Summary
I recorded this video to demonstrate how Riley, our Bookly chatbot, handles real‑world user requests while enforcing guardrails and leveraging dynamic policies. I show the agent first detecting an unclear intent and immediately clarifying it with Claude Haiku. When the user tries to start a return, the system hits an authentication guardrail, prompts for a code, and then stores the session flag for subsequent calls. I then illustrate a request for an early refund; the agent pulls the relevant Agent Operating Procedure (loyalty early refund) into its context, runs a tool to validate eligibility, and approves the refund. After the resolution score exceeds 90 %, a resolution event triggers an upsell workflow that checks inventory, finds the title in stock, and places a new order. The demo ends with a recap of read/write guardrails and an invitation for live discussion.