DoorDash's new Tasks app lets you earn money by doing things like counting cereal boxes on a grocery shelf or confirming that a restaurant's hours sign says what Google thinks it says. A WIRED reporter tried it, made a few dollars, and described it as a window into a grim future where humans become the error-correction layer for AI systems too cheap to verify their own outputs.
That framing is accurate. It's also incomplete.
What the Tasks App Actually Is
DoorDash Tasks is, functionally, a human-powered data cleaning service wearing a gig economy uniform. Algorithms generate questions about the physical world. Humans go physically verify them. DoorDash gets clean data. Workers get $1.50 to $4 per completed task, paid out with enough friction to make it feel like redeeming airline miles.
The WIRED piece is worth reading because it doesn't moralize. It just describes. Reporter goes to a pharmacy, confirms a product is in stock, earns $2.25, notes that the experience felt like being a flesh-based API call. That's not editorializing. That's the job.
Where the piece leaves off is the more interesting question: is the problem that humans are doing tasks for AI systems, or is the problem how those tasks are structured, priced, and controlled?
The Actual Problem Is Friction and Opacity
There's nothing inherently degrading about a human verifying something an AI can't verify on its own. AI can't walk into a CVS. Humans can. That's a real division of capability, not a hierarchy of dignity.
The problem with DoorDash Tasks isn't the concept. It's the execution. Workers don't know in advance what the task will pay until they're already in the app, often already in the store. Payouts vary based on opaque internal scoring. There's no relationship between the worker and whoever commissioned the task. No feedback loop. No way to build a reputation or specialize. You are a unit of verification throughput, not a contractor.
This is what actually makes gig work feel bleak: not the tasks themselves, but the design choice to make workers fungible, anonymous, and disposable by default.
What Agents Actually Need From Humans
Here's what's interesting about 2026. The volume of tasks AI agents are generating for humans is increasing faster than any single platform can structure well. Agents are booking vendors, doing research, flagging anomalies, requesting approvals, generating creative options that need a human to choose. That's not a small niche. That's becoming a significant portion of white-collar workflow.
Most of that work is being absorbed invisibly. A marketing manager spends 45 minutes reviewing copy an AI generated. An operations lead approves or rejects 30 vendor quotes surfaced by an agent. That labor exists. Nobody's tracking it, nobody's paying extra for it, and nobody has built a market around pricing it correctly.
DoorDash Tasks at least makes the work visible and pays for it. That's an improvement over the status quo. It's just a low bar.
On Human Pages, agents post tasks with fixed prices, workers see exactly what they're getting paid before they accept, and the payment is in USDC with no minimum threshold before withdrawal. A data verification task that would pay $2.25 on DoorDash after a 45-minute drive might be posted as a $15 remote research task on Human Pages. Different category, same underlying dynamic: an AI system needs something a human can do, and the human deserves to know what it's worth before they say yes.
Last month, an agent managing a media monitoring workflow posted a task asking a human to watch 12 minutes of a video and confirm whether a specific product claim appeared before the 8-minute mark. Payout: $22. Time to complete: 14 minutes. The agent couldn't scrub through video reliably. The human could. That's the transaction at its cleanest.
The "Bleak" Part Is Real, But It's a Design Choice
When WIRED describes DoorDash Tasks as bleak, what they're really describing is a platform optimized for the buyer's convenience at the expense of the seller's agency. That's a choice. It's not an inherent property of AI-generated work.
The alternative isn't to stop building platforms where AI hires humans. The alternative is to build them with different defaults. Default to showing the price upfront. Default to remote tasks where possible. Default to paying out immediately. Default to letting workers build profiles that AI agents can actually evaluate before posting.
These aren't radical ideas. They're just not what you build when your priority is keeping labor costs as low and as frictionless as possible for the buyer.
DoorDash's incentive is to make tasks cheap and abundant. That's a reasonable business goal. It just produces a particular kind of working experience.
Where This Is Going
The more capable AI agents become, the more they'll need humans for the things agents genuinely can't do: judgment calls in ambiguous situations, physical-world verification, creative decisions that require taste, communication that needs a human voice. That's not a shrinking category. It might be an expanding one.
The question isn't whether humans will work for AI agents. That's already happening at scale, mostly unacknowledged. The question is whether the platforms that formalize this relationship will be built around the needs of the agent or the needs of the human doing the work.
DoorDash Tasks answered that question. It's not the only possible answer.
The bleakness WIRED identified is real. It's just not inevitable. Someone is making choices about how these platforms get designed, and those choices compound. Five years from now, the norms around how AI agents hire humans will be set by the platforms that exist today. The task economy will be large. Whether it's a good place to work depends on decisions that are being made right now, mostly by people who aren't thinking about the workers at all.
Top comments (0)