There’s a new side hustle going around, and it pays $15 an hour to record yourself doing your laundry.
DoorDash — yes, the food delivery company — quietly launched an app called Tasks this week. It has nothing to do with delivering burritos. Instead, it pays gig workers to strap a smartphone to their chest and record themselves folding clothes, scrambling eggs, pouring cement, or wandering through a park. That video data then gets fed into AI and robotics training pipelines.
This is the AI data economy in its rawest form, and it’s worth paying attention to.
What DoorDash Tasks Actually Is
The Tasks app works like any gig platform: open it, browse available jobs, pick one, do it, get paid. But instead of dropping off a pizza, you’re generating labeled training data for computer vision models and humanoid robots.
Current task categories include:
- Household chores — laundry, dishwasher loading, repotting plants
- Cooking — frying, poaching, and scrambling eggs (a lot of eggs)
- Handiwork — lightbulb changes, pouring cement
- Navigation — walking through parks, museums, apartment complexes
- Language tasks — natural conversations in Russian, Mandarin, and other languages
Pay is displayed upfront and typically comes out to around $15/hour with hard caps per task. Complete laundry in 90 seconds and you might earn $0.37. Spend 20 minutes in a park and you might top out at $5. A journalist from WIRED completed three tasks and earned less than $10 total.
Why This Matters More Than It Looks
The surface reading is “underpaid gig workers do chores for robots.” The deeper reading is more interesting.
AI models are getting trained on human motion at scale. The whole point of Tasks is to collect physical-world data that robotics developers can’t easily synthesize. A humanoid robot learning to fold laundry needs thousands of real examples with real hands in real environments. DoorDash is essentially crowdsourcing an embodied AI dataset using its existing gig worker network.
This is a structurally different problem than generating text training data. Physical tasks — the way you grip a fork, navigate a doorframe, read a room — are hard to fake. Human-generated video at scale is genuinely valuable. DoorDash is sitting on a distribution channel with millions of registered workers and is now repurposing that channel for a completely different market.
The gig economy’s next frontier is training AI, not serving humans. For years, gig platforms monetized human labor to serve human needs: food, rides, errands. Tasks inverts this. The customer is now an AI company or robotics lab, and the product is human behavioral data. The worker is still doing the same low-paid physical work, but instead of a person on the other end, there’s a model.
The consent and privacy implications are real. DoorDash explicitly blocks workers in California, New York City, Seattle, and Colorado from using Tasks — almost certainly because of those states’ stronger labor and data privacy protections. That’s a tell. Navigation tasks that require recording in public spaces (parks, hotel lobbies, museums) are extremely difficult to execute while following DoorDash’s own “no filming people without consent” rules. The WIRED reporter abandoned a park navigation task after five minutes when a jogger approached.
What This Signals
DoorDash is not the only player here. Platforms like Scale AI, Appen, and Remotasks have been doing AI data labeling work for years. What’s new is a consumer-facing, mass-market app from a household-brand company. It normalizes the idea of doing paid chores for AI.
This moment matters for a few reasons:
The physical AI data shortage is real. Companies building humanoid robots (Figure, 1X, Boston Dynamics, Tesla’s Optimus) all need this kind of training data. Whoever solves the physical data acquisition problem at scale has real leverage.
Platform labor is pivoting to AI infrastructure. As AI automates digital tasks — customer support, data entry, image tagging — the remaining human work is shifting to tasks that require physical presence and embodied intelligence. For now, that work still pays the same pittance.
The economics are unsustainable for workers. $0.37 for a laundry session. Less than $10 for a morning of physical work. The value created here is captured almost entirely by the AI companies and robotics labs downstream, not by the workers who generated it.
The robots are being trained on human labor. And the humans doing the training are being paid in egg money.
This post is based on reporting by Reece Rogers at WIRED (March 21, 2026).