The famous delivery company DoorDash confirmed that reliance on external AI agents to order meals is still in a very early stage and has weak usage, according to statements by company CEO Tony Xu during the company's third-quarter earnings call.
Weak figures for external AI orders
Responding to analysts' questions, Xu explained that AI agents have generated only several hundred orders on the DoorDash platform from their launch until now. He described this volume as an extremely meager use case, especially when compared to the total orders processed by the company, which approached one billion orders during the third quarter of 2026 alone. He noted that while he sees potential benefits for AI agents in complex, long-term purchases that require proactive background work, simple tasks like ordering food usually default directly to the DoorDash app itself.
The novelty of the technology and its first step
This limited performance comes in light of DoorDash's recent launch of the Model Context Protocol (MCP) less than a month ago, which provides a standard and open way for third-party AI applications to order food through the platform easily. This integration is still available to a limited number of early partners and is often used to organize corporate lunch orders, while other agents are forced to deal via traditional web interfaces.
What changes practically?
In contrast, DoorDash's proprietary internal AI tools show a different level of engagement, such as the Ask DoorDash feature launched during the summer to build shopping carts via recipe photos or general descriptions. Although detailed statistics are not available, the company indicates that carts built through this tool are formed five times faster and achieve an average value about 50 percent higher compared to traditional orders, contributing to the discovery of new local restaurants.