The typical cycle takes under a minute. The customer writes “Friday night, two of us, any chance?”; the agent reads the request, queries the diary, answers with two or three genuinely free slots — “8:30 or 9:15” — and the customer picks one. Confirmed, the agent writes to the calendar: name, contact, covers, notes, automatic reminder for the day before. The owner touches nothing and finds everything logged. The decisive point is that every step uses real data: the agent never invents availability or times, because it doesn't “imagine” — it looks things up. It's the difference between an assistant that promises and one that executes, and in booking work only the second counts.
Acting on your behalf takes discipline, not just intelligence. Three rules separate a serious agent from a toy. Least privilege: it can read the diary and write bookings, not wipe history or touch payments unless necessary. A full trail: every action ends up in a readable log — who wrote what, when, at whose request — so the owner can check and correct. A clear action limit: certain operations stay human, from discounts to changes to large groups, and the agent must know how to hand over without resistance. Without these three you don't have an agent: you have a risk with a nice voice.
Frequently asked questions
- Can an AI agent get a booking wrong?
- It can misread an ambiguous request — “for two, actually four” — like a distracted person would. The difference is that it leaves a trail: every action is in the log, correctable in a minute. And on real data it doesn't get it wrong: it never offers full slots.
- What's the difference between an AI agent and a secretary?
- Coverage, not substance. The agent covers all hours — Saturday night, Sunday, the small hours — at a fixed cost, and handles repetitive volume without tiring. The person stays essential for judgement: complaints, off-script requests, special customers, delicate situations. The two roles integrate: the agent takes the first contact, the person closes the real cases.