Eight cents. That is what DoorDash pays gig workers per task on its new Tasks app—roughly $0.08 to record yourself folding laundry, scrambling eggs, or walking around a park. The setup takes 15 minutes, the task another 10, and the humiliation less than a minute to set in.
DoorDash Tasks is a micro-labor platform where gig workers perform and record household activities to generate training data for embodied AI systems—robots designed to eventually do these same tasks autonomously. The premise is logical: if robots will fold your shirts, someone must first record thousands of humans doing it. That someone, apparently, costs $0.08.
The math reveals the design. When you divide the pay by the time required, DoorDash's effective wage comes to less than a dollar per hour—approximately $0.08 per task at the rates workers reported to Wired. This is not a glitch. It is the intended outcome of a system built to extract the cheapest possible training data from workers with the fewest alternatives. The platform frames these rates as "fair market value," but fair market value is meaningless when supply vastly outstrips demand and workers are desperate for any income.
The investors funding this model understand what they are building. Embodied AI—the category of systems that interact physically with the world—requires vast quantities of human movement data to function. Hiring professional data labelers at living wages would cost millions. Outsourcing the work to gig workers at poverty rates costs thousands. The difference is not efficiency. It is the deliberate externalization of labor costs onto workers and the social safety net.
The pattern extends beyond DoorDash. Amazon Mechanical Turk, Appen, Scale AI—all operate on variations of this model, paying pennies for tasks that generate enormous value for the companies that commission them. DoorDash Tasks simply adds a layer of physical labor to a template already proven in cognitive gig work. The workers are not merely annotating images or transcribing audio; they are producing the raw material from which future robotic systems will be built.
The cruelty is structural. Workers creating this training data may be training the systems that will automate their own jobs. Each folded shirt recorded is a step toward a robot that folds shirts without humans. Each scrambled egg filmed is footage for a machine that replaces the worker holding the spatula. This is not a cycle—it is a machine, and gig workers are both its fuel and its intended output.
What makes this "innovation" rather than exploitation is never made clear. The language of disruption and technological progress does not change the underlying transaction: workers bear the risk, bear the labor, and receive a fraction of the value they create. The companies receive data, build moats, and scale systems that reduce dependence on human labor. This is the gig economy perfected—not for workers, but for the platforms built atop their labor.