What "a table for four at eight" is actually asking for.

A restaurant gets asked, through an AI agent booking on a customer's behalf, for a table for four at eight. That sentence has a time and a party size in it, and nothing else: no date, no sign of a birthday that means a cake gets brought out, no mention of a dairy allergy that rules out half the tasting menu. Answered as a search query, it gets a yes. Answered as what it is, a request with conditions still attached, it gets a shorter list of real questions first.

Requests have conditions

A time and a party size are only the beginning of what decides whether Friday at eight works. Add a date, a dietary need, an occasion, a policy about how long a table can be held, and the honest answer can change completely, from an immediate yes to "not that table, but this one, if that's fine." A product order, an appointment, a quote carry the identical shape: a sentence that reads complete but is only complete once its real conditions are on the table.

If those conditions get lost, a result can look helpful without being useful. The restaurant confirms Friday at eight, and the customer only discovers at the door that the table doesn't suit a birthday party bringing their own cake, because nobody asked.

Ask the missing question early

An agent shouldn't invent an answer just to keep a booking moving. It should work out which single missing detail would change the outcome, the occasion, the allergy, how long the table's needed for, and ask that before returning a confident "table for four at eight is booked." A customer gets a better experience from one relevant question up front than from a confirmed booking that turns into an awkward conversation at the door.

For the restaurant, this produces a better inbound request too. The booking team sees a party of four, eight o'clock, birthday, a dairy allergy to flag to the kitchen, rather than a bare booking they have to unpack from scratch or discover the hard way.

Match the request to the source of truth

Some of what decides the answer is stable, whether the restaurant does birthday add-ons at all. Some of it needs checking against something live: whether eight o'clock is free tonight, whether the kitchen can flag a dairy allergy on short notice, whether an exception needs a manager's sign-off. The booking route needs to know which is which, rather than answering every question the same way.

The point isn't exposing every system behind the booking desk to an AI agent. It's connecting each part of the request to whichever system or person can give the next true answer, a live table plan for availability, a person for the exception.

Keep the context as it moves

When a booking gets handed onward, whether to a host confirming the table or a kitchen flagging an allergy, the details that shaped it need to travel with it. That lets the team pick up where the request actually is, rather than starting the conversation over from a bare "table for four, eight o'clock."

This is the whole shape of an agent-ready transaction: a specific request, birthday, allergy, eight o'clock, stays exactly that specific from the first question an agent asks to the moment someone can confirm it. Selfe connects a business's real booking system, the one that knows tonight's actual table plan, to something an AI agent can ask these questions against directly, rather than a form that only collects a bare time and party size.

Does this only apply to restaurant bookings?

No. The same shape applies to a product order, an appointment or a quote, anything that reads as a complete request but is missing a detail that would change the answer.

What if a business doesn't want an agent asking follow-up questions?

The alternative is a booking that looks confirmed and then falls apart on arrival, which costs more than one well-placed question up front.