Tony Xu built DoorDash on the stubborn realities of the physical world. Last week in Turin, the chief executive told an audience that artificial intelligence agents won’t shrink the need for people moving packages. They will expand it.

“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 at the Wave by Vento event. The Next Web reported his remarks from the stage. He spoke with DoorDash board member Diego Piacentini, a former Amazon executive.

The logic is straightforward. AI agents already outnumber humans in digital tasks. Their value lies in handling repetitive requests. Many of those requests end in the real world. More agents mean more orders that require a driver, a robot, or a drone. Demand for physical networks grows. So does the call for human labor in the last mile.

But. This view clashes with assumptions that automation displaces workers. Xu sees the opposite. Physical fulfillment becomes the scarce resource. DoorDash aims to own it.

The company has spent years mapping streets, restaurants, and customer habits. That data doesn’t live on the open internet. It comes from billions of deliveries. Xu calls this the hard part. “We’re actually solving the end-to-end job for a customer, which is to get them some item brought to them in the condition they expect, on time, every time,” he told analysts earlier this year. Restaurant Business Online covered his comments on an earnings call.

He views external AI agents as top-of-funnel channels. Much like Google or Facebook once drove traffic. They discover and order. DoorDash handles what happens after checkout. Late drivers. Missing items. Substitutions. Those problems sit in the physical realm. Only operators with real infrastructure solve them at scale.

DoorDash moved fast to embed itself in the agent future. At its Dash Forward event in late September, executives unveiled a text-to-order system inside Apple Messages. Users text requests such as “order my usual” or “find me a protein bowl for the office.” The agent pulls order history, builds a cart, shows photos, and checks out. No app needed. TechCrunch detailed the rollout, which opened to a waitlist for U.S. iPhone users.

Early pilots reached about 20,000 customers. The system learns preferences over time. It handles group orders with mixed diets. It even accepts a photo of an empty refrigerator and suggests recipes with matching groceries.

At the same event Tony Xu’s team announced DoorDash Air. The company earned Part 135 certification for drone operations. Purpose-built drones will launch from partners including Chipotle and Popeyes. Early tests in Northern California clock in under five minutes for some deliveries. Xu’s September 30 post on X laid out the vision. More retail categories now arrive in under an hour. Five hundred thousand items available to the average U.S. customer. Returns come with a flat fee.

Robots fill another slot. DoorDash Labs developed DOT, an autonomous delivery robot roughly one-tenth the size of a car. It travels roads, bike lanes, and sidewalks at up to 20 miles per hour. It carries 30 pounds. The company deployed it in Arizona cities with plans to expand. An AI-powered Autonomous Delivery Platform decides in real time whether a human Dasher, a drone, or DOT makes the most sense for each order. Cost, speed, distance, all factored.

This hybrid network matters. Pure autonomy faces limits. Weather disrupts. Traffic changes. Robots struggle with stairs or apartment doors in many markets. Human couriers still handle the majority. And Xu believes AI will push volume higher, not lower.

Competitors sense the shift. A small startup called Bites lets users order directly through ChatGPT. It charges a flat dollar fee instead of DoorDash’s typical 15 to 30 percent commission. Restaurants keep more. Diners sometimes pay less. One Bay Area location saw 65 percent of its orders move away from traditional platforms, according to recent discussions on X. The Verge examined the tension just days ago.

OpenAI, TikTok, and Pinterest experiment with agents that guide users to checkout. Each wants a piece of the transaction. Yet the economics of delivery stay messy. Packaging, timing, insurance. Agents compress the digital side. The atoms remain expensive.

DoorDash processed 970 million orders in its second quarter. Revenue hit $4.5 billion. About $4.64 per order funds the marketplace. That margin supports the infrastructure Xu bets will become more valuable.

He frames the economy in two battles. One for attention, fought in bits. Another for physical execution, fought in atoms. DoorDash chose the latter years ago, even before ChatGPT arrived. The decision carries costs. Every burrito involves roughly 20 interconnected systems. Traffic breaks plans daily. Trust resets with each order.

That’s why every DoorDash employee still makes deliveries. The practice dates to the earliest days. Xu once interviewed engineers from his Honda while dropping off orders. He wanted people who understood real problems, not just code.

The strategy shows in results. DoorDash holds more than 60 percent of the U.S. food delivery market. Orders grew 32 percent in one recent quarter. Marketplace spending rose 39 percent.

Yet challenges loom. AI agents could erode discovery and advertising if users bypass the app entirely. DoorDash counters by making its own agent available inside the tools where people already work. Slack bots for corporate lunches. Internal agents that restock supplies. A connector based on the Model Context Protocol lets companies keep employees inside their existing workflows.

Xu sees these as complementary. Agents drive demand. DoorDash fulfills it. The company expanded into returns, groceries, and retail. Macy’s, Vans, The North Face, and Anthropologie joined the platform. DashOS gives merchants a unified view whether orders arrive through the app, a restaurant website, an AI agent, or in-person dining.

The bet rests on execution. Mapping the physical world. Operating at scale. Fixing issues when packages go astray. Those tasks resist pure software solutions. Data must be collected the hard way, through actual deliveries.

Xu’s message in Turin was clear. AI won’t render delivery networks obsolete. It will strain them. Then it will force them to evolve. Robots and drones will take certain routes. Humans will handle the rest. The network grows more capable. Volume rises. Labor demand follows.

Plenty of uncertainty remains. Regulation on drones. Public acceptance of sidewalk robots. How quickly agents gain trust for high-stakes tasks like medical deliveries or expensive returns. DoorDash tests in limited markets first. It measures outcomes, not hype.

One thing seems likely. The conversation around AI and jobs often focuses on displacement. Xu flips the script. In delivery, intelligence creates more work for people. At least for now. The physical world doesn’t digitize easily. Someone still has to hand over the bag.

And that someone, or something, will probably come from a DoorDash network.