# DoorDash’s AI shopping agents moved a few hundred orders out of 993 million > Source: > Published: 2026-10-30 19:30:52+00:00 Several hundred. That is the total number of orders DoorDash has received from outside AI shopping agents, cumulated over however long those agents have had access to its marketplace, according to chief executive Tony Xu on the company’s third-quarter earnings call on October 29, 2026, as [TechCrunch reported](https://techcrunch.com/2026/10/30/doordash-said-its-only-received-several-hundred-orders-from-ai-agents/). In the same quarter DoorDash processed 993 million orders, up 28% year over year, and $33.9 billion in gross order value, up 36%, per the [Motley Fool’s transcript of the call](https://www.fool.com/earnings/call-transcripts/2026/10/29/doordash-dash-q3-2026-earnings-call-transcript/). Do the division. Take the generous end of “several hundred” — say 500 — and third-party agents account for one order in roughly 2 million. Take the low end, 200, and it’s one in 5 million. Either way the share of total volume rounds to somewhere between 0.00002% and 0.00005%. That is not an early data point on a growth curve. It is below the noise floor of a platform that handles close to a billion orders a quarter. Xu named the agents customers have actually used: Meta’s Muse, the startup Instinct — backed by Spear Street and valued near $10 billion according to JPMorgan Chase’s own disclosure about restricting it, [per CNBC](https://www.cnbc.com/2026/10/13/jpmorgan-ai-agents-account-access.html) — and Wajo’s Fo. Each takes a plain-language instruction from a user (order dinner, reorder the usual) and then acts on a site or app on the user’s behalf, rather than being DoorDash’s own interface. That these agents can order from DoorDash at all is itself a choice, not a given. Amazon has spent 2026 shutting the same category of software out of its storefront, most recently [blocking Meta’s Muse](https://rews.cc/a/meta-s-muse-ai-agent-shops-and-checks-out-for-users-and-amaz-2f2d46) after accusing it of not disclosing itself and of capturing customer credentials. DoorDash has gone the other way: it operates an API built on the Model Context Protocol, a standard for letting a chatbot call structured actions on a service, that third-party agents use to place orders, and it has not revoked access for Muse, Instinct or Fo. Xu’s framing was deliberately noncommittal. My take on all things technology is you never want to fight gravity. But at the same time, you don’t want to be delusional. **Two routes to a DoorDash order** How AI-driven orders reach the marketplace, per DoorDash's account | From | To | How | |---|---|---| | Customer | Third-party agent *(Muse, Instinct, Fo)* | delegates shopping | | Third-party agent *(Muse, Instinct, Fo)* | DoorDash MCP / API | calls API | | DoorDash MCP / API | DoorDash marketplace | | | Customer | Ask DoorDash / Text DoorDash *(DoorDash's own AI)* | chats or texts | | Ask DoorDash / Text DoorDash *(DoorDash's own AI)* | DoorDash marketplace | builds basket | | DoorDash marketplace | Restaurant or store | order placed | Based on [DoorDash Q3 2026 earnings call](https://www.fool.com/earnings/call-transcripts/2026/10/29/doordash-dash-q3-2026-earnings-call-transcript/) The diagram shows why DoorDash’s own order volume isn’t threatened by letting outsiders in through the side door: almost all of the AI-assisted ordering happening on DoorDash right now runs through DoorDash’s own tools, not a third party’s. DoorDash has been building that in-house stack since June 2026, when it launched Ask DoorDash, an in-app chatbot that can turn a photo of a recipe or a vague prompt like “dinner for a family of four” into a shopping cart. A command-line version followed in July. On September 30 the company launched Text DoorDash, which lets a user place and pay for an order over SMS, no app required, including ordering “the usual” with known modifications remembered from past orders, [per DoorDash’s announcement](https://about.doordash.com/en-us/news/text-doordash). Co-founder Andy Fang described it as working like “a personal shopper that helps you discover and compare thousands of items.” **Ask DoorDash lifts basket size** Increase over the same customers' traditional orders, % | | Increase with Ask DoorDash (%) | |---|---| | Basket value | 50 | | Unique items per order | 60 | Source: [DoorDash](https://about.doordash.com/en-us/news/text-doordash). DoorDash's own figures, self-reported from internal marketplace samples, June–September 2026; not independently audited DoorDash says grocery baskets built with Ask DoorDash come together five times faster than baskets built without it, and that over 30% of Ask-built grocery orders are routine restocks rather than one-off purchases. On the restaurant side, DoorDash says users discovered more than 40,000 restaurants they had never ordered from before through Ask DoorDash in its first three months, with almost half of all Ask-generated restaurant orders going to a spot new to that customer. All of these figures come from DoorDash’s own press materials, drawn from internal samples rather than an outside audit, which is worth keeping in mind before treating a 50% or 60% lift as a settled fact about consumer behavior rather than a company’s characterization of its own feature. The contrast with how other large platforms are treating outside agents is sharp. JPMorgan Chase’s chief executive Jamie Dimon told analysts on the bank’s October 13 earnings call that the bank would not let outside agents operate on customer accounts, naming Muse and Instinct specifically and warning about “rogue agents” that could number in the thousands, [per CNBC](https://www.cnbc.com/2026/10/13/jpmorgan-ai-agents-account-access.html). Charles Schwab has taken a similar line. Amazon has blocked Muse, Perplexity’s Comet and ChatGPT’s shopping features in turn. DoorDash’s position — leave the door open, charge nothing extra for now, see what happens — is the exception, and the volume numbers suggest it costs DoorDash little to be the exception: there’s essentially nothing yet to protect against. [Ars Technica](https://arstechnica.com/ai/2026/10/doordash-says-outside-ai-agents-have-only-led-to-hundreds-of-orders-so-far/), which first reported Xu’s remarks, drew a comparison to Pizza Hut’s first online-ordering experiments in 1994, noting that online orders now make up more than half of the chain’s business. It’s a comparison worth treating the way an engineer treats any two-point trend line: 1994 and 2026 are the only data points, thirty-two years apart, and nothing about an online order form from the dial-up era tells you whether an AI agent placing a few hundred orders a quarter is an early version of the same curve or a feature nobody outside a demo video actually wants. DoorDash has so far declined to charge agents or agent makers anything for the orders they place, even as [other merchants discuss per-transaction fees](https://rews.cc/a/ai-shopping-agents-want-a-fee-on-every-purchase-you-just-don-3e544c) for agentic commerce and even as [questions mount](https://rews.cc/a/your-ai-shopping-agent-counts-as-an-authorized-user-and-so-d-f0f268) over who bears liability when an agent places a bad order on a user’s behalf. Xu said DoorDash sees a future opportunity to advertise inside third-party agents instead. The number to watch next quarter isn’t whether that monetization plan launches — it’s whether “several hundred” becomes a number DoorDash stops disclosing, because it’s grown large enough to matter, or because it hasn’t moved enough to be worth another sentence on an earnings call.