DoorDash says outside AI agents have placed only several hundred orders
On DoorDash's third-quarter earnings call on Thursday, co-founder and CEO Tony Xu answered an analyst's question about AI agents with a very small number. In his words, "AI agents are generating several hundred orders [on DoorDash] across the time that they've been up and running." That figure is cumulative over the whole period the agents have been running, not a quarterly or daily rate.
Xu called this a "very low use case" for a company that handled just shy of 1 billion orders in the third quarter of 2026 alone. Ars Technica calls the phrase a marked understatement and says that, so far, a vanishingly small number of people seem to be using an outside AI agent with the food delivery app.
Speaking more generally, Xu said he sees potential for AI agents to be genuinely helpful for "long consideration purchases" and for major decisions where agents would be "doing work on your behalf proactively in the background." But for simple tasks the pattern is different. In his words: "when it comes to a task that's as simple as ordering food, we tend to see that come directly to DoorDash."
Ars Technica adds context in fairness to the AI boosters: DoorDash only rolled out its Model Context Protocol (MCP) connector less than a month ago. The connector gives third-party AI tools a standardized, open-standard way to order through DoorDash's platform. It is still available only to a small number of "early beta partners" and is mainly used to streamline corporate lunch orders. Without that kind of link, AI agents have to go through more effort to reach DoorDash, for instance through a web interface.
Key facts
- CEO Tony Xu said on the Q3 earnings call that outside AI agents have generated "several hundred orders" across the time they have been running.
- DoorDash handled just shy of 1 billion orders in Q3 2026 alone, so Xu called agent ordering a "very low use case".
- Xu sees agents helping with "long consideration purchases", but says simple tasks like ordering food tend to come directly to DoorDash.
- DoorDash's MCP connector launched less than a month ago and is available only to a small number of early beta partners, mainly for corporate lunch orders.
Why it matters
Many companies are trying to use LLMs to improve how people buy goods and services online. This is a concrete data point from a large consumer platform: several hundred agent-placed orders in total, set against nearly 1 billion orders in a single quarter. Xu's argument is that agents may suit big, slow decisions, while a simple task like ordering food still goes straight to the DoorDash app.
Who it affects
Companies building AI shopping and ordering agents, and the platforms they want to order through, are the most directly affected. The figure also bears on DoorDash's early beta partners, who use the MCP connector mainly for corporate lunch orders.
How to use it
DoorDash's MCP connector offers a standardized, open-standard way for third-party AI tools to order through the platform. Access is currently limited to a small number of early beta partners. Without it, agents have to go through a web interface, which takes more effort.
How solid is it
The figure comes from the CEO himself, speaking on an earnings call in response to an analyst's question, as reported by Ars Technica. It is a rough number: "several hundred", cumulative, with no exact count, no per-day or per-week rate, and no breakdown by agent or platform. The Q3 total is also given only as "just shy of 1 billion".
Risks and caveats
The sample is young. The connector is less than a month old and restricted to a few beta partners, so the number says little about what agent ordering could reach with wider access. Ars Technica frames this as "Give it time?". The article gives no revenue or order-value figures for agent orders and does not name which agents placed them.
“when it comes to a task that’s as simple as ordering food, we tend to see that come directly to DoorDash.”
— Tony Xu, DoorDash co-founder and CEO, on the Q3 2026 earnings call