DoorDash has revealed that outside AI agents have placed only several hundred food orders on its platform in total, a strikingly small number for a company that completes an average of three million deliveries a day. Chief executive Tony Xu shared the figure on the company’s third-quarter earnings call, in answer to an analyst question about agentic commerce. The disclosure offers a rare, concrete measure of a much-hyped trend, and it suggests that agent-led ordering remains an experiment rather than a shift in how people actually buy dinner.
What Xu told analysts
Xu described agent ordering as a fast-moving area and said DoorDash did not want to fight the direction of technology. He warned, however, against overstating where it stands today, and said the company had seen perhaps several hundred orders in total from third-party agents.
He named Muse, Instinct and Fo among the agents involved, and stressed that the total covers all outside agents combined rather than a single service. That makes the gap with DoorDash’s daily volume wider still, and it frames agent ordering as a niche activity for now.
The chief executive added that customers still place orders predominantly through DoorDash’s own app, rather than handing the task to an outside agent. He said the company wants to learn more about agentic use cases and to build a healthy relationship with the makers of third-party agents.
Looking ahead, he said there could be an opportunity for DoorDash to advertise inside those agents, turning a future channel into revenue. That hint suggests the company sees agents as a future channel to monetize, rather than as a threat it must block.
DoorDash’s own agents
DoorDash is not standing aside from agent technology while it waits for outside agents to grow. The company launched a chatbot called Ask DoorDash in June that lets people place orders through prompts and photos, and it introduced a command-line agent in July.
In September it added a text-to-order agent, giving customers another way to order without opening the full app. The company has also released a Model Context Protocol tool that lets other chatbots handle ordering for corporate customers directly and at scale.
That step matters because it separates two different agent markets: consumer agents that order on a person’s behalf, and business tools that order for a workplace. DoorDash’s own agents already cover both sides, which may explain why it can afford to be relaxed about the slow start of third-party agents.
Why rivals are split on agents
The wider industry has not agreed on how to treat shopping agents, and the split is now visible in company policy. Amazon has been blocking personal agents from its storefront, treating them as a risk to its control over the customer relationship and its product pages.
Walmart, by contrast, has welcomed agents, laying out tools for them in the hope of winning business from competitors that shut them out. That disagreement mirrors an older question about who owns the customer when a machine does the shopping.
Retailers that block agents keep the checkout, the recommendations and the data, while retailers that welcome them bet that being first in the agent’s answer matters more. DoorDash sits between those poles, neither blocking agents nor depending on them, and its numbers give the debate something it lacked: an actual count.
Yelp has taken a similar middle path with its own phone agent for booking services, showing that established platforms are testing agents carefully on their own terms rather than surrendering the customer experience to outsiders before the technology is ready.
Why It Matters
The figure matters because agentic commerce has been sold as the next platform shift, with forecasts of agents booking travel, buying groceries and ordering meals at scale. DoorDash’s several hundred orders, set against three million deliveries a day, show how far reality lags that pitch.
For investors and founders, the lesson is to ask for volumes before believing a category has arrived. It is also worth distinguishing between an agent that can technically place an order and one that customers actually choose to use every single day.
For retailers, the split between Amazon and Walmart now has a data point behind it. Blocking agents costs little while volumes are tiny, and welcoming them costs little too, which means both strategies can survive for now without much risk.
The real test will come when an agent can reliably handle substitutions, dietary restrictions, delivery windows and refunds without a human stepping in to fix mistakes. Until then, DoorDash’s disclosure suggests the agent economy is still in its demonstration phase.
Background
DoorDash disclosed the agent order figure during its third-quarter earnings call on October 30, 2026, in response to an analyst question. The company reported an average of three million deliveries a day in its 2025 report, the comparison used to put the several hundred agent orders in context.
Source: TechCrunch report
0 Comments