DoorDash CEO: AI Agents Have Placed Only a Few Hundred Orders

DoorDash used its third-quarter earnings call to deliver a notably sober number: orders placed through external AI agents have totaled only a few hundred. Asked by an analyst about the topic, CEO Tony Xu said agent ordering remains a fast-moving area. The company, he said, has no intention of fighting the technology trend, but it also should not be under an illusion about where things stand.

As reported by TechCrunch on October 30 and covered by DigitalToday, that figure sits in stark contrast to DoorDash's overall scale. In its 2025 report the company said it handles an average of about three million deliveries a day. Orders arriving via third-party AI agents are therefore a rounding error, and the overwhelming majority of customers still open the DoorDash app to order.

Xu's framing was not a rejection of agents but a statement about pacing. He said DoorDash plans to understand agent use cases more clearly and to build healthy relationships with outside agent developers, and he noted that advertising inside those agents could become an opportunity later. That points to a view of agents as a potential new distribution surface, not merely another checkout channel.

DoorDash has in fact invested heavily in its own agents. In June it launched “Ask DoorDash,” a chatbot that takes orders through prompts and photos; in July it shipped a command-line interface agent; in September it introduced an agent that accepts orders over text messages. It has also published a Model Context Protocol interface so enterprise customers can place orders from inside other chatbots.

Running its own agents while watching outside ones amounts to a two-track strategy: package ordering as a callable capability, and simultaneously test whether third-party agents generate genuinely incremental demand. Xu's comments on the call suggest the sequence is deliberate — first establish where agents are actually used, then talk about scale.

The value of the few-hundred figure is that it sets a baseline. For the past year there has been abundant discussion of AI agents bypassing apps and transacting on users' behalf, but in food delivery — a high-frequency category with obvious intent — agents have not produced meaningful order volume. The binding constraint may not be technical feasibility at all, but whether people are willing to hand an agent the decision of what to eat tonight.

Three things are worth watching next: how the relationship with outside agent developers takes shape, what form in-agent advertising eventually takes, and whether DoorDash discloses finer agent-order data as its own text, CLI and MCP agents reach more users. Those details will determine whether agents become a new growth curve or remain a long-running demo.