DoorDash Now Pays Couriers to Film Themselves Training AI

DoorDash announced this week that it will pay couriers to film themselves performing household tasks, with the footage used to train artificial intelligence systems and robotics. The new app, called Tasks, is available to the company's 8 million couriers across the United States, though workers in California, New York City, Seattle, and Colorado are excluded from the program.

The company's announcement describes Tasks as offering workers body cameras and paying them to record activities including washing dishes, folding clothes, scanning supermarket shelves, photographing hotel entrances, and recording multilingual audio. Pay is shown upfront and varies based on what DoorDash calls 'effort and complexity.' The footage is then provided to DoorDash's in-house AI development teams and to corporate partners across retail, insurance, hospitality, and technology sectors.

The goal of Tasks is to help more businesses understand what's happening on the ground and gather new insights, all while giving Dashers a new way to earn on their own terms.

— Ethan Beatty, General Manager, DoorDash Tasks

The excluded markets are notable: California's AB5 law creates stricter tests for independent contractor classification, New York City has minimum wage requirements for delivery workers, Seattle has similar protections, and Colorado has strengthened worker rights legislation. DoorDash appears to have concluded that offering Tasks in these jurisdictions would create legal exposure, suggesting the company itself recognises the program sits in regulatory grey areas.

The gig economy becomes a training layer

DoorDash is not the first platform to recognise that its distributed workforce represents a valuable source of training data for AI systems. Uber launched a similar initiative in October 2025 through its Uber AI Solutions division, paying drivers to collect visual and audio data that feeds into the company's autonomous vehicle development and other machine learning projects.

The timing of DoorDash's announcement aligns with broader industry developments in robotics training, and on March 16, 2026, Universal Robots and Scale AI announced a joint imitation learning system that trains robotic arms to perform physical tasks by watching human demonstrations. The demand for this type of training data is substantial and growing. Companies developing embodied AI systems need millions of hours of footage showing human hands performing everyday activities, and gig workers represent a scalable, on-demand source of exactly this content.

The commercial logic follows a familiar pattern for platform companies seeking to extract additional value from their existing networks. Gig workers are already distributed across residential areas, already have relationships with the platform, and can be directed to perform specific tasks through app-based instructions. Converting them from delivery workers into data collectors requires minimal infrastructure investment beyond the body cameras and an updated app interface.

Who owns the training data?

The legal framework for this type of work remains underdeveloped. When a courier films themselves folding laundry and that footage is used to train a robotic system that might eventually fold laundry in warehouses or homes, questions arise about ongoing rights to that contribution. Standard work-for-hire arrangements in traditional employment typically vest all intellectual property in the employer, but gig workers are classified as independent contractors.

DoorDash's terms of service will almost certainly include broad assignments of any intellectual property rights in the captured footage, but whether such assignments would survive scrutiny under various state laws is untested. The Federal Trade Commission has shown increasing interest in data practices by large technology platforms, and the use of gig worker footage to train AI systems that may reduce demand for human labour could attract regulatory attention.

European regulators have been more active in this space, and the EU's AI Act includes provisions around training data transparency that would require companies to disclose the sources of data used to train AI systems. If DoorDash or its partners deploy AI trained on Tasks footage in European markets, they may face obligations to document and potentially compensate the workers who provided the underlying training data.

Training your replacement

Labour advocates have raised concerns about the broader implications of programs like Tasks, and the Gig Workers Collective, a US-based organisation representing platform workers, described the program as asking workers to 'train their own replacements.' The comment reflects anxiety within the gig economy about the trajectory of automation and whether today's flexible work arrangements are a transition phase toward workforces that are substantially smaller and more automated.

These concerns reflect observable trends in robotics development, not hypothetical scenarios. The same robotics companies purchasing training data from gig platforms are developing systems designed to perform physical tasks currently done by human workers. A dishwashing robot trained on footage collected through Tasks might eventually reduce demand for human dishwashers. A shelf-scanning system trained on supermarket footage might replace stocktaking workers. The workers providing the training data are, in effect, contributing to systems that could eliminate their own future job opportunities.

DoorDash would likely respond that Tasks represents additional earning opportunities for its couriers, not a threat to existing work. The company's announcement emphasises worker choice and the ability to earn on their own terms. But the structural dynamic remains: platforms are monetising their workforce not just for the labour they perform today, but for the data they can generate to enable automation tomorrow.

The regulatory gap

Current US labour law was not designed for work that involves generating training data for AI systems. The Fair Labor Standards Act addresses wages and hours but does not contemplate ongoing royalties or residual rights for workers whose contributions enable machine learning. The National Labor Relations Act protects collective bargaining but gig workers are typically excluded from its coverage as independent contractors.

State-level legislation has begun to address some aspects of platform work, but the specific question of AI training data rights remains largely unregulated. California's consumer privacy law, the CCPA, gives residents some rights over their personal data, but whether filmed demonstrations of household tasks constitute personal data under the law is unclear. The footage shows hands and actions, not faces, and may not fall neatly into existing privacy categories.

Congressional interest in AI regulation has focused primarily on safety, bias, and national security implications of advanced AI systems. The question of how workers who generate training data should be compensated or protected has received less attention, though the SAG-AFTRA strike of 2023 did establish some precedents around AI training data rights in the entertainment industry that could influence future legislative approaches.

What happens next

DoorDash's Tasks program will likely expand in coming months, potentially adding new types of data collection and new corporate partners. The excluded markets may become a pressure point: if the program proves profitable, DoorDash may seek to extend it into California and other currently excluded jurisdictions, potentially triggering legal challenges that would clarify the regulatory boundaries.

Other gig platforms are monitoring the program's performance and may launch similar initiatives. If Tasks succeeds commercially, expect similar programs from Instacart, Grubhub, and other delivery platforms. The gig economy's workforce represents an attractive asset for companies developing AI systems, and the platforms that can effectively monetise this asset will gain competitive advantage.

For workers weighing participation, the calculus involves balancing immediate income against uncertain long-term consequences. Tasks offers immediate income for relatively simple activities, but the longer-term implications of contributing to AI training datasets are abstract and uncertain. In a gig economy where income is unpredictable and workers have limited bargaining power, the offer of additional paid work will be attractive regardless of downstream consequences. That asymmetry between immediate benefit and long-term risk is familiar in labour markets, and regulators have historically intervened when it becomes sufficiently pronounced.