AI Agent Assistants

Every persona is an AI Agent Assistant: a job title for an assistant. Each one has its own skills, its own integrations, and its own private workspace, so you can hand it a goal the way you would hand it to a colleague. You are the CEO. You give the goal, the assistants do the work, and anything risky waits for your Approve, as your house rules require.

What an AI Agent Assistant is#

An AI Agent Assistant bundles four things:

• A role. What the assistant is for, such as writing code or drafting a campaign.
• Skills. Step-by-step playbooks it follows for common jobs, such as "fix CI" or "draft a payment reminder".
• Integrations. The tools it is allowed to use, such as GitHub for the Coder.
• A workspace. A private computer of its own, so one assistant's work never lands in another's files.

The AI Agent Assistants#

Hire any of them from the AI Agent Assistants screen in the app. You can hire more than one of the same kind, each with its own name and focus.

For universities and science#

StellarFirm for universities and StellarFirm for scientists put two assistants first.

The rest of the team is there when you need it: the Coder builds a course site or turns an analysis into a tool, and the Copywriter helps you write in your own words. See the university and science lab playbooks.

How the CEO briefs them#

You do not need to pick the right assistant yourself. You write a goal in chat, and the CEO assistant reads it, chooses the AI Agent Assistant that fits, and briefs it with the details it needs. The chosen assistant then works in its own workspace and reports back in the same chat.

1. You give a goal in plain words.
2. The CEO assistant routes it to the right AI Agent Assistant. See routing.
3. The assistant works and shows its progress, its terminal, and its changes.
4. If a step is risky, it waits for your Approve in Approvals.
5. You get the result in chat.

If you already know who should do the job, name the assistant at the start of your message and routing is skipped.

Coder, look at the open issues on [repository] and tell me which one you would pick up first, and why.

Several assistants, one goal#

A chat in StellarFirm is a session. A session can hold several assistants at once, all reading the same conversation. You can address one of them, or everyone.

• Address one assistant by name at the start of a message: "Coder, ...".
• Write a goal with no name and the CEO assistant plans it and hands each part to the right assistant.
• Each assistant works in its own workspace, so they do not overwrite each other.
• Every risky step from any of them lands in the same Approvals list.

So a Product owner writes the ticket, a Coder builds it, and Marketing announces it, all in one session. In a university, the Scientist turns a new result into notes and the Teacher turns those notes into a lesson.

Team, our goal this week is [goal]. Break it into parts, tell me who does what, and start.

Several Coders at once#

You can run more than one Coder. Each Coder has its own name, its own focus (for example "UI engineer" or "backend engineer"), its own repository, its own login to your source host, and its own workspace. They run side by side, so a UI change and a backend change can move at the same time.

Address a Coder by its name, or let the CEO assistant route the message by focus or by the repository it names. Set them up under Coders in Settings. See the Coder for details.

UI Coder, build the [component] on [repository-one]. API Coder, add the endpoint it needs on [repository-two]. Tell me when each pull request is ready.

In this section#

Ready to put them to work for a specific kind of business? See the playbooks.