Your Next Customer Might Not Have Eyes

Something odd happened this month. DoorDash let people order lunch from a command line. A payment code that sat dead since 1997 came back to life. And a founder raised money to build a button that fires off AI agents. Different stories, same trend line: the thing buying from you might not be a person clicking a screen. Let me catch you up.

Ordering a salad by typing code

DoorDash shipped a beta tool called dd-cli that lets you order food straight from an AI agent. You can search stores, find deals, and check out without opening the app. Co-founder Andy Fang put it on X, and yes, it looks like a joke at first. Ordering three salads by parsing JSON and running Python scripts is absurd on purpose.

But look past the gag. DoorDash is opening its ordering platform so developers build their own checkout flows, or bolt food ordering onto other tools. The app is no longer the front door. The company already feeds its service to ChatGPT and Claude. The screen you spent years polishing is now one option among many, and maybe not the main one.

The dead code that became a cash register

HTTP has a status code, 402 Payment Required, that sat unused for 29 years. Backend developers treated it as a joke, an IOU for internet money nobody built. Now AWS and Cloudflare have turned it into working payment rails for agents through a protocol called x402, laid out well in this walkthrough of the handshake.

Here is how it runs. Your server answers with a 402 and a machine-readable price, usually in USDC stablecoins. The agent signs a payment proof and retries. A facilitator verifies and settles, then your server hands over the goods. No cart. No checkout page. No human reading your pricing tiers. A companion piece notes you can wire this into an API in about 20 lines of middleware.

Why now? Agents break the old money models. Ads, subscriptions, and click-through flows all assume eyeballs. A bot does not watch ads or read your upsell copy. It just wants the resource and will pay for it if you let it.

A button instead of a keyboard

The hardware side is chasing the same shift. Aina, run by Ultrahuman's former hardware VP Apoorv Shankar, raised $5.5 million for devices that control agents rather than record you. Its first product, Dune, is a three-key pad that runs shortcuts and triggers workflows.

Shankar drew a clear line. He is done with always-listening rings and notetakers that just capture what happens around you. "We are building an action-oriented device that will use the context to help you control and trigger workflows," he said. OpenAI just shipped a keypad for Codex too. Qualcomm says it is testing more than 40 devices. Nobody knows the winning form factor yet, but the bet is consistent: people want to fire off agents, not narrate their day.

The screen stops being the store

String these together and the pattern is plain. The transaction is moving off the screen. A command line places the order. A revived status code moves the money. A physical button kicks off the task. In each case, your carefully designed interface is not where the decision gets made.

Roblox shows the flip side of this in creation, not commerce. Its new Build tool lets anyone make a game from a text prompt, no code needed. To handle the flood of AI-made junk, Roblox is leaning on ranking: "If no one plays it, no one can find it." That is a preview of your problem. When creation and buying get cheap and automated, discovery and trust become the scarce thing, not the interface itself.

The deep cut

If agents can buy from you, your pricing and terms have to be readable by a machine, not just persuasive to a person. The x402 flow assumes a price the agent can parse and act on in one step. That means your value has to survive with zero UX around it: no hero image, no comparison table, no free-trial nudge. Go pull one product decision this quarter and ask a blunt question. If an agent hit our API or checkout with no human watching, would it know what to buy, what it costs, and why to pick us over the next endpoint? If the answer lives entirely in your visual design, you have a gap to close before the traffic shows up.

Three questions for your team

• If an agent could reach our product tomorrow, what is the smallest path to purchase, and does it work without a single screen?
• Is our pricing machine-readable and stable enough that a bot could parse it and settle payment without a human in the loop?
• When creation and buying get automated, what makes us the endpoint an agent picks: trust, data quality, retention, or just habit we no longer own?