Short answer
Yes, and you decide how far it goes. The sensible split is by what can be undone. Lookups change nothing and can run on their own; changes you can reverse can run on their own if they’re logged; and anything you can’t undo, like canceling, deleting or refunding, should be confirmed by a person, because if the agent gets it wrong there, nobody will notice in time.
In detail
The three levels
Look up, change and reply. The first is controlled with permissions, meaning which queries you open. The second, with which actions you give it and which ones need confirmation. The third, by reviewing what it says. Almost all the value is in the first, and almost all the risk in the second.
Start with lookups
Checking a plan, a usage figure or an order changes nothing and resolves a good share of conversations. Start there, review it for a few weeks and expand once you have confidence and data. Opening up changes on day one is how the incidents that give these tools a bad name happen.
Where the limit goes
In the system, not in the instructions. Writing "don’t cancel subscriptions" in the prompt is a request. The cancel action being unavailable, or requiring a person to confirm it, is a limit that holds every time.