Plan requirement
| Subscription | Suite Team or higher, Support Team or higher |
| Access | Admin |
Who may build and change AI agents. There are three roles, and the one thing to understand before you assign them: only one of the three can be limited to particular agents. The other two get all of them.
The three roles
| Client admin | View, add and edit all AI agents, and edit other users' role and agent access. |
| Client editor | View and edit all AI agents. |
| Client user | View and edit specified AI agents. |
So "let this person work on the retail agent only" is a Client user, and nothing else. Giving somebody Client editor because it sounds smaller than admin hands them every agent you run.
Set it
- Assign the role that matches what the person should reach.
- For a Client user, specify which AI agents they may work on.
- Check the list again after any team change, because access follows people rather than projects.
A client admin can change somebody's role and agent access after the fact, which is who to ask when the scope was wrong.
Everybody who is a Support admin already has the top role. Any user with the Admin role for Support is a Client admin for AI agents by default. So keeping this small is a matter of taking access away, not of granting it carefully: look at who your admins are before you assume the list is short.
Why keep it small
An AI agent speaks to customers directly, without review, at volume. A careless edit to a reply is published to everybody who asks that question, and there is no draft between the change and the customer.
Most settings affect how work is handled. This one affects what your company says.
Who should have it
Two or three people who understand both the configuration and the tone you want. In practice that is whoever owns the knowledge base plus an admin, which is a good pairing because the content and the agent are the same problem.
Contributing without access
Most improvements are article improvements, and those need no AI permissions at all. Someone who spots a bad answer should be able to fix the article, and that is the fix in the majority of cases.
Make that route obvious, or you get requests to change the agent for problems that are not in the agent.
Watch the transcripts together
Reading conversations does not have to be restricted to the people who can edit. Widening who looks, while keeping narrow who changes, gets you better information and the same safety.
Comments
0 comments
Article is closed for comments.