The Salary Hiding Inside the Robot - by Jaimin

Atoms to Algorithms

SubscribeSign in

The Salary Hiding Inside the Robot<br>Thursday, June 11, 2026 · Logistics & Warehousing

Jaimin<br>Jun 11, 2026

Share

Last post I promised the number that turns a robotics story back into a labor story. Here is the awkward part: nobody will publish it. Starship says its 3,000 sidewalk robots make more than 125,000 road crossings a day, about two per second, with nobody actively watching. Serve Robotics reports 812 robots out delivering on an average day and a fleet closing in on 2 million lifetime deliveries. What neither company discloses is how often a remote human still has to reach in and help. That one hidden number decides whether a delivery robot is a machine that replaces a courier, or a courier with extra steps.

All week we have been inside warehouses, where a stuck robot is somebody else’s problem and help is a few meters away. Today the robot rolls out the door, onto a public sidewalk, into rain and dogs and curious teenagers. Out here it does not have to be brilliant. It has to be cheaper than a person on a bike, every single trip. In the name of last mile problem, companies are trying to figure out strategy that gives them better economics over the period of time.

How it actually works

“Autonomous” does not mean no humans. It means the robot handles everything it recognizes, and when it hits something it does not, a blocked sidewalk, a confused customer, a dead patch of cell coverage, it stops and phones home. The whole business comes down to how often it phones, and what answering costs. Thats where the real cost is right now to the companies which builds this robots (of-course other than the cost of robot army build out).<br>When the call comes in, the response climbs a ladder. At the bottom is monitoring: one person watching a wall of green status icons for a whole fleet, doing nothing most of the time. Next is tele-assistance: the robot waits politely while a human clicks “go ahead” or points to a spot on a map. Above that is tele-driving: a person takes the wheel from a desk, steering through cameras with none of the body’s feel for the street, work that operators describe as surprisingly draining. And at the top, when screens fail, a human in a van drives out to rescue the robot in person. Each step up the ladder costs more human minutes, and human minutes are the one ingredient that never gets cheaper as the fleet grows. This is also kind of exact thing few other companies are trying out at shipping ports, it helps reduce the accidents and allow them to work in good condition in all types of weather, may be thats the story for some other time

The arithmetic is the kind you can do on a napkin. One remote operator earns a salary; that salary gets divided across every delivery the operator’s robots complete. Researchers who interviewed people in this industry found that a typical company today has one operator for roughly every three robots , while the most advanced fleets are estimated to be near one to seven . Stretch one salary across seven robots instead of one and the human cost per delivery falls sevenfold without changing anything about the robot itself.<br>But here is the catch: a company does not get to choose that ratio. Its robots choose it, by how often they call for help. Think of a lifeguard at a pool. One lifeguard can watch twenty swimmers if rescues are rare. If someone needs saving every few minutes, you need a lifeguard per swimmer, and at that point you have reinvented swimming lessons with extra steps. Robot trouble also arrives in bunches: the rainstorm that confuses one robot’s cameras confuses every robot in the city at the same time, so companies have to staff for the worst hour, not the average one. That is why the rate of human interventions per mile is the decisive number in this industry, and why progress compounds so strangely: going from 99 percent autonomous to 99.9 percent is not a one percent improvement, it cuts the salary line ten times over.<br>Since nobody publishes that number, we read shadows. Serve’s latest quarter shows $3 million of revenue against $12 million of delivery costs, meaning each delivery still consumes far more than it earns, though the gap narrowed. Starship claims the opposite shore: robot delivery already $3 to 4 cheaper than a human rider, aiming at roughly $1 per drop. Somewhere between those two claims sits the truth, and the hidden number decides where.<br>New this week

Starship just made the boldest move in the sector: it is shutting down the US university campus business where it spent a decade learning, and redeploying over 1,200 robots to grocery delivery in open cities across Europe and the US. In Finland, the company says about one in five grocery deliveries already arrives by robot. Campuses were the easy mode; this is a public bet that its robots now need help rarely enough to survive real streets at scale (Starship, June 4). Meanwhile Serve’s...