What changed
DoorDash CEO Tony Xu said third-party AI agents have generated only several hundred DoorDash orders since access became available. That is a tiny figure beside the nearly 1 billion orders DoorDash handled in the third quarter of 2026. The company’s new MCP connector, launched less than a month ago, is still limited to a small group of early beta partners, mainly for corporate lunch orders.
Why it matters
For now, food ordering looks like a poor showcase for autonomous agents. Ordering dinner is already a short, familiar task, so most people still go straight to DoorDash rather than ask another tool to do it for them.
The more revealing contrast is inside DoorDash itself. Its Ask DoorDash feature can turn a recipe photo into a shopping cart or suggest dinner from a vague prompt. DoorDash says those baskets are built five times faster, have nearly 50% higher average value and include about 60% more unique items than the same customers’ traditional orders. Nearly half lead customers to local spots they have not tried before.
That suggests the near-term opportunity may be less about agents replacing the DoorDash app and more about making the app smarter once customers are already there. External agents have to earn a place in a simple transaction; DoorDash’s own tools can remove friction and shape the choice at the same time.
It is still too early to know whether third-party ordering will scale beyond these early experiments. DoorDash’s MCP connection has barely launched, and the several-hundred figure does not include usage of the company’s internal AI tools. As Xu put it, the sensible approach is to avoid fighting technology’s momentum without pretending every new interface already has demand.
Comments
No comments yet.