DoorDash pays delivery drivers to train AI and robots

If you thought delivery couriers only moved burritos and groceries, DoorDash wants you to think again. The company is now paying its drivers to strap on body cameras and do the dishes.

On Thursday, DoorDash officially unveiled DoorDash Tasks alongside a pilot for a new, standalone app. The idea is straightforward: turn the company's fleet of more than 8 million delivery workers into an on-demand workforce that collects real-world footage and audio to train AI models and robotics systems.

From food delivery to physical data collection

Training frontier language models on internet text has hit its limits. If companies want robots and multimodal AI to actually understand how the physical world works, they need messy, first-person data from real life. DoorDash happens to have millions of people already walking through that physical world every day.

Couriers can pick up short gigs between deliveries or whenever they have spare time. Pay is shown upfront and varies based on how much effort and time the job requires.

Some of these tasks have actually been running quietly inside the main Dasher app for a while. Drivers have been photographing restaurant dishes, mapping tricky hotel entrances, and even helping Waymo self-driving vehicles by manually shutting open doors. According to DoorDash, workers have completed over 2 million of these micro-tasks since 2024.

The new standalone Tasks app dials this up specifically for AI labs. Bloomberg reported that one assignment asks a courier to put on a body camera and film their hands washing at least five dishes, pausing to hold each clean plate in front of the lens for a few seconds. Couriers can also earn money by recording speech in different languages. DoorDash says the data will feed both its internal models and external partners across retail, hospitality, insurance, and tech.

The feature is currently live in select US regions, but conspicuously excludes California, New York City, Seattle, and Colorado—places where gig-economy labor laws are notoriously strict. DoorDash isn't working in a vacuum either; Uber announced a similar initiative late last year to let drivers snap photos for AI training.

Why it matters

For delivery drivers, this is an easy win during slow hours. Instead of idling in a parking lot between lunch and dinner rushes, they can pick up micro-income doing odd jobs on their phones. For AI startups, this solves a brutal bottleneck: gathering first-person physical data used to require expensive, specialized staging. DoorDash can do it at a fraction of the cost simply by redirecting a fraction of its fleet.

It also highlights where the AI race is heading. The battleground is shifting from web scrapers to embodied physical intelligence. The platforms that win might not just be the ones with the best compute clusters, but the ones that control millions of pairs of eyes and hands moving through the real world.

We think DoorDash's move is clever, but it walks a fragile line. Turning a delivery workforce into a distributed data lab is brilliant unit economics, but quality control will be a headache. How do you stop workers from half-assing a dishwashing video just to grab a quick payout? More importantly, the deliberate exclusion of California and New York proves DoorDash knows the regulatory ice is thin. If courts decide these micro-tasks look too much like hourly labor subject to minimum wage guarantees, the math behind this crowd-sourced data machine could fall apart fast.