The DoorDash CLI tool is real, it works, and it just ordered three salads without a human lifting a finger. The company has launched a limited beta of dd-cli, a command-line interface that lets AI agents place food and grocery orders directly through DoorDash’s platform.
Andy Fang, DoorDash’s co-founder and CTO, unveiled dd-cli on Wednesday evening. The tool bypasses the consumer-facing app entirely, connecting AI agents straight to DoorDash’s ordering infrastructure via the command line. Developers in the U.S. and Canada running macOS can apply through a waitlist.
What the DoorDash CLI Tool Actually Does
The beta allows developers to search stores, browse deals, and complete checkout, all through code. The sign-up form asks applicants what they would build with the access, which is either a good onboarding question or an extremely efficient way to crowdsource product ideas. Probably both.
A public demo showed Anthropic‘s Claude completing a full end-to-end DoorDash order autonomously, processing real payments rather than a simulation, with no human input required. AI Weekly reported that the demo showed Claude searching stores, adding items to a cart, and finishing checkout without anyone touching a keyboard.
The video Fang posted on X leans hard into the over-engineering angle. To order those three salads, the agent reads Slack, recalls memories, parses JSON, inspects menu structures, runs Python scripts, recovers from errors, and calculates totals. At one point the interface displays the word “Flibbertigibbeting,” which may be the most accurate description of modern AI pipelines yet committed to a product demo.
Why Developers Should Care About Agentic Commerce
The joke format, part homage to the classic XKCD “sudo make me a sandwich” comic, is doing some heavy lifting here. In that comic, a programmer says “make me a sandwich,” gets told “What? Make it yourself,” then replies “sudo make me a sandwich,” and the other person says “OK.” The gag lands because sudo invokes superuser privileges in Unix systems, making the command impossible to refuse. DoorDash clocked the reference and ran with it.
Underneath the humour, though, the DoorDash CLI tool is a serious infrastructure move. By exposing its ordering platform to AI agents, DoorDash is positioning itself as a building block in the agentic commerce stack rather than just an app you open when you’re hungry. Developers could fold DoorDash’s capabilities into their own tools, their own workflows, or their own AI assistants, without ever routing users through the DoorDash interface itself.
That’s the actual bet here. DoorDash has already been experimenting with surfacing its service through non-traditional channels: ordering via iMessage, its own “Ask DoorDash” AI chatbot, and integrations with OpenAI’s ChatGPT and Claude. The CLI fits the same thesis. If AI agents are going to handle more of the routine friction in daily life, sitting inside those agent workflows matters more than owning the front end.
Whether developers actually build anything interesting with it remains the open question. The waitlist, the macOS-only constraint, and the beta label all suggest DoorDash is still feeling out demand. But the real test is whether the sign-up form’s “what would you build?” field returns answers DoorDash hasn’t already thought of. That answer will tell you more about dd-cli’s future than any product announcement will.
