DoorDash Co-founder Andy Fang: Dot Robot is Completing Multiple Real Orders for Unmanned Delivery
DoorDash co-founder Andy Fang stated that the Dot robot is completing multiple real orders for unmanned delivery: after meals are placed in the compartment, the robot checks for missing items before departing. He admitted that the inspection process could be faster and mentioned that daily real orders are allowing Dot to improve rapidly.
Dot is a self-developed Level 4 delivery robot, led by Stanley Tang since around 2018, and has been operating fully autonomously in the Phoenix metropolitan area for over two years. The robot weighs approximately 300 to 350 pounds, stands about 4.5 feet tall, is about 3 feet wide, has a maximum speed of about 20 miles per hour, can carry about 30 pounds, has a range of about 6 to 8 hours, and can navigate lanes, bike paths, and sidewalks, with a size roughly one-tenth that of a car.
Humans are still often needed for the last few feet of delivery: in Arizona, Dashers pick up meals from restaurant counters and place them into parked Dot robots, with a single trip costing about $5 and taking about 5 minutes. The company is shifting its bottleneck focus from algorithms to fleet operations, merchant integration, and mass production, and is collaborating with Rivian's micro-mobility spin-off Also to manufacture vehicles. The platform uses an autonomous delivery system to dispatch orders among people, Dot, drones, and other partners based on speed, cost, and location.
The data barrier is referred to as the "first and last 100 feet": over 10 billion deliveries have recorded where people actually placed the meals, rather than just map pins. In agent-placed orders like those from DoorDash, about half of the restaurant trips are directed to merchants that have never been ordered from before, with grocery baskets being about 40% larger. The company claims that in ten years, multimodal capacity will lead to more orders, and the number of Dashers will not decrease. The costs for users are the same as those for human delivery.
In terms of market mechanisms, this is about training with real orders to replace people from long-distance rides: what is bought is the reduced unit cost of predictable mid-range routes, while what is sold is the segment that still requires humans for rainy days, apartment access, and item verification. Beneficiaries are platforms that need to lower peak subsidies; those under pressure are riders who rely on mileage for income and merchants who must redesign their pickup flow. Until the inspection process speeds up, the robot is not yet a complete closed loop.
Additionally, DoorDash Air has obtained FAA Part 135 certification for drone deliveries over 3 to 5 miles; the early goal for Phoenix was to cover about 1.5 million people. Internal AI spending has leveled off after increasing about 20 times in the first half of the year.
Source: Public Information
ABAB AI Insight
Fang and Tang have treated the delivery robot as an internal project since 2018, initially testing pedestrian robots and outsourced Robotaxis, then determining the need to create vehicles that can operate on roads. Dot is not a laboratory prototype showcase, but rather uses daily hot meal orders in Phoenix as road test data. The acknowledgment of slow item checks indicates that Level 4 has passed the road, not the merchant counter. Paying Dashers to move meals from the counter to the robot's compartment effectively buys back the "last 100 feet" temporarily.
Capital is shifting from subsidizing riders to vehicle manufacturing and infrastructure. Also is responsible for expanding the fleet from dozens of vehicles to a manageable operation, while ADP is tasked with assigning each order to the cheapest mode. The 10 billion data points allow DoorDash to understand better where to park for apartments and whether fast food should be picked up from the front door or the lane. Agent-placed orders bring new merchants into the funnel, while robots handle the incremental orders that people are less willing to run.
In contrast to Starship pedestrian boxes, Nuro's closed community vehicles, Amazon's Scout suspension, and Serve Robotics' narrow expansion post-IPO. The industry phase is characterized by software claiming to be sufficient, while hardware and merchant interfaces are not. Humans will not disappear immediately but will be transformed into loaders and exception handlers; the ten-year prediction of "more Dashers" is based on the expansion of order volume due to lower unit prices, not on rising wages.
The structure belongs to labor reconstruction within technological substitution. Long-distance driving is the most expensive and best standardized, so it is the first to be replaced by vehicles; the most expensive coordination from the counter to the roadside and from the address to the doorstep still requires humans temporarily. The mechanism is that the platform, possessing demand density and delivery truth, has the ability to train a closed loop; those without daily hot meal orders can only demonstrate in a controlled environment.
ABAB News · Cognitive Laws
Autonomous driving first navigates the roads, ultimately getting stuck at the counter
Real orders are training data, demonstrations are not
Robots replace mileage, not all humans