Sierra Horizon is useful as a more mature form of long-running agent. It is not a “chat that lasts for weeks”, but a state machine around the customer journey.
The agent works with signals, playbooks, suppression rules, and a measurable terminal state. It can move a customer toward a mortgage, renewal, or sale through calls, email, messages, and internal systems. The important detail: payment is tied to the outcome.
This kind of agent requires a different operating model. It needs states, prohibitions, events, escalation routes, measurable outcomes, and protection against unnecessary customer touches.
Editorial read: a long-running agent becomes a product only when it has a terminal state and a business outcome, not merely a long message history.