DoorDash launches dd-cli beta for AI agents to order food

DoorDash opened a limited beta of dd-cli, a command-line tool that lets AI agents search stores, compare deals, check menus, build carts and place food orders from the terminal. The release marks a clear turn toward services built not only for people tapping an app, but for software acting on their behalf.

DoorDash described dd-cli as a way to order “directly from your agent,” and one account of the launch said the tool runs as a macOS command-line utility. The pitch is simple and consequential: an agent can now look through restaurants, weigh prices, choose items and complete a real order without a human sitting at the screen for each step.

The announcement traveled quickly across developer circles on X, where Paul Graham reacted to the idea as “agentic food delivery.” That phrase captures where the product sits in the market: not a novelty wrapper around the terminal, but another piece of infrastructure for the agent economy, where the customer may be a model, a script or an autonomous workflow rather than a person.

For DoorDash, the beta extends a platform the company has spent years widening. Founded in 2013, DoorDash says it now operates in more than 30 countries, a reach that gives the company a large base of restaurants, couriers and consumers to expose to machine-driven ordering. The company has also said it is building a collaborative AI ecosystem, signaling that dd-cli fits into a broader strategy rather than a one-off experiment.

The company’s developer stack already includes the DoorDash Drive API for on-demand delivery, though production access is restricted and the developer documentation says there is no guaranteed timeline for certification after development. That limitation matters because it shows how tightly DoorDash is still controlling access even as it opens a new path for automation.

As AI agents start placing food orders, the practical questions move beyond convenience. Restaurants will need to know how menus, substitutions and pricing are interpreted by software. DoorDash will need rules that define who is responsible when an agent misorders, overspends or triggers a delivery dispute. And with a limited beta now in the wild, the company is effectively testing how much of its commerce layer can be handed to machines without loosening platform control.