Each one does a single job well, works from the full account context and stops at the points where a person should decide.
Reads every new ticket with the account behind it: plan, seats, usage, open invoices and renewal date. It tags and routes the ticket, links known issues and drafts the first reply.
Hands off to your support team for review before anything is sent.
Watches new accounts through setup and notices the ones that stall. It sends help that matches where they are stuck, and brings in customer success when a nudge is not enough.
Hands off to customer success with the full setup history.
Picks up failed payments, works out why they failed and decides whether to retry or reach out. It writes to the billing contact with the right context instead of a generic dunning email.
Hands off to your billing owner before any customer email goes out.
Set a rule for each action once. Routine steps run automatically, anything that reaches a customer or touches money waits for a person, and some actions are simply off limits.
Each agent connects only to the tools and fields its job needs, with read and write access set separately.
What the agent read, what it decided and what it changed, so any action can be traced and undone.
Agents use Anthropic's Claude models for reading tickets and documents, reasoning about the account and drafting replies.