DoorDash just built an ordering interface for AI agents, not humans, as the app layer starts splitting in two

DoorDash just built an ordering interface for AI agents, not humans, as the app layer starts splitting in two

DoorDash co-founder and CTO Andy Fang announced dd-cli this week, a limited-beta command-line tool that lets AI agents search stores, find deals, and check out on the platform without a human tapping through an app. 1 It is open to U.S. and Canadian macOS developers via a waitlist. 1

A command-line tool for ordering food sounds like a programming joke. TechCrunch draws the parallel to the classic XKCD "sudo make me a sandwich" comic, and DoorDash's announcement video leans into the absurdity: it shows an agent reading Slack, parsing JSON, running Python scripts, and calculating totals just to order three salads. 1 But dd-cli is not a joke. It is a structural move.

DoorDash is exposing its entire ordering platform to programmatic agents, positioning itself as infrastructure that other software can build on. 1 Developers could use dd-cli to construct their own tools for ordering food, finding local deals, or combining DoorDash capabilities with other services as building blocks. 1

Here is what makes this bigger than a product launch. The application layer is splitting in two. On one track, human-facing apps with buttons, menus, and checkout flows designed for thumbs. On the other, machine-facing interfaces where agents navigate commerce programmatically, never loading a webpage. DoorDash is now building for both tracks, and the agent track is where the momentum is.

The evidence is already visible in DoorDash's own product lineup. The company exposes its service to AI chatbots including OpenAI's ChatGPT and Claude. 1 It has experimented with iMessage as an ordering channel and operates its own AI chatbot called Ask DoorDash. 1 dd-cli extends that infrastructure to any developer building agent-powered software. Each step moves DoorDash from a destination app toward a commerce API that agents can invoke on their own.

The sign-up form for the beta includes a field asking developers what they would build if granted access. 1 That is the posture of a platform soliciting ecosystem partners, not a company running a side experiment.

This is not a DoorDash story. It is a preview of a question every product team will face within 18 months: does your product have an agent interface? In 2010, "do you have a mobile app?" went from novelty to baseline expectation within roughly two years. The agent-interface question is arriving on a similar curve, possibly a faster one, because the consumer-facing pieces already work. The dd-cli demo shows an AI agent chaining together Slack reads, JSON parsing, Python execution, error recovery, and total calculation to complete a real purchase. 1 If an agent can already place a food order end-to-end through a command-line tool, the gap between agent experiment and agent commerce is closing fast.

For product teams and investors, the signal is this: the interface fork is happening, and it is asymmetric. Human-facing apps will continue to exist, but the growth surface, the developer activity, and the leverage are shifting toward the machine-facing layer. The companies that expose their platforms to agents early will aggregate demand the way mobile-first companies captured users a decade ago. The ones that wait will build agent interfaces under competitive pressure, the way every company eventually shipped a mobile app after the App Store made it unavoidable.

The next time a consumer platform launches a command-line tool for agents, do not file it under developer novelty. It is a bet on where the next wave of transactions originates: not from a thumb on a glass screen, but from an agent executing a chain of commands it assembled on its own.