Contributing editor Peter Backman is a long-term foodservice sector guru and founder of theDelivery.World, a platform that connects the delivery sector and makes sense of the myriad changes and challenges that affect the sector across the globe.
Around the world, food delivery companies are building connections to artificial intelligence assistants so diners can order a meal simply by asking for one. My October article on second-party ordering ended by suggesting that, if diners began asking AI assistants what to eat, what would matter most is who holds the menus that AI orders from.
In China, Alibaba’s Qwen app announced the ability to order food through Taobao Instant Commerce in January, Alibaba’s own delivery business. Meituan CEO Wang Xing said during a June earnings call that the company would serve AI agents as well as consumers and merchants. Meituan is also among the early services being tested in WeChat’s AI agent system.
KFC is the exception—a restaurant brand with its own route into an AI assistant rather than one that runs through a marketplace. Its ordering assistant, built on Qwen, launched in February; Alibaba subsequently announced that KFC would have its own agent inside the Qwen app, covering pickup from more than 13,000 restaurants in China. Few restaurant businesses have the scale or resources to develop something comparable.
In India, food ordering and delivery platform Swiggy enabled ordering through ChatGPT, Claude, Gemini and other AI tools. Payment arrangements with Razorpay and the National Payments Corporation of India also let agents in Claude pay for orders from Swiggy, Zomato and Zepto. The government-backed Open Network for Digital Commerce had announced no comparable connection as of August.
Elsewhere, tech firm Kakao selected Coupang Eats as the first delivery service for its forthcoming AI agent in Korea. In Brazil, iFood’s Ailo assistant lets customers complete orders in WhatsApp. In Europe, Just Eat Takeaway.com introduced voice ordering inside its own app in eight countries.
Some forms of AI ordering involve external agents placing orders; others add AI capabilities to existing marketplace services. But the delivery marketplaces are putting themselves at the center of these new arrangements.
Nevertheless, traffic is light. On Uber Technologies’ November 4 earnings call, the company reported about 500 Uber Eats orders from external AI agents during Halloween week, against roughly 320 million rides and delivery orders across Uber that week. On DoorDash’s October 29 earnings call, CEO Tony Xu told analysts that third-party agents had produced “several hundred orders in total.”
Noah Glass, founder and CEO of restaurant technology company Olo, recently posed a question to me that addresses what AI might mean for restaurant delivery.
“How will AI decide amongst various ordering options for pickup/delivery?” he asked. He said the lowest price and shortest path to checkout and fulfillment would be the key.
AI assistants already place orders through delivery marketplaces. They connect to a restaurant’s own ordering system or use another intermediary. Which route they take determines who earns the commission, who controls the customer relationship and what is left for the restaurant.
But Glass made the further point that restaurants need to establish their own connections to AI assistants now, before consumer habits become entrenched. I would add that AI companies and delivery platforms are striking deals at the same time. By the time AI ordering becomes commonplace, where agents send orders may already have been decided.
This test assumes an assistant can choose between routes. If a marketplace is the only service connected to it, there is no choice. If consumers become accustomed to ordering through those marketplace connections, other routes may find it harder to gain a foothold later.
In the U.S., restaurant technology providers are developing routes that do not require orders to pass through delivery marketplaces.
Since July, Square has allowed restaurants to accept orders through ChatGPT and Claude without marketplace commissions, although standard processing fees still apply.
Olo is building connections intended to let restaurants receive orders placed directly through AI assistants while retaining the customer relationship and associated data. Its new consumer app provides another ordering route for participating restaurants.
Glass pointed to Shopify as a model. Shopify has developed ways for merchants to sell through AI interfaces without depending on conventional marketplaces. The model is based on the principle that individual businesses should not need to build separate connections to each new way consumers choose to shop.
A direct route is not necessarily the cheapest or easiest, either. Delivery still needs arranging, and marketplaces may offer discounts, stored payment details and reliable fulfillment. An AI assistant might prioritize the customer’s total price, convenience or previous ordering history rather than the restaurant’s margin.
It remains uncertain whether AI assistants will consistently favor the lowest-cost, most direct route.
The international evidence gives weight to a need for urgency. Marketplaces are busily establishing technical connections and commercial relationships while very few customers are currently using AI to order food.
Waiting until customers are ordering through AI in significant numbers may mean missing the boat if the available routes, and the businesses controlling them, are already well established.
None of this means widespread AI ordering is inevitable. An agent may offer little advantage for routine purchases, although it could help with unfamiliar restaurants or complicated orders.
Restaurants should be asking which assistants their ordering systems can connect to, what those connections cost, who receives customer information and whether alternative routes remain available. Accurate menus, prices and availability that machines can read are a practical starting point.
Restaurants that wait for consumers to start ordering through AI may find that the ordering channels have already been established, largely without their involvement.
