A restaurant's evening is governed by the phone. The kitchen is hot, the pass is busy, and the host stand is the chokepoint where reservation confirmations, takeout pickups, and walk-in waitlists all pile up. A bad night is when the host is on hold with a guest while three more guests are physically standing at the door waiting to be greeted. The economics are brutal: a missed reservation slot is empty revenue; a miscommunicated takeout order is a refund and a 1-star review.
The right AI agent for a restaurant runs the phone work the host should not be running. The host stays on the floor; the agent answers, books, takes orders, sends confirmations. This post walks through what fits and what does not.
What the agent does, ordered by frequency
1. Reservations. The single most-frequent call. The agent checks the booking system (Resy, OpenTable, Tock, or your in-house tool), offers the next two open windows that match the party size, books, and sends a confirmation text. Total call time: under 90 seconds.
2. Takeout orders. The agent reads the menu, takes the order, confirms substitutions, captures payment via Stripe link, and reads the kitchen the order via your KDS or by SMS to the expo's phone. Pickup time is auto-calculated from the kitchen's current ticket queue.
3. Hours and menu questions. "Are you open Sundays?" "Do you have gluten-free options?" "What is the kids' menu?" These calls are 80% of the inbound volume in a casual restaurant; the agent handles them in 20 seconds without ever reaching the host.
4. Same-day cancellations. A guest calls to cancel a 7 PM. The agent confirms, releases the table to the waitlist, and texts the next person on the waitlist to fill the slot. The slot does not go empty.
5. Special requests. "We have a birthday tomorrow, can you do something special?" The agent flags the reservation with a kitchen note + house-cake recommendation, and sends the chef a one-line FYI.
What it does not do
A restaurant has things the agent cannot help with, and being honest about them matters.
- The agent cannot greet guests at the door.
- The agent cannot tell the kitchen the steak is overcooked.
- The agent cannot diffuse an angry table.
- The agent cannot describe the wine list with a sommelier's expertise (although a serviceable knowledge of the menu plus pricing is well within its capabilities).
These are still the host's job. The agent is the phone-and-takeout layer, not the floor layer. The right framing for the restaurant owner is "this saves the host from being on the phone for 4 hours of every shift," not "this replaces the host."
Integration with the restaurant stack
Three integration points that determine whether the agent fits cleanly.
Reservation system. Native integration with OpenTable, Resy, Tock, SevenRooms. For long-tail tools we use a generic Google Calendar bridge. The agent reads availability and writes bookings; the system is the source of truth.
POS / kitchen display. For takeout orders, the agent writes directly to the POS (Toast, Square for Restaurants, Lightspeed) or sends an SMS to the kitchen's phone if the POS integration is not native. Most restaurants are on one of three or four major POS systems; we have integrations for the major ones.
Payment. Stripe Connect for takeout deposits and prepayments. For dine-in, the agent does not handle payments; that stays with the server's flow.
- 1Reservation systemOpenTable, Resy, Tock, SevenRooms, or Google Calendar bridge.
- 2POSToast, Square, Lightspeed for takeout writes.
- 3PaymentStripe Connect for prepayments + deposits.
- 4Kitchen alertSMS to expo phone if KDS integration is not native.
What the host's shift looks like with the agent on
A representative Tuesday evening at a 60-seat dinner restaurant.
6:00 PM. Doors open. 4 reservations confirmed for the next hour.
6:14 PM. Phone rings. Agent picks up. Takeout order, $42.
Kitchen sees it on the KDS at 6:15. Expected pickup 6:35.
6:22 PM. Phone rings. Agent picks up. Question about gluten-free
menu. Answered. Caller does not book.
6:31 PM. A 6:30 walks in. Host greets, seats, turns the phone over
to the agent.
6:38 PM. Phone rings twice in 90 seconds. Both reservations for
8 PM. Both booked. Host did not know.
7:01 PM. Phone rings. Cancellation for 7:30. Agent releases the
table, texts the next person on the waitlist, gets a
confirmation in 2 minutes.
7:14 PM. Host gets a text from the agent: "Birthday at 7:45.
Table 12. Want me to flag the chef?"The host has been on the floor the whole time. The phone has rung 9 times. The host took zero of the calls.
What changes about the customer experience
Three things diners notice over time.
The reservation confirmation feels professional. Two-way text confirmations, clear party-size expectations, easy reschedule via reply.
The takeout flow is faster. No "let me put you on hold" to clarify the kitchen's queue; the agent already knows.
The repeat-customer recognition. The agent remembers preferences: "Last time you mentioned a peanut allergy; we'll flag the kitchen." That detail used to live in the host's memory; now it lives in the customer record.
- Host is on the phone during peak. Floor experience suffers.
- Hold times for takeout. Lost orders.
- Waitlist gaps when cancellations happen.
- Repeat-customer details remembered or not, depending on which host took the reservation.
- Host stays on the floor. Phone is silent at the host stand.
- Takeout orders captured cleanly with no hold time.
- Cancellations rebooked to the waitlist within minutes.
- Customer history loads on every call. Allergies, preferences, last visit.
Frequently asked
Will guests notice they are talking to an AI?
What about non-English-speaking guests?
How does it handle waitlists?
Can it handle private events or large parties?
What about delivery platforms (DoorDash, Uber Eats)?
Try it
Call your own restaurant's main number at 7 PM on a Tuesday. The host's voice, or the silence, is the test.
