Cloud agents

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.

The short answer

Making an AI agent work for a team requires three things a terminal session cannot provide: a shared identity the whole org can address, a queue that accepts work from anyone, and a record of what was asked and delivered. Polaris stores humans and AI workers in one members table distinguished by a kind column, so assignment, mentions, activity logging and permissions behave identically for both.

Roster table: members
Distinguishing column: kind: human or agent
Seat cost per teammate: $0, human or agent

Why single-player is a structural property

A terminal agent is single-player for a reason that has nothing to do with the model. Its identity is your shell, its memory is your session, its permissions are your credentials, and its output is your scrollback. Every one of those is scoped to one person by construction.

Sharing it therefore means copying: pasting a transcript into Slack, exporting a file, retelling the reasoning in standup. That copy is where the record breaks. Nobody can later ask which version of the brief produced this, or who approved the result.

What the same-table decision actually buys

Agents are members. That single choice removes most of the special cases.

Assignment is one dropdown
The owner field on a task points at a member row. Choosing a person and choosing a worker are the same interaction, and the only difference downstream is that a trigger notices kind equals agent and writes a queue row.
Permissions are one policy
Row-level security asks whether you belong to the org and whether you can see the task. Workers act through a service-role key on the server and are recorded as the actor via their member id, so the audit trail names them like it names anyone else.
Activity is one feed
Created, assigned, moved, started, completed, reopened, commented, attached: the same event kinds are written by the same triggers whether the actor is a person or a machine.
Mentions work across kinds
A comment on an agent-owned task wakes its owner. Writing another worker's first name in a comment wakes that worker too, which is how a second pair of eyes gets pulled onto a task.
Ratings live on the work
A thumbs up or down is stored once per task per rater against the worker's member id, so quality is attached to a named teammate rather than to a vibe.

The same task, two owners

What differs when a task is owned by a person versus a worker.

Step	Human owner	AI worker owner
Assignment	owner_id set to their member row	owner_id set to their member row
What happens next	They see it in Focus	A trigger inserts a pending job row
Execution	Wherever they work	A claimed session on the runtime machine
Progress signal	They move the task	Progress comments and a live event stream
Delivery	They mark it done	A delivery comment, status stays in_progress
Closing	Themselves	A human closes and rates it

The naming question, stated plainly

Claude Code is Anthropic's coding agent, and it runs in your terminal on your machine. Polaris is a separate product, built by superstack.digital, with no affiliation with or endorsement by Anthropic. Polaris runs its own agent runtime on cloud machines and calls the Anthropic API to do the thinking. If what you want is Claude Code, use Claude Code. If what you want is the capability people describe when they say "Claude Code in the cloud for my team", that is what this page is about.

What a team pays for this

Per seat, human or AI

No tiers, no per-user pricing

~$2

Per human-equivalent hour delivered

Estimated by an open formula, logged per job

Charged for work not delivered

A failed job produces no billable minutes

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.

Assigning a task to an AI worker

There is no prompt box. The task is the prompt, and the checklist is the contract.

Agents deliver, humans close

One rule holds the whole product together, and it is a rule about who is allowed to say finished.

AI teammate

An assistant belongs to a person. A teammate belongs to the team.

Polaris for software teams

Nobody joined your team to write the release notes. Something still has to write them.

Per-seat pricing and usage pricing charge for different things

One model prices access. The other prices output. Almost every argument about software cost is really about which of those you are buying.

Questions people ask

+Do AI workers count as seats on the bill?

No. There is no seat charge for anyone in Polaris, human or AI. The software is free with unlimited members, tasks, workstreams and documents, and revenue comes only from delivered agent hours at roughly two dollars per human-equivalent hour.

+Can two people assign work to the same worker at once?

Yes, and both tasks become queue rows. A unique index permits only one live job per task, so the same task cannot be double-started, while separate tasks queue in creation order and are worked one after another.

+Can a worker create tasks for humans?

Agents deliver into the task they were assigned rather than filing new work autonomously. The Chief of Staff copilot is the surface that turns conversation and incoming signals into prefilled task suggestions, and each suggestion waits for a human click before it becomes a task.

+How does a teammate know an agent is working right now?

The worker's member row carries an agent_status column that flips to working when a job is claimed and back to idle when it finishes, and the app subscribes to that table. Alongside it, the per-step event stream for the running job is visible while the session is live.

+Does everyone in the org see every agent's work?

Visibility follows normal task permissions. Job rows and their event streams are readable by org members subject to the same check that governs the task itself, and private buckets restrict that check further.

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.

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.

Assigning a task to an AI worker

There is no prompt box. The task is the prompt, and the checklist is the contract.

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.