The Frictionless Fallacy: Why AI Agents are Failing the Food Delivery Test

In the world of e-commerce, the 'frictionless' experience is the holy grail. The promise is simple: an AI agent handles the logistics, and the consumer simply receives the goods. But as Ars Technica first reported, the actual adoption of this technology in the food delivery sector is bordering on non-existent.

During a third-quarter earnings call on Thursday, DoorDash co-founder and CEO Tony Xu revealed that third-party AI agents have generated only "several hundred orders" since they began operating. To put that in perspective, DoorDash handled nearly 1 billion total orders in the third quarter of 2026 alone. In the eyes of an operator, this isn't just a slow start; it is, as Xu described it, a "very low use case."

***Opinion:*** *From a commerce perspective, this failure highlights a critical misunderstanding of the 'friction' AI is meant to solve. The friction in food delivery isn't the act of clicking buttons—it's the nuance of the order. Food is an emotional, real-time purchase. When a customer wants a specific substitution or a particular mood for dinner, they don't want an agent guessing in the background; they want direct control. The 'friction' of the app is actually where the decision-making happens.*

Even with the introduction of the Model Context Protocol (MCP) connector—an open-standard tool launched less than a month ago to help third-party AI tools integrate with the platform—the needle hasn't moved. Currently, this connector is limited to a few "early beta partners" and is primarily utilized for corporate lunch orders. Without such a link, agents are forced to navigate the web interface, adding a layer of technical instability to a process that consumers already find intuitive.

Interestingly, there is a distinction between *agentic ordering* and *AI-assisted discovery*. DoorDash's internal "Ask DoorDash" tool, launched this summer, shows significantly more promise. According to reporting from Ars Technica, DoorDash claims these AI-assisted baskets are built "five times faster" and boast nearly 50% higher basket values and 60% more unique items compared to traditional orders. Furthermore, nearly half of these orders are placed at local restaurants the customer has never tried.

This suggests that consumers value AI when it acts as a concierge—helping them discover new foods or turning a recipe photo into a shopping cart—but reject it when it attempts to act as the proxy for the final transaction.

Xu himself seems cautious about the hype, stating that while one shouldn't "fight gravity" regarding technology, one also shouldn't be "delusional." He noted that AI agents may eventually find a home in "long consideration purchases" and major decisions where background work is required. But for the immediate gratification of a meal? The human remains the essential operator. As it stands, the dream of a fully agentic, frictionless dinner is a myth.