone other than the worker itself queues a follow-up session, with the latest human comment treated as the new brief.
What travels with the work
Everything a second person needs in order to trust the output.
The instructions the worker ran under
Instructions and attached skill documents are org-visible records, not a prompt hidden in someone's shell history.
The searches it ran
Each web search is written to the session event stream with the query text, so a reviewer can see what it looked at rather than inferring it.
The acceptance criteria it ticked
Ticked items record the worker as the owner of the tick, which makes an unticked box a visible gap rather than a silent one.
The effort counters behind the bill
Searches, characters written, items ticked, comments addressed and files produced are stored on the job row next to the estimate they generated.
Related reading
Turning a single-player agent into a team member
The change is not a better prompt. It is putting the worker in the same table as the people.
Agents deliver, humans close
One rule holds the whole product together, and it is a rule about who is allowed to say finished.
When the deliverable is a file
A chat reply is not a deliverable if the thing you needed was a document somebody can open.
Delivery comment
Work arrives where the task already lives, attributed and reviewable, and the task stays open.
Work log
The record that makes an unwatched run reviewable afterwards.
Connect Slack to Polaris
Six scopes, no access to direct messages, and a bot that only reads the channels somebody invited it into.
Questions people ask
+Can my teammate change how the worker behaves?
Yes, if they have access. Instructions and the skill library are org-wide records readable and editable by members, so tuning a worker is a normal edit rather than a request routed through whoever created it. Connection credentials are the exception and are restricted to owners and admins.
+How does a colleague ask for a revision?
They comment on the task. A comment from anyone other than the worker itself inserts a follow-up job, and the runtime is instructed to treat the latest human comment as the brief, answer questions in-thread and avoid redoing finished work that nobody asked about.
+Can I keep some agent work private?
Buckets can be private, and job rows and their event streams are readable only where the underlying task is visible to you. Work in a private bucket stays inside it rather than appearing in the org-wide activity feed for everyone.
+Does sharing a worker mean sharing my API key?
No. An organisation stores one Anthropic key in a table with no select policy, writable only by owners and admins and readable only by server-side code. Members use the worker without ever holding or seeing the key.
Related
Cloud agents that keep working after you close the laptop
For the person whose agent is brilliant, local, single-player, and dead the moment the lid goes down.
Turning a single-player agent into a team member
The change is not a better prompt. It is putting the worker in the same table as the people.
Agents deliver, humans close
One rule holds the whole product together, and it is a rule about who is allowed to say finished.
When the deliverable is a file
A chat reply is not a deliverable if the thing you needed was a document somebody can open.
The audit trail behind every agent session
If you cannot reconstruct what happened three weeks later, you do not have an audit trail. You have a feeling.
Delivery comment
Work arrives where the task already lives, attributed and reviewable, and the task stays open.
AI teammate
An assistant belongs to a person. A teammate belongs to the team.
Connect Slack to Polaris
Six scopes, no access to direct messages, and a bot that only reads the channels somebody invited it into.