DoorDash co-founder and CEO Tony Xu used a wide-ranging interview in Turin to argue that the next wave of artificial intelligence will not simply replace human workers. Instead, he said, it will increase demand for the physical delivery networks that move goods to people’s doors — and, in many cases, for the human labour that keeps those networks running.
Xu spoke on Wednesday at an industry event in Turin, Italy, during a session titled “Inside the Making of DoorDash.” The conversation, led by a DoorDash board member and former Amazon executive, covered the company’s origins, its robotics and drone efforts, and its ambitions to become a last-mile logistics network that can deliver almost anything.
According to Xu, AI agents have probably already overtaken humans in the number of things they do on the internet. To be useful, he said, these agents look for recurring tasks that people ask them to perform again and again. The most valuable of those tasks, he argued, often have a physical component: shopping, eating, picking something up, or returning an item.
“One of the most useful things to do is things in your daily life which tend to have a physical component, like shopping or eating or picking something up or returning something,” Xu said. As the number of agents keeps growing, he added, so will the demand for physical networks and for human labour. In September, DoorDash launched an ordering agent that runs inside Apple Messages.
AI agents and the physical bottleneck
The idea that AI agents could increase rather than reduce demand for delivery may seem counterintuitive. Much of the debate around automation focuses on software replacing white-collar tasks, from writing to customer service to coding. But Xu’s argument is that digital agents are only as useful as their ability to complete real-world errands. An AI agent can compare prices, place an order, and schedule a return, but someone or something still has to pick up the item, transport it, and hand it over.
That final step is where DoorDash sees its opportunity. The company began as a restaurant delivery service, but it has expanded into groceries, convenience items, retail, and other categories. Each new category adds more physical complexity: different store hours, different packaging, different temperature requirements, and different return policies. An AI agent might make ordering more seamless, but it does not eliminate the need for a courier network.
Xu suggested that agents will make routine transactions more frequent because they lower the effort required from consumers. If a person can ask an agent to reorder household staples, send a gift, or return a package, they may do those things more often. That, in turn, could create more delivery jobs, not fewer. The demand for human labour may also persist in tasks that robots cannot yet handle, such as navigating apartment buildings, dealing with uncooperative weather, or resolving an unexpected problem at a restaurant.
From Palo Alto driveways to a global platform
Xu said the goal of DoorDash was never food delivery. It was to help local business owners like his mother, who moved from China to the United States. In 2013, he recalled, almost no restaurants in the US offered delivery. He and his co-founders started doing deliveries themselves, driving their own cheap cars around Palo Alto, where they studied. They completed the first few hundred deliveries and kept going for two years.
Xu also poked fun at the early projections for the business. His Y Combinator application estimated $100 million of revenue in year seven, a figure he called made up. He was off by a couple of orders of magnitude, he said. The lesson, according to Xu, is that when there is no data, a company has to start from a point of view. Then it should ship something fast and collect information. DoorDash keeps the stakes low so anyone in the company can test an idea with customers without needing approval.
That experimental culture has shaped DoorDash’s expansion. The company now operates in more than 40 countries, but Xu said each market had to be learned from scratch. Streets in cities such as Rome, Turin, or London are not laid out on a grid. Homes are denser, with more apartments. There are also far more small, family-run restaurants and shops than in the United States. These differences affect everything from routing to packaging to customer expectations.
Robots, drones, and the dispatch layer
DoorDash has invested heavily in automation. Its Dot robot can travel on the road, in bike lanes, and on the pavement. Xu called it the world’s first commercial autonomous vehicle of its kind. It is running in Phoenix and has made live deliveries to tens of thousands of customers, he said. The same week, Deliveroo started putting delivery robots on pavements in five UK towns, a sign that pavement delivery is becoming a competitive arena.
Building autonomous vehicles is like building three to five companies, Xu said. They need software, hardware, policy and permits, and operations. Vehicles break down and sometimes need rescuing. In September, DoorDash also unveiled its own drone system. But the most important piece, according to Xu, is the platform behind those vehicles. For every order, the platform decides whether a land robot, a drone, or a human Dasher should take it. A person and a robot can also share one route, with the human handling the final steps that a machine cannot.
“No customers care about the technology. They just care about getting what they ordered,” Xu said. That focus on the customer experience rather than the novelty of the technology helps explain why DoorDash is pursuing multiple modes of delivery at once. The company is not betting on a single robot or drone design. Instead, it is building a logistics layer that can route each order to the most efficient available option.
Human labour in the age of automation
DoorDash’s reliance on human Dashers has been a central part of its business model. In many markets, delivery workers are independent contractors, a classification that has been challenged by regulators and labour groups. The rise of AI agents could intensify those debates. If agents drive more orders, platforms may need more couriers to meet demand. But the terms of that work — pay, benefits, and flexibility — remain contested. Xu’s argument that AI will raise demand for human labour does not address those questions directly, but it does suggest that the company expects human workers to remain essential for the foreseeable future. Robots and drones may handle some routes, especially short or predictable ones, but human couriers are likely to remain the most flexible option for complex urban deliveries.
The economics of low margins and last-mile density
The board member who led the interview, Diego Piacentini, said DoorDash measures its margins in basis points, which forces attention to detail and a lack of waste. Building a business with low margins is very hard, he said. That observation underscores the challenge of last-mile delivery: every cent counts, and small inefficiencies can multiply across millions of orders.
DoorDash’s ambitions extend beyond food. Xu said that in five years, he wants DoorDash to be a last-mile network that can deliver anything. Every city holds tens of millions of items, he said, and no digital product yet knows where they all are. That gap between the physical world and digital inventories is both a problem and an opportunity. If DoorDash can map more of what is available in local stores, it could become the default delivery layer for a wide range of goods.
Asked about his mistakes, Xu said most came when he did not trust his gut, on products and on hiring. He also sits on Meta’s board. The best boards, he said, ask tough questions without acting as if they run the business. Founders spend 70 hours a week or more on their companies, he said, while board members spend about 10 hours a quarter. That asymmetry, in his view, means directors should focus on oversight and long-term judgment rather than day-to-day management.
For Xu, the growth of AI agents fits into a broader pattern: technology changes how people interact with the physical world, but it does not make the physical world disappear. The more that software can automate the decision to buy, the more important it becomes to have a reliable network that can actually deliver. Whether that network is staffed by humans, robots, drones, or a combination of all three, DoorDash is betting that demand for the last mile will only increase.