DoorDash pilots Ask DoorDash AI chatbot for food and grocery ordering

Ask DoorDash is part of a broader wave of agentic commerce experiments where assistants stop being bolt‑on chat bubbles and start controlling core transaction flows. By letting users describe meals, upload grocery lists or share recipes and then automatically constructing carts and reservations, DoorDash is testing whether customers will trust an AI to directly transform intent into orders. That’s a far richer feedback loop than search clicks; the model learns what people actually buy, in what combinations, at what budgets and with what dietary constraints. ([stechtimes.com](https://stechtimes.com/en/article/doordash-tests-ai-ordering-with-prompts-photos-and-reservation-search-mq9mqcj9))

For the AGI race, these deployments are valuable not because they are “smart” in a sci‑fi sense, but because they expose models to structured decision‑making under real stakes. Every time Ask DoorDash chooses substitutes, balances price vs preference, or recovers gracefully from a mismatch, it is effectively running micro‑scale planning and reinforcement learning in the wild. As Uber Eats and Instacart roll out competing agents, we’ll see a natural experiment in whether tightly scoped, high‑frequency assistant tasks can bootstrap more general reasoning skills. If they work, commerce apps may become one of the most important training grounds for pragmatic, tool‑using agents long before we see anything branded as AGI in the lab.