DoorDash, best known as a food delivery platform, has quietly entered the AI training data market with a product called Tasks — a gig-work app that pays users to record themselves performing everyday physical activities for use in training artificial intelligence models.
A journalist from Wired downloaded the app and completed several assignments, recording video of themselves doing laundry, scrambling eggs, and walking through a park. Each task arrived with specific instructions about camera angles, movement pace, and environmental conditions — suggesting the data is being collected to tight specifications likely tied to robotics or embodied AI development.
Workers are being paid to help develop systems that may eventually compete with other workers on the same platform.
What Tasks Actually Asks Workers to Do
The app presents workers with a queue of assignments categorised by physical activity type. Accepted tasks require filming the action — often from a first-person or over-the-shoulder perspective — followed by a review period before payment is released.
The pay structure reflects the broader economics of micro-task platforms: individual assignments yield small dollar amounts, and cumulative hourly earnings depend entirely on how quickly tasks are completed and approved. The Wired account does not disclose specific per-task figures, but characterises the compensation as consistent with existing AI data-labelling platforms such as Scale AI's Remotasks or Amazon Mechanical Turk.
DoorDash has not issued public statements about which AI developers or robotics companies are purchasing the data, nor about the number of workers currently enrolled.
Why Robotics Companies Need This Footage
The emergence of Tasks reflects a well-documented bottleneck in AI development: robotics systems require enormous quantities of labelled, real-world human behavioural data, and text scraped from the internet cannot provide it. Video of humans performing physical tasks in authentic domestic and outdoor environments is considerably harder to acquire at scale.
Companies including Google DeepMind, Physical Intelligence, and Figure AI have all publicly discussed the challenge of obtaining sufficient training data for robots intended to operate in unstructured environments — homes, kitchens, warehouses. The specific activities described in the Wired piece — laundry, egg preparation, outdoor walking — map directly onto manipulation and navigation problems considered unsolved in general-purpose robotics.
DoorDash's existing infrastructure gives it a structural advantage here. A large, geographically distributed workforce already accustomed to app-based task assignment and payment allows rapid scaling at lower overhead than building a dedicated research operation.
The Labor Conditions Underneath the App
Workers on Tasks are classified as independent contractors, meaning no benefits, no guaranteed minimum wage across total working time, and no formal protections governing how their recorded data is used or retained. This is not unique to DoorDash — Appen and Remotasks have faced repeated criticism from researchers and labor advocates for opacity around pay rates, unexplained task rejections, and the absence of any worker negotiating power.
The Wired journalist's firsthand account reproduces several of these dynamics: precise instructions that must be followed exactly, payment contingent on approval decisions workers cannot contest, and no disclosed information about downstream data use.
The numbers bear this out. A 2023 study published in the Oxford Internet Institute's journal found that median hourly earnings on leading micro-task platforms — accounting for unpaid time spent browsing and completing rejected tasks — fell below U.S. minimum wage thresholds for a majority of surveyed workers.
DoorDash's Strategic Irony
For DoorDash, Tasks represents a diversification of its core delivery business at a moment of sustained pressure on margins. The company has invested in autonomous delivery technology and has a disclosed interest in reducing its dependence on human couriers over time.
That context sharpens the tension: workers are recording their physical movements to train systems that may eventually displace other workers on the same platform. DoorDash did not respond to Wired's requests for comment on data licensing arrangements, worker volume, or the identities of downstream AI clients.
Regulatory Pressure on the Horizon
The Tasks model is likely to attract scrutiny from labor regulators as it scales, particularly in jurisdictions where gig worker classification and AI data transparency are active legislative concerns — including California and the European Union. The EU AI Act, which entered force in 2024, includes transparency provisions for high-risk AI systems, though its application to training data pipelines remains legally unsettled.
More broadly, the app illustrates how platform-economy infrastructure is being repurposed to solve AI's data problem. The same dispatch logic that sends a worker to pick up a burrito can, with minimal modification, send them to record themselves folding a T-shirt. The labor pool is already assembled. The payment rails already exist.
What does not yet exist — at least not in any form visible to workers completing Tasks assignments — is clarity about what their recorded movements are worth, who profits from them, and whether the people who generated that data will share in any benefit from the AI systems their labor helped build.
What This Means
DoorDash's Tasks app signals that the infrastructure of the gig economy is becoming a primary sourcing mechanism for physical AI training data — and workers entering this market do so without meaningful transparency about the value they are creating or how it will be used.