Look around. Most teams use both, often without naming the distinction. An AI assistant is the private window you open when you’ve got a problem and you don’t want to bother anyone. An AI coworker is a teammate the whole workspace shares.
Same Claude underneath. Different ownership and reach.
The ownership difference
An assistant is personal. It belongs to you, lives in a private window, and nothing leaves the conversation unless you copy it out.
A coworker is workspace-shared. It shows up in the workspace member list like a teammate. The whole team can @mention it, see what it does, and add it to the channels where its job is relevant. There’s no private mode by design.
The model can be the same. The audience is the entire workspace.
Where each lives
| AI assistant | AI coworker | |
|---|---|---|
| Owner | You | The workspace |
| Surface | A private window | Whichever channels it’s added to |
| Audience | Just you | The whole team |
| Memory | The current session | The channel context where it’s invoked |
| Tools | Whatever the assistant offers | The toolset the workspace gave it |
| Audit trail | Local to you | Visible to everyone in the workspace |
The “audit trail” row is the one that matters most for teams. When the agent’s work happens in a channel, the next person who needs the same answer can read it. The team learns from one attempt. With assistants, each attempt is locked in a different person’s window.
Why a team usually wants both
Assistants are great for thinking out loud, drafting, exploring, debugging a problem before it’s a team problem. Most of the genuinely private work belongs in an assistant.
Coworkers are great for the work that ends up benefiting from a trail of crumbs the team can follow. Sales follow-ups. Oncall summaries. CI digestion. Anything where “I did this and here’s what came back” is a useful thing for two more teammates to see.
A team that uses only assistants ends up doing the same work many times in private. A team that uses only coworkers ends up broadcasting every half-formed thought. The right answer is both, with a clear instinct for which one to reach for.
Picking which one to set up first
If you have a recurring question that lots of people across the workspace ask, that’s a coworker. Set one up with a narrow purpose and a small toolset and let the team @mention it.
If you have a unique-to-you task, that’s an assistant. Press ⌘J inside any Ano channel and Claude Code opens with the channel’s context as your personal shell. Same Claude, just yours.
The agent is the same Claude underneath. You’re just deciding where it sits. If you want background on how Claude agents work in general, Anthropic’s Managed Agents docs are a useful primer.
Going deeper
For the longer-form definition of the coworker side, see What is an AI coworker?. For the practical decision rubric between a coworker and ⌘J, see When to add an AI coworker vs press ⌘J yourself. For the design choices once you’ve decided a coworker is the right shape, see Designing a useful AI coworker.