DoorDash launches Tasks: gig workers film themselves to train AI

DoorDash, which most people know as a food delivery platform, has quietly launched a new product called Tasks. It is an app where ordinary people record video of their daily lives for a small reward. Doing laundry, frying eggs, walking in the park: all of it can now be monetized. A Wired journalist tried it firsthand and reached a discouraging conclusion: Tasks is not a side gig but a window into a bleak future of the gig economy.

The scheme is absurdly simple. The user opens the app, sees a list of assignments and picks one. A task might read: "Record a 3-minute video of yourself cooking in your kitchen." You hit record, do the task, upload the file. Each clip earns a reward, usually a few cents or dollars depending on complexity. The money accumulates and can be withdrawn through standard payment systems.

Who buys these videos? Large technology companies that need training data for AI systems. This concerns computer vision, motion recognition and understanding of household scenes. To teach a neural network how a person fries an egg, you must show it thousands of recordings of real people frying eggs in different conditions, in different kitchens, under different lighting. DoorDash acts as the middleman: it aggregates contributors and sells the result to corporate clients. In essence, it is Amazon Mechanical Turk, but for visual data in the era of generative AI.

The problem is the economics. Whereas DoorDash gig workers could previously earn $15–25 an hour delivering food, before transport costs, Tasks offers significantly less. The journalist, after spending several hours in the app, earned amounts that are hard to call even supplemental income. Meanwhile the barrier to entry is minimal, since all you need is a smartphone with a camera. That inevitably creates competition among a huge number of contributors and pushes rates down. It is the classic trap of the platform economy: the easier it is to join, the cheaper the labor.

The nature of the work is also interesting. Unlike delivery or ride-hailing, Tasks requires no physical travel; assignments are completed at home. That lowers the barrier for people who cannot or do not want to work outside the home. But it also means competitors are not limited by geography: a contributor in Manila or Lagos can complete the same task as a user in New York, for far less money. The globalization of the training data market in action.

There is also the question of privacy. When you film video at home, you inadvertently hand over data about your interior, your habits and any family members who accidentally end up in frame. Where these recordings go, who processes them and how long they are stored: according to the journalist's description, Tasks does not make this sufficiently transparent to contributors.

The market for AI training data is valued in the billions of dollars and keeps growing as major labs compete on the quality and diversity of datasets. Tasks did not create this niche: Scale AI, Remotasks, Appen and CloudFactory have long worked with gig contributors on data labeling. DoorDash simply arrived in this market with a recognizable brand and millions of users in its base. It shows where the platform economy is heading: from physical delivery of goods to delivery of human behavior as raw material for AI. The same mechanisms, the same precarity, a new type of task.

As long as tech companies pay for training data, platforms will find ways to collect it: cheaply, at scale and with the participation of people who need extra income. Tasks is not a distant future. It is the present, already available in the App Store.