techcrunch.com web signal

DoorDash opens dd-cli beta for AI agents to place orders

agents openai anthropic ai-business

TL;DR

  • DoorDash launched a limited beta of dd-cli, a command-line tool that lets developers and AI agents search stores, find deals and check out.
  • Access is gated to macOS developers in the U.S. and Canada via an invite-only waitlist, with applicants asked to submit intended use cases.
  • A launch demo showed Anthropic's Claude using dd-cli to pick a restaurant, add items to a cart and complete checkout without human intervention.

A food delivery company shipping a command-line tool sounds like a joke, and DoorDash is leaning into that reading. The announcement nods at the famous XKCD comic about programmers automating mundane tasks with sudo. But the substance is that DoorDash has quietly published a real transactional interface that a language model can drive end to end, and TechCrunch reports the demo alongside the announcement showed Anthropic's Claude searching, adding to a cart and completing checkout without human intervention.

The rollout itself is deliberately narrow. Andy Fang, DoorDash's co-founder and CTO, announced dd-cli this week as a limited beta gated by an invite-only waitlist, restricted to macOS developers in the U.S. and Canada. Applicants have to submit the use case they intend to build. Read as product distribution, that is tiny. Read as positioning, it stakes an early claim on being the commerce endpoint an AI agent reaches for when it needs to buy food.

The strategic move is treating the CLI as the primary agent surface rather than an afterthought on top of the app. DoorDash has already tried ordering through iMessage, launched an in-house chatbot called Ask DoorDash, and wired itself into ChatGPT and Claude. dd-cli is the natural next step in that arc, giving the agent a scriptable, deterministic interface instead of asking it to click through a mobile UI. If agentic commerce turns out to be real, being the default endpoint for one of the biggest recurring transaction categories in a person's life is a valuable place to sit.

The honest caveat is that the reporting is thin on the parts that will actually determine whether this matters. It does not spell out how dd-cli handles authentication and spending limits, whether it rides on an open agent protocol standard or something proprietary, or who eats the loss when an agent orders the wrong burrito to the wrong address. A waitlist-gated developer beta on one operating system in two countries is a controlled experiment, not a product launch, and the interesting questions are the ones a broader rollout would force DoorDash to answer.

What to watch is whether the other food delivery platforms feel pressured to publish their own agent-facing endpoints in the coming months. If they do, agentic commerce stops being a slide in a keynote and starts being a real competitive surface. If they do not, dd-cli will look, in retrospect, like a developer-culture flex.