DoorDash’s agent-commerce reality check arrived during its third-quarter earnings call: outside AI tools have produced only “several hundred” orders in total, according to co-founder and CEO Tony Xu.¹ For a platform that says it completes roughly 3 million deliveries a day, the figure barely registers.

Xu’s message was neither dismissal nor surrender. “You never want to fight gravity,” he said of technology shifts. “But at the same time, you don’t want to be delusional.”¹ The immediate evidence, he said, is that customers still overwhelmingly come straight to DoorDash’s app for food.

That gap reflects the practical limits of the current rollout. DoorDash’s Model Context Protocol connector, which lets selected third-party AI services place orders through the platform, was introduced only recently and is available to a small group of early beta partners, with corporate lunch ordering among its main uses.² Outside agents otherwise face a more cumbersome route through the web.

DoorDash has not waited for third parties to prove the model. It introduced a CLI agent in July, launched its text-to-order tool in September, and rolled out Ask DoorDash over the summer. The chatbot can build a cart from a recipe photo or make suggestions from a broad request, such as dinner for a family of four.¹

The company has disclosed no comparable order totals for those internal tools, but says Ask DoorDash carts are built five times faster, with orders carrying nearly 50% higher basket value and about 60% more unique items than traditional orders from the same consumers.²

The contrast is stark: third-party agents have yet to move volume, while DoorDash sees signs that its own AI can improve shopping. Xu’s longer-term case is that agents may prove more useful for high-consideration purchases, where they can work proactively in the background—not for the quick, habitual task of ordering dinner.²