The next delivery app may be your restaurant’s AI chat

For years, ordering food meant opening an app, accepting its rules and paying for access to demand the restaurant often helped create. Now an order can start inside an AI conversation and reach the restaurant directly. This is not a distant scenario: US experiments are already running.

The Verge reported on October 7, 2026 that Bites, a roughly ten-person startup, had about 300 restaurants signed up in the Bay Area. Its model uses a flat US$1 per-order fee—typically paid by diners, though some restaurants cover it—compared with DoorDash’s published 15%, 25% and 30% delivery commission plans. In July, Square announced ChatGPT and Claude integrations that let eligible restaurants surface menus and receive orders without an additional marketplace commission. These are US signals, not Colombian availability. They do show where leverage may move: the chat can become a doorway to the direct channel.

The app was the storefront. Conversation may be the next door

The interface is not the main change. The ownership of context is. A useful agent needs the live menu, prices, inventory, opening hours, delivery zones, payment rules and a clear human handoff. If the order lands in the restaurant’s POS instead of a marketplace, the economics and the relationship can change with it.

Bites matters precisely because it is small. The Verge’s October 7, 2026 report said the ten-person team had about 300 Bay Area restaurants signed up and described DoorDash’s warning emails to some restaurant partners. That does not prove Bites will replace an incumbent. It shows that a thin software layer can challenge a large platform when it reduces the cost of taking an order.

Square’s announcement is a different signal. Its ChatGPT app and Claude plugin are designed to help eligible sellers get discovered and, for supported restaurants, accept orders directly. Square says those orders do not carry an additional marketplace commission. The platform remains involved, but the restaurant can retain more control over the transaction and the customer relationship.

A menu is now an operating interface

An AI agent cannot order from a menu that only looks good. It needs structured answers: what is available, ingredients and allergens, final price, preparation time, delivery area, substitutions, cancellation terms and the moment a human must approve an exception.

This is where many “AI ordering” projects fail. They connect a conversation to stale data and then call the result automation. The customer experiences a confident mistake: an unavailable dish, a wrong price, or a delivery promise the kitchen cannot meet.

Direct does not mean alone

Restaurants may still need marketplaces for discovery, logistics and coverage. A direct channel can be stronger for repeat customers, high-margin orders and first-party data. The right architecture can include both.

Ask five questions before choosing a connector:

1. What does each fee actually pay for?
2. Where does the customer data go?
3. Can the channel respect brand, menu and hospitality rules?
4. What margin remains after fulfillment and support?
5. Can the restaurant earn permission to continue the relationship?

The operating checklist

Choose one source of truth for menu and price. Make the POS capable of receiving structured external orders. Set permissions for what an agent can confirm. Define consent and data retention. Measure contribution margin, repeat rate, errors and kitchen time — not order count alone.

For Colombia and Latin America, the current integrations cited here are not a promise of immediate availability. The preparation work is still clear: clean data, an operational menu, explicit policies, connected ordering and a direct experience that can receive the customer after the conversation ends.

Restaurants building this capability can connect the menu, orders and operations as one system instead of adding another isolated screen. Pimentone works on these challenges through digital architecture, AI agents and conversion engines. The right solution depends on order volume, existing systems and how much control the business needs.