A short story
Ahmed is the workspace admin. People across the company have started building agents.
He sets the rules once. In the workspace capability settings, he locks the email rule for every agent: each agent needs an allow-list, and email to anyone outside it waits for approval. Owners still choose who is on their own list, but they can’t switch the rule off.
He looks at the whole picture. In Agent management he filters on the connection to the company’s CRM. Five agents come up. One is a personal agent. Ahmed can see who owns it, what it’s connected to and how many credits it uses, but not its instructions, chats or documents. That’s by design. He asks the owner what it’s for.
Another is a restricted agent that is open to outside systems through a webhook. To check it, Ahmed gives himself access. The step is logged, and the agent’s owners can see it in the agent’s activity log.
He hears when something happens. One morning he gets an alert: attack detection blocked an email with hidden instructions sent to Sara’s invoice agent. Thanks to the email rule, the agent couldn’t have sent anything to an unknown address without Sara’s approval anyway.
Ahmed doesn’t read every agent. The rules hold everywhere, and he can see what each agent can reach.



