Director of Product & Business Operations · 2018 to 2021
I owned the product for Postmates' sidewalk robots, now Serve Robotics
Serve Robotics started inside Postmates and spun out as its own company. I owned the product: the autonomy spec, the tools that let one operator watch 8 robots, and the handoff from merchant to customer. The robots made 1,000+ deliveries, and I launched them in 5 markets.
Serve Robotics2018 to 2021
- one operator for 8 robots
- 1:8
- deliveries by robot
- 1,000+
- markets launched
- 5

Where things stood
Deliveries under a mile lost Postmates money on every order. Sending a car and driver for a burrito made no sense, and couriers avoided short, low-paying trips, so driver churn hit 40%. No one had built the whole experience of a robot delivery. How does a merchant hand over an order, and how does a customer get it out?
Calls I made
I scored routes by how robots behaved on them
I first routed robots the way we routed cars, on map data, and assumed permit zones would hold. Then the city changed its curb-cut designations and 40% of our approved routes broke overnight. The fallback sent robots up steep grades and past construction, and two got stuck in three days. So every time a robot slowed or paused for no clear reason, we tagged the spot and lowered that route's score. Dispatch avoided low-scoring routes even when they were legal.
What happened: Once the route scores settled, one operator could watch 8 robots.

A Serve robot on a busy sidewalk. Picture: Serve Robotics One operator, eight robots
Showing: one operator watching 8 robots
1:8
Operator, remoteRobot
Once the route scores settled, one operator could watch 8 robots.
The merchant puts the order in the robot.
The robot drives itself. An operator steps in remotely only when it meets something it can't handle.
The customer unlocks the lid and takes the order out. I redesigned both ends through user research, and pickup time fell 35%.
I set the bar a robot had to clear before it left the lot
I owned the Level 4 sidewalk autonomy spec. I wrote the test sets the perception team worked against, scenario by scenario, including blocked views and sidewalk shapes the robots hadn't seen. No robot left the depot until those tests passed.
What happened: Operators stepped in remotely on the rare cases the model couldn't handle on its own.

A Serve fleet from above. Picture: Serve Robotics I designed the handoff at both ends
A merchant had to load the robot and a customer had to unlock it, and no one had designed either. I owned the whole delivery from merchant handoff to customer unlock, and redesigned it through user research.
What happened: Pickup time fell 35%.

Loading an order into a Serve robot. Picture: Serve Robotics
What changed
- one operator for 8 robots, once route scores settled
- deliveries by robot
- markets launched
- faster pickups