DoorDash Pays Couriers to Train AI with Real-World Video
DoorDash has more than 8 million delivery couriers across the United States. On March 19, 2026, the company launched DoorDash Tasks, a stand-alone app that pays those couriers to complete assignments designed to train AI and robotic systems. Tasks include filming everyday activities while wearing a body camera, photographing restaurant entrances, and recording speech in different languages. The data helps AI systems understand the physical world, and couriers earn money for work that goes beyond delivering food.
How DoorDash Tasks Works
- Couriers download the stand-alone Tasks app or find tasks listed in the existing Dasher app
- Pay is shown upfront and based on effort and complexity
- Example tasks: film yourself washing five dishes with a body camera, photograph a hotel entrance, record multilingual speech
- Video and audio footage trains both DoorDash’s in-house AI models and those developed by partners in retail, insurance, hospitality, and tech
- Available in select U.S. locations, excluding California, New York City, Seattle, and Colorado
Why Gig Workers Are Becoming AI Data Collectors
DoorDash is not the only company tapping gig workers for AI training data. In late 2025, Uber announced plans to let drivers earn extra income by uploading photos to help train AI models. The trend reflects a growing need for real-world, diverse training data that cannot be scraped from the internet.
AI models designed for robotics, autonomous vehicles, and spatial understanding need video footage of how humans interact with physical objects. A dataset of someone washing dishes, navigating a hotel lobby, or describing a restaurant in Spanish has direct value for training embodied AI systems.
“There are more than 8 million Dashers who can reach almost anywhere in the U.S. and who want to earn flexibly beyond delivery. That is a powerful capability to digitize the physical world,” said Ethan Beatty, general manager of DoorDash Tasks.
What Couriers Actually Do
Bloomberg reported that one example task asks a courier to capture footage of their hands washing at least five dishes while wearing a body camera, holding each clean dish in frame for a few seconds before moving to the next. Another involves photographing restaurant menu items so DoorDash can show real photos instead of stock images.
The Waymo partnership, where couriers close the doors of self-driving cars after deliveries, is also listed as a task. These assignments range from simple photo uploads to structured video recordings with specific framing instructions.
The Data Supply Chain Behind AI Products
Training data is one of the most expensive and least visible parts of building AI products. Companies like Scale AI and Appen built entire businesses around data labeling and annotation. DoorDash’s approach is different. Instead of hiring dedicated annotators, it uses an existing workforce that is already distributed across the country and familiar with navigating real-world environments.
For couriers, Tasks offers a new income stream. For DoorDash and its partners, it creates a scalable pipeline of real-world data that would be difficult and expensive to collect any other way. The question going forward is how this data gets used, who profits from it, and whether the pay reflects the value it generates for the companies training their models on it.
