How did DoorDash use Tasks to train AI?

DoorDash’s “Tasks” app: paying drivers for training data

DoorDash has launched a new “Tasks” app in some markets that pays delivery couriers for completing short assignments intended to train AI models. Instead of only driving deliveries, participating workers can be asked to submit videos and perform other task requirements through the standalone app.

The underlying shift is that DoorDash is treating everyday on-the-ground activity as an input to machine learning, turning human observations and recorded footage into training signals. That matters because it shows how platforms are moving from traditional “gig work” (logistics execution) toward “AI work” (capturing data for model development).

What the program is described to involve includes:

• Submitting video clips from the field.
• Completing short activities specified by the task prompts.
• Compensation tied to completing assignments rather than delivery completion alone.

This also highlights how training data can be collected at scale: delivery drivers already move through varied environments and encounter real-world conditions, which can produce the kinds of examples AI developers often need for perception, context understanding, and quality control.

For workers and regulators, the important takeaway is that the gig platform is expanding the scope of what drivers may do. Rather than strictly delivering orders, some workers may be asked to record or otherwise provide information that becomes part of an AI training pipeline.

As of the information available in the stories, it’s not fully specified how tasks are selected, what model types they target, or how privacy and consent are handled at the data level. But the core announcement is clear: DoorDash has formalized AI-data collection as a paid app-based gig category.