San Francisco Startup Pays $20 an Hour for Humans to Walk Robot Dogs and Train AI Systems

San Francisco Startup Offers Unusual Job Walking Robot Dogs

AI startup Dimensional has posted an unusual job offer in San Francisco, hiring humans to walk robot dogs for $20 an hour with no experience required1

1

2

2

. The San Francisco startup posted the robot dog walking gig to a Bay Area jobs group on Facebook, though the listing has since been deleted. Shifts run four to six hours, and the company provides paid training with payment after each completed shift. Workers use handheld remotes or mobile applications to guide the machines along designated routes while cameras and sensors record information about their surroundings.

The Physical Challenge Behind the Job

The unusual job offer comes with a notable catch. Workers must be comfortable carrying the roughly 33-pound machines back to Dimensional's office in San Francisco's Dogpatch neighborhood if they overheat or stop working mid-walk1

1

2

2

. The hardware isn't proprietary—Dimensional owns more than 100 robot dogs, and the ones on the street are $2,000 models from Chinese maker Unitree. These four-legged machines can sometimes overheat or stop functioning during outdoor testing, requiring human intervention to complete the mission.

Training AI Agents to Learn Hardware Control

Dimensional is building an open-source operating system designed to let AI agents control robots and eventually perform physical tasks with minimal human involvement. The 26-year-old founder Stash Pomichter told The New York Times that the company wants its technology to help control robots for manual labor in environments such as factories, data centers, and construction sites1

1

2

2

. By hiring humans to walk robot dogs, the startup collects real-world data that helps train AI systems to navigate city streets independently. Pomichter stated, "It's honestly barbaric that humans are still doing these physical tasks," expressing his vision for robots taking on heavy lifting and construction work.

Current Limitations of AI Hardware Control

Despite ambitious goals, Pomichter admits that AI agents remain far from ready for independent operation. "What we determined from our dog-walking experiment was that agents are still very bad at controlling hardware," he told The New York Times, adding, "And humans are still needed in the loop."2

2

Getting AI systems to operate physical machines independently in real-world environments remains a significant challenge. The human walkers serve a critical purpose: they steer the robot dogs while the machines' cameras and sensors record navigation data, helping Dimensional examine how the robots respond to their surroundings and improve robot control and navigation capabilities.

What This Means for the Future of Manual Labor

Dimensional's approach reveals both the promise and current limitations of AI-powered robotics. While the company envisions robots hauling heavy boxes and working construction sites, the reality is that even basic navigation requires extensive human-guided training. The experiment highlights a transitional phase where humans remain essential to teaching machines tasks we hope they'll eventually perform independently. Watch for how quickly Dimensional can reduce human involvement in robot operations—that timeline will signal whether AI agents can truly handle complex physical tasks or if we're still years away from autonomous robots in manual labor settings.