The gap between agent hype and agent economics just got a number: several hundred orders versus 3.2 billion.

The Summary

The Signal

Several hundred orders. That's the entire AI agent economy for DoorDash right now. Tony Xu dropped this reality check during the company's Thursday earnings call, putting a hard number on what agent maximalists have been promising for months. Against DoorDash's 3.2 billion deliveries last year, agent orders round to zero.

The story isn't that agents don't work. It's that humans don't trust them yet. Business Insider found users willing to let agents track prices for holiday gifts or flights, but they stop short of handing over purchase authority. When they do, they use single-use credit cards, the digital equivalent of a very long stick.

"For the everyday ordering, what we see is people come to DoorDash, and they come back to DoorDash."

The trust gap makes sense. Ordering food requires context agents don't have yet:

  • Dietary restrictions that change based on who's eating
  • Address nuances like "side door" or "don't ring the bell"
  • Tip calculations that factor in weather, distance, and whether you're feeling generous

These aren't technical problems. They're delegation problems. People will let an agent book a meeting because the downside is mild. They won't let it order dinner because the downside is explaining to their kid why the pizza has mushrooms.

DoorDash's response is telling: build your own agents instead of waiting for Meta's to figure it out. The company is testing text-to-order features that keep the human in the app but reduce friction. Smart. If agents are going to reorder your Tuesday burrito bowl, they should be agents you trained, not agents guessing.

The Implication

Watch what DoorDash builds, not what third-party agents promise. The winning agent play for commerce isn't a universal agent that orders everything. It's category-specific agents that know your burrito order cold because you've trained them inside the app you already use. The trust barrier drops when the agent lives where the transaction happens, not in some Meta layer above it.

For agent builders, the lesson is clear: stop pitching universal purchase authority. Start with visibility and suggestions. Earn trust in low-stakes decisions before asking for the credit card.

Sources

Business Insider Tech | TechCrunch AI