To build a complex real-world business, the founding team did every job themselves. This hands-on experience provided critical insights that algorithms or data analysis alone could never uncover, such as knowing not to assign a driver if food isn't ready.

Creating the Dot delivery robot wasn't just a hardware challenge. DoorDash had to build the vehicle hardware, a custom L4 autonomy software stack, integrate them, and then plug the entire system into its complex logistics and merchant platform—a multi-year, first-principles effort.

Instead of focusing on the 'how' (chat vs. voice), DoorDash's AI strategy starts with the 'what': the customer's complete, end-to-end job. For DoorDash, that's getting a physical item delivered. This grounds AI development in solving a real problem, preventing teams from chasing shiny tech without purpose.

As DoorDash's robotics program matures, the primary challenge has shifted. Five years ago, the focus was on making autonomy possible. Today, the harder problems are scaling operations (fleet management, maintenance) and manufacturing reliable hardware at scale.

The adoption of humanoid robots will mirror that of autonomous vehicles: focus on achievable, single-task applications first. Instead of a complex, general-purpose home robot, the market will first embrace robots trained for specific, repeatable industrial tasks like warehouse logistics or shelf stocking.

DoorDash's recruiting pitch for top AI and robotics talent isn't about research prestige, but the opportunity to ship. Many experienced engineers are frustrated with working on demos for years and are drawn to the chance to deploy their work at scale in the physical world.

DoorDash leverages data from its billions of human-powered deliveries to solve the "last 100 feet" problem for its robots. This data, showing precise drop-off locations at complex venues like apartments, provides a competitive advantage over generic mapping solutions.

Xiaomi is developing humanoid robots for internal use in its manufacturing facilities first. This creates a controlled R&D environment and a guaranteed first customer (itself). This 'dogfooding' approach de-risks development and aims to perfect the technology on its own massive operational needs before ever tackling the consumer market.

The true test for scaling robotics isn't the core autonomy but handling mundane, real-world edge cases. DoorDash encountered issues like leaves affecting wheel torque, dirt on camera sensors, and slow boot-up scripts becoming massive productivity sinks when managing a fleet at scale.