Platforms like DoorDash act as a middleman between restaurants and their customers. AI-based ordering could reshape the industry — if customers and restaurants get on board.
DoorDash handled 970 million orders in Q2 2026 and generated $4.5 billion in revenue, roughly 10.7 million orders a day, with more than 500,000 merchant partners, keeping it the clear leader in US food delivery. The counterpoint sits in the Bay Area: Bites has just 10 full-time employees, remains at the pre-seed stage, and has signed about 300 restaurants, all in the Bay Area, with no second city yet. Its product works differently from a conventional delivery app. Instead of opening an app and scrolling through listings, users describe what they want in natural language, including budget, party size, cuisine and dietary limits, and an AI agent handles search, filtering, p
The fight over delivery entry points is not new. In the 2010s platforms used subsidies to move users from phone orders to apps, and for the next decade the app icon served as the moat: whoever owned demand could charge restaurants commissions of 15% to 30% of order value. That structure assumed users made choices inside an app. After 2025 the assumption began to erode. OpenAI launched Instant Checkout with Etsy and Shopify in September 2025 and pushed the Agentic Commerce Protocol, letting models complete payments inside a conversation; Google tested age
Liu Gong argues that Bites' 300 restaurants settle nothing about scale, but the migration of the entry point it points to deserves a closer look. From a technical standpoint, what has blocked AI ordering was never model capability but the shape of supply-side data. Menus, inventory, opening hours and sold-out items are scattered across POS systems, platform dashboards and paper menus, with almost no common standard. For an agent to order reliably, someone must first structure those fields, then solve payment authorization, delivery address validation and order status callbacks. That a 10-person team signed 300 restaurants shows this work currently has to be ground out merchant by merchant. Liu Gong expects the winner to be whoever turns restaurant-side interfaces into a standard product, because models can be swapped while the integration layer cannot. On the business side, the heavier damage may land on advertising rather than delivery. Within DoorDash's $4.5 billion quarterly revenue, commissions and ads carry the gross margin, and ads depend on the platform controlling ranking. When a user hands the agent a request like 'Thai food under $30, not spicy', the agent reads only price, ratings and delivery time, routing around paid placements and shaving a layer off monetization. If agents become the default entry point, delivery platforms get compressed into fulfillment APIs and bargaining power shifts from demand to supply. For the industry, the rules of agentic commerce remain unwritten: who pays when an agent orders wrong, whether restaurants may refuse agent traffic, and whether platforms can technically block scraping all lack clear answers, and such disputes will arrive before scale does. Liu Gong predicts that over the next 12 months the more realistic use cases are group dining and corporate catering, where high ticket sizes and complex decisions make the saved decision cost worth the agent's friction.
- DoorDash processed 970 million orders and booked $4.5 billion in Q2 2026, about $4.60 in revenue per order.
- That equals roughly 10.7 million orders a day, more than 35,000 times the 300 restaurants Bites has signed.
- On merchant coverage, DoorDash works with over 500,000 merchants versus about 300 for Bites, a gap of roughly 1,600 times.
- On headcount, Bites has 10 full-time employees against DoorDash's roughly 20,000, a ratio of about 1 to 2,
- – Mainstream delivery platforms charge restaurants 15% to 30% commissions, while an AI agent comparing
For the industry, in the short term (three months) leading platforms will treat AI ordering as a feature rather than a threat, shipping conversational ordering and voice reordering while tightening data access for third-party agents. Within a year, restaurants will face demand to optimize for agent retrieval, making structured menus and open APIs standard rather than niche; third-party agents could take a single-digit share of order volume but cannot match incumbent delivery density. For ordinary users, the near-term gain is convenience: a sentence reord
Questions worth exploring:
- How agent entry points erode the ad and ranking-based monetization of delivery platforms
- Whether standardizing restaurant menu and inventory data creates a new middleware layer
- The rule vacuum around agent authorization, liability for wrong orders, and platform efforts to block scraping
This item comes from The Verge AI (published 2026-10-07 12:00:00). Wenfeng Radar provides a Chinese/English summary and commentary on overseas AI developments; the original content is owned by its author.
Daily aggregation of overseas AI developments with in-depth analysis. Bookmark us to never miss an important signal; back to homepage.