Preparing for backlog refinement sessions can be a time-consuming and repetitive task. Artificial intelligence excels at handling such tedious efforts, like reviewing all backlog items and team comments. However, a significant risk emerges when an AI agent is tasked with "helping with refinement." Such an agent can inadvertently steer the conversation away from the team. Even without writing acceptance criteria, the agent can suggest actions like closing, merging, splitting, or creating backlog items. This preemptively answers questions the team was meant to explore during refinement. To mitigate this, clear boundaries must be established before the agent is used. The agent's role should be to identify evidence and formulate questions. The Product Owner is then responsible for reviewing these questions and setting the agenda for the actual refinement session. The team then undertakes the refinement process based on this agenda. Documenting these boundaries in an A3 Handoff Canvas solidifies the agent's responsibilities and allows the Product Owner to ensure adherence.