Field guide · AI agents
Training your Front Desk Agent on house policy
Karol Białek
Head of Messaging · 1 Mar 2026 · 5 min read
A Front Desk Agent that does not respect house policy is worse than no agent at all. Encoding policy is not a one-time exercise; it is a working agreement between the GM, the FOH lead, and the agent’s policy file. The mechanics below are the ones we walk new customers through.
Three layers
- Hard rules, the four to eight things the agent must never do (e.g., never quote below floor rate).
- Soft rules, the twenty things the agent should do unless the supervisor flags otherwise.
- Examples, fifty real conversations annotated as "good" or "needs revision" by the FOH lead.
How the agent uses each layer
Hard rules are deterministic blocks at the tool layer. Soft rules are part of the system prompt. Examples are retrieved at inference time when the conversation matches a similar past situation. The combination produces an agent the FOH lead can review on Friday and trust on Monday.
Run your hotel on the platform we write about.
Bring your existing data and your team’s habits. We’ll show you a like-for-like Prostay setup on a sample of your last 30 days.