DoorDash Counts Several Hundred Orders From Outside AI Agents Against 993 Million in One Quarter
A.I. / news
DoorDash Counts Several Hundred Orders From Outside AI Agents Against 993 Million in One Quarter
Chief Executive Tony Xu gave the figure on the Oct. 29 earnings call, while the company's own release says its strategy stays on delivering physical goods.
DoorDash has received "maybe several hundred orders in total" from third-party AI agents, Chief Executive Tony Xu said on the company's third-quarter earnings call, according to TechCrunch's report of the call. In the same quarter the company processed 993 million orders.
DoorDash published its results on Oct. 29, 2026 in a Form 8-K press release. The release does not contain the agent figure. It appeared only in Xu's answer to an analyst.
What Xu said about agents
Xu said agent ordering is "still a very dynamic area" and that "you never want to fight gravity," but that DoorDash does not want to be "delusional" either, TechCrunch reported. He said most customers still order through DoorDash's own app.
The outside agents named in the report are Muse, Instinct and Fo, plus "others." The report does not say over what period the several hundred orders accumulated, or how many of them were completed deliveries.
Xu also said advertising inside agents could be an opportunity later. DoorDash wants to learn from agent makers and work with them, he said.
What the 8-K says about the same technology
The release lists the agent products DoorDash itself shipped: Ask DoorDash, Text DoorDash, and a Model Context Protocol connector, a standard that lets outside AI agents call a service's functions. The release says the connector reaches external agents, including ones used for corporate ordering.
It then states that AI personal agents are becoming more mainstream, but that the company's strategy "remains focused on winning the battle for atoms." The release names autonomous delivery as an investment area.
- Total orders28 %
- Revenue34 %
- Marketplace GOV36 %
- Net income91 %
Source: DoorDash Form 8-K press release, Oct. 29, 2026
| Q3 2026 measure | Figure | Change on Q3 2025 |
|---|---|---|
| Total orders | 993 million | up 28% |
| Marketplace GOV | $33.9 billion | up 36% |
| Revenue | $4,616 million | up 34% |
| Net income | $466 million | up 91% |
Two figures that do not match
TechCrunch put DoorDash's daily volume at about 3 million deliveries, citing the company's 2025 report. The 8-K's 993 million orders across a 92-day quarter works out to roughly 10.8 million a day, by The Terminal's arithmetic.
The release does not reconcile the two. Either the 3 million figure is out of date or it counts a narrower set of deliveries than "total orders." At either rate, several hundred agent orders is a rounding error: about 0.00003 percent of the quarter's total, if the entire count fell in Q3.
How the agents fit alongside the in-house tools
DoorDash's own assistants are the larger bet. AI Weekly's summary of the call lists Ask DoorDash, a chatbot that takes text and photo prompts, a command-line agent and a text-to-order agent. That summary dates them to June, July and September 2026. The 8-K itself gives no launch dates.
The retail picture is split. TechCrunch reported that Amazon has been blocking personal agents while Walmart has welcomed them. The question of which sites let agents in is the same one the shared robots.txt token draft tries to answer for crawlers, and tools such as Vercel Labs' agent-desktop show the other route, driving an app through its interface when no connector exists.
DoorDash has not published a count of orders placed through the Model Context Protocol connector. Its fourth-quarter guidance, also in the release, is $35.4 billion to $36.4 billion in marketplace GOV, and the next earnings call is the first chance to see whether the agent number moves.
Sources
More in A.I.
- 01John Langford Says He Left AI Frontiers After Microsoft AI ReorgMicrosoft's lab page confirms the Sept. 14 move into Microsoft AI, while Langford's blog post says the new owner's philosophy is scaling and not research.
- 02Nemotron 3.5 Transcribe Posts 4.62 Word Error Rate Across Seven Languages, NVIDIA's Own NumberThe offline speech model has 4.3 billion active parameters, 40 languages and a licence that permits commercial use.
- 03Cloudflare's Security Audit Skill Has an Open Issue Saying a Made-Up Finding Can Pass ValidationThe MIT-licensed skill runs six phases of agent audits. Its tracker shows 43 open issues, including a profile-wipe report and a Windows failure.
- 04Anthropic's Agent Memory Study Puts Opus 5.5 at 0.838 With Memory and 0.655 WithoutThe Nov. 6 post tests notes, wikis and offline dreaming, and reports where a wiki beat search and where it did not.