DoorDash has seen only 'several hundred' orders from AI agents

DoorDash has received only "several hundred" orders from third-party AI agents, the company disclosed on its third-quarter earnings call. That number sits next to an average of 3 million deliveries per day, which DoorDash cited in its 2025 report. In other words, agent-placed orders make up a fraction of a fraction of the company's volume.

The gap matters because the industry has spent the past year talking up AI agents as the next interface for commerce. DoorDash's own numbers are the first hard, named figure from a major delivery platform on how much agentic ordering is actually happening. The answer, at least for now, is almost none of it.

CEO Tony Xu put the figure on the record while answering an analyst's question about agents such as Muse, Instinct and Fo. His framing was cautious rather than dismissive.

What did Tony Xu actually say?

Xu gave the quote verbatim on the call:

"I would say that this [using agents to order] is something that is still a very dynamic area. And 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."

He then quantified the "delusional" part directly:

"What I mean by that is if you look at where AI agents are today, we're seeing maybe several hundred orders in total from some of these third-party agents."

The takeaway in plain words: Xu is not betting against agents, but he is not pretending the demand is here today either.

The "several hundred" figure covers orders placed through named third-party agents including Muse, Instinct and Fo, according to the company. That's the total across all of them — not per agent, and not per month. DoorDash did not break the number down further on the call.

For scale, DoorDash completes an average of 3 million deliveries every day, per its 2025 report. Even if every single agent-placed order arrived in one day, it would still amount to a vanishingly small share of that daily volume.

Where do DoorDash's orders actually come from?

Users are still predominantly placing orders through DoorDash's own app, Xu said. That detail is easy to skip, but it carries most of the story's weight.

The app is the incumbent interface, and it is winning by several orders of magnitude. Whatever momentum agent-based ordering has in the broader AI conversation, DoorDash's customers have not moved. The takeaway: the app remains the front door, and no agent is close to challenging it.

That said, Xu signaled openness rather than resistance. He said DoorDash wants to learn more about agentic use cases and develop a healthy relationship with the makers of third-party agents.

He also floated a commercial angle. In the future, Xu said, there could be an opportunity for DoorDash to advertise within these agents. That is a notable positioning move: instead of treating agents as a threat to the app, DoorDash is exploring them as another ad surface.

The takeaway: DoorDash is choosing engagement over gatekeeping — and looking for revenue in the process.

What has DoorDash built itself?

DoorDash is not waiting for third-party agents to define the space. The company has shipped its own set of agent products across 2025:

• A text-to-order agent, launched in September
• A CLI agent, introduced in July
• Ask DoorDash, a chatbot launched in June that lets users place orders through prompts and photos
• An MCP (Model Context Protocol) release, which enables ordering from other chatbots for corporate ordering

That lineup shows a company hedging across interfaces. Text, command line, chat and protocol-level integration each cover a different way an AI system might reach DoorDash's marketplace.

The corporate-ordering MCP is the most structurally interesting piece. By publishing a protocol, DoorDash makes itself pluggable into chatbots it does not control — the same channel through which Muse, Instinct and Fo currently send their handful of orders.

The takeaway: DoorDash has built agent infrastructure on its own terms, even while third-party agent volume stays near zero.

Why are websites split on AI agents?

The backdrop to Xu's comments is a live fight over access. The space of agent-based ordering is still evolving, and websites are still deciding whether to let these agents reach their content at all.

The industry has split into two camps, and the contrast is stark:

• Amazon has been blocking personal agents from its platform
• Walmart has laid out a red carpet for them, in an effort to snag business from competitors

Same technology, opposite strategies. Amazon treats agents as a threat to its interface and its customer relationship. Walmart treats them as a customer-acquisition channel — a way to intercept shoppers who might otherwise default to a rival.

DoorDash, based on Xu's comments, sits closer to the Walmart camp: accommodate the agents, learn from them, and potentially monetize them through advertising. The takeaway: the fight over agent access is really a fight over who owns the customer relationship.

What does this mean for agentic commerce?

The gap between the hype and DoorDash's "several hundred" orders is the story. For a year, AI agents have been pitched as the future of how people buy things — ordering dinner, restocking groceries, booking services, all without opening an app.

DoorDash is the first major delivery platform to put a public number on that future, and the number is tiny. Three million deliveries a day versus several hundred agent orders total is not a transition in progress. It is a proof-of-concept stage at best.

But Xu's "you never want to fight gravity" comment cuts both ways. He is saying the technology's pull is real, even if the usage isn't. His company's own product cadence — a June chatbot, a July CLI agent, a September text-to-order agent, plus an MCP for corporate ordering — shows DoorDash preparing for a shift it cannot yet measure in its order logs.

The forward question is whether agent volume grows fast enough to justify the advertising opportunity Xu floated, or whether the app stays dominant and the agents stay a rounding error. Amazon's blocking and Walmart's welcoming show there is no industry consensus on the answer. DoorDash's next earnings call — and whether "several hundred" has grown — will be one of the clearest public signals to watch.