Dimensional Pays $20 an Hour to Walk Robot Dogs in San Francisco

By Amara Osei

Dimensional, a San Francisco AI startup, is paying people $20 an hour to walk robot dogs through the city, according to The New York Times. The role requires no experience and runs in four- to six-hour shifts.

What does the paid walk involve?

The New York Times reported that Dimensional recently posted the gig to a Bay Area jobs group on Facebook. Walkers steer the dogs while cameras and sensors record each route through San Francisco.

The robot weighs roughly 33 pounds. If its equipment overheats, the walker must carry it back to the office.

That last requirement puts a physical limit on a job built around teaching software. The people doing the walking are also the fallback for moving an overheated machine home.

Why is Dimensional hiring people to guide robots?

The New York Times reported that Dimensional is building an open-source operating system that lets AI agents control robots. It wants those agents to handle manual labor, including hauling heavy boxes and working construction sites.

“It’s honestly barbaric that humans are still doing these physical tasks,” 26-year-old founder Stash Pomichter told The New York Times. His stated ambition is to remove people from those jobs once robots can navigate and operate reliably.

The walkers supply the missing street knowledge. By steering the dogs and recording the outing, they create data for agents that must first learn to move through a city.

That makes the arrangement more than a labor gig. Dimensional uses paid human guidance to test navigation while its operating system learns from the route.

It also exposes the gap between human-directed motion and dependable machinery. A robot steered down a street still has to endure the heat of a four- to six-hour shift.

How large is Dimensional’s robot fleet?

Dimensional owns more than 100 robot dogs, The New York Times reported. The units used on the street are $2,000 models from Chinese manufacturer Unitree.

The hardware is not proprietary, according to the report. Dimensional is building the control layer in software while relying on another company’s physical platform.

More than 100 dogs give the startup repeated opportunities to collect walking data. Unitree’s $2,000 street models also put a clear price tag on each machine in that testing program.

Using nonproprietary hardware makes performance a live constraint on the software rollout. Any failure with the $2,000 Unitree platform directly affects Dimensional’s ability to collect usable navigation data.

What still limits the automation?

Pomichter described the current state bluntly: AI agents are “still very bad at controlling hardware.” The admission keeps the focus on control software, while overheating adds a separate constraint on the hardware.

That admission frames the walking program as necessary infrastructure. Dimensional needs human steering and recorded routes before agents can handle navigation without a walker.

The listing lowers the experience barrier but leaves a physical one. Applicants need not have prior robot experience, yet they must carry roughly 33 pounds if a unit stops during the walk.

The $20 rate sets a clear labor cost for each hour of guidance. Dimensional is paying people not only to move a robot, but to generate the training record for its control system.

Overheating failures also limit how much autonomy Dimensional can claim from a successful walk. A completed outing proves the robot moved under human direction, not that its AI agent can repeat the route alone.

What happens next?

Dimensional’s near-term test is whether recorded walks can improve control enough to reduce human guidance. The company’s larger goal is physical work without people, but the current fleet and founder’s warning point to an earlier stage.

For now, each street walk links three measurable constraints: $20 human labor, a $2,000 robot and an AI agent that still struggles with hardware control. Dimensional must turn that data collection into reliable autonomy before the walkers can disappear.