DoorDash 的自主配送栈：代理商业、Dot 机器人与最后一百英尺

Natural-language agents can reveal latent commerce demand by helping users plan meals, discover products, and complete broad shopping goals.

Agentic commerce works only when merchants expose reliable inventory, pricing, fulfillment, and preference context to software agents.

DoorDash designed its autonomy stack backward from real delivery constraints instead of building a general robot first and searching for a use case later.

Dense suburban delivery requires a vehicle faster and longer-ranged than a sidewalk robot but smaller and more task-specific than a robotaxi.

Autonomous delivery must solve the first and last one hundred feet because packages cannot walk from an inconvenient curb to a merchant or customer.

Real-world fleet data exposes rare operational conditions that simulation and carefully staged demonstrations systematically miss.

Dot is a purpose-built, roughly 300-pound delivery robot designed to travel up to about 20 miles per hour across multiple road environments.

As autonomy improves, fleet operations, merchant integration, maintenance, charging, boot time, and hardware production become the dominant scaling constraints.

DoorDash expects a multimodal network to route each order among human Dashers, robots, drones, and autonomous vehicles according to context.

Enterprise AI adoption improves when teams benchmark models and agent harnesses on real coding, analytics, and operational tasks rather than asking for subjective impressions.