DoorDash did 993 million orders in Q3. Third-party AI agents contributed 'several hundred'

Three layers of material, kept separate. The first is DoorDash's third-quarter earnings release published on its investor relations site on October 29, 2026, the official source, which covers financials only. The second is the earnings call transcript from the same day (Motley Fool has the full text), where the agent-order remark came in the Q&A. The third is TechCrunch's October 30 report, which combined the call with background such as the company's 2025 figure of three million deliveries a day. Every financial number here comes from layer one, every quote from layer two, and layer three is used only for cross-checking.

The official numbers first: total orders of 993 million, up 28% year over year and 17% excluding Deliveroo; Marketplace GOV of $33.9 billion, up 36% (23% excluding Deliveroo); revenue of $4.6 billion, up 34% (23% excluding Deliveroo); net income attributable to common stockholders of $466 million; adjusted EBITDA of $1,060 million, up 41% and 3.1% of Marketplace GOV. The company also said it will stop disclosing figures excluding Deliveroo from Q4 2026, once the acquisition anniversary passes.

The key exchange was in Q&A. Xu called it a very dynamic area, then offered the line worth quoting: my take on all things technology is you never want to fight gravity, but at the same time you do not want to be delusional. Only then came the number: if you look at where AI agents are today, we are seeing maybe several hundred orders in total from some of these third-party agents. Asked whether there could be revenue sharing with consumer agents, he repeated the magnitude: right now, AI agents are generating several hundred orders across the time that they have been up and running, so it has been a fairly low use case.

The gap only becomes tangible once the number goes into the denominator. At 993 million orders per quarter and 92 days, that is roughly 10.79 million orders a day, or about 450,000 an hour. If third-party agents contributed 500 orders in total, that is 0.00005% of quarterly volume, one in about two million; even at 900 orders it is only one in roughly 1.1 million. In other words agent ordering today is not the early stage of a new channel so much as test traffic on one. The arithmetic is mine; DoorDash did not make this comparison.

Who the third-party agents are matters. TechCrunch names consumer assistants such as Muse, Instinct and Fo, none of which is a delivery platform's own product; these are general-purpose assistants trying to complete the order-a-meal action on a user's behalf. For that path to work it must clear three hurdles at once: the assistant has to understand real menus and delivery zones, the user has to be willing to hand over payment and address, and responsibility has to be assignable when something goes wrong. Fail any one and conversion stalls there.

DoorDash has not been idle on its own side. Per TechCrunch, it launched the Ask DoorDash chatbot in June (ordering through prompts and photos), a CLI agent for command-line users in July, and a text-to-order agent in September, plus an MCP that lets other chatbots place orders, aimed mainly at corporate ordering. The interfaces were not the bottleneck; they were laid down and left waiting for demand.

One easily missed detail in Xu's answer: he did not dismiss the path, he framed a hypothesis for working with it. He said he believes there will be opportunities for win-win arrangements with third-party agents, and that the company wants to learn more about agentic use cases and build a healthy relationship with agent makers. TechCrunch adds a more specific conjecture, that there could be an opportunity to advertise inside these agents. That is the outlet's framing rather than a company commitment, but it points at a real problem: when the entry point moves from your own app to someone else's assistant, the monetisation model has to move with it.

Why demand has not materialised can be inferred from one platform-side observation: DoorDash says users still predominantly order through its own app. That is habit plus friction. Ordering food is high-frequency and low-deliberation, so adding an assistant inserts steps rather than removing them; agents earn their keep on tasks that span several apps under multiple constraints, such as not wanting to open five apps to compare. Single-platform food ordering happens to be one of the cases where an agent's advantage is smallest. That is my reading, not the company's explanation.

For developers there are two actionable layers. The interface layer first: MCP, CLIs and agent ordering are already on a mainstream platform's roadmap, so wiring them up is not the hard part; the hard part is that wiring them up produces no volume. Then the product layer: if your value proposition is that AI orders on the user's behalf, you now have to answer one concrete question, namely why a user would hand payment authorisation to a general assistant instead of opening the app they already use every day. Until that has an answer, more interfaces just mean more test traffic.

Boundaries, clearly. First, the company gave one vague magnitude, several hundred orders, with no exact figure, no time window and no quarter-on-quarter comparison, so neither the slope nor whether this is a steady state can be determined. Second, the release contains nothing about agent orders; the remark exists only in the call's Q&A. Third, background numbers TechCrunch cites, such as three million deliveries a day in 2025, are that outlet's framing and are not on the same basis as this quarter's 993 million orders, so they should not be divided into one another. Fourth, the idea of advertising inside agents is a conjecture in reporting, not an announced plan. Fifth, the company did not disclose which third-party agents or how many orders each contributed.