DoorDash Tests iMessage AI Agent for Food Orders on iOS

Choosing dinner has entered its next, apparently inevitable, stage: delegating the indecision to a robot over text.

DoorDash is testing an AI agent through iMessage that is designed to turn a conversational food request into restaurant suggestions and, eventually, an order. Rather than opening the delivery app and browsing the familiar tiles, a user can send a message describing what they want. The agent is intended to consider both the request and the user’s prior DoorDash orders before proposing nearby options.

It is a small change in interface, but an interesting one. Food-delivery apps have spent years condensing the ordering process into search fields, recommendation rails, saved addresses and prominent reorder buttons. DoorDash’s experiment asks whether the next layer of convenience is not a faster menu, but avoiding the menu until a machine has narrowed it down.

For now, the trial is limited. People can sign up through DoorDash’s website to test it on iOS, while Android support is planned for a later date. There is no confirmed timing for the end of the beta or for a broader release.

What the DoorDash AI agent is meant to do

The basic flow is conversational. A customer sends the agent an idea of what they would like for dinner. DoorDash’s system then searches restaurants in that person’s area and makes suggestions informed by the message and previous orders.

That distinction matters. This is not presented simply as a text-message shortcut to a specific restaurant. It is designed to interpret an open-ended prompt—something closer to “what should I have?” than a conventional search query—and use order history as part of the response.

DoorDash has used the example of requesting a usual Friday-night meal. In that scenario, the agent can remember the familiar order and the modifications normally attached to it. A modification is the restaurant-ordering term for a requested adjustment to an item, such as leaving something off or otherwise changing the standard preparation. These details are exactly where a repeat purchase can become inconvenient if a system fails to remember them.

An AI agent, in this context, means software that receives a natural-language request and helps carry out a task through a series of steps. The important practical distinction is that users are not expected to browse every restaurant or formulate the exact structured search themselves. They state an intention in ordinary language; the software attempts to translate it into choices within DoorDash’s service.

That is the pitch, at least. The test does not establish that text messaging will be the best way to order food for every person or every meal.

The real question: is chat actually faster?

DoorDash’s standard app experience already makes repeat ordering relatively direct by surfacing recent purchases near the top. That means the new agent is not solving a completely unsolved usability problem. For someone who wants the exact same restaurant, items and preferences as last time, opening the app and using a reorder-oriented option may remain the more obvious route.

Text could be more useful when the user has a preference but not a decision. “Something spicy,” “pizza for tonight,” or a request framed around a routine are more ambiguous than tapping a past order. The proposed benefit is therefore less about shaving a second off a known purchase and more about outsourcing part of the choice.

There is also a possible irony to the chat-first approach. A customer may need to write a request, read recommendations, choose among them and confirm details. Depending on the conversation, that could involve more interaction than selecting a previous order in the app. A conversational interface only feels frictionless if it reaches the right suggestion quickly and retains the details the customer actually cares about.

In other words, the agent’s value is likely to depend on the gap between “I know exactly what I want” and “I would prefer not to think through dinner.” The former has already been addressed reasonably well by reorder tools. The latter is where DoorDash is placing its bet.

A return to text ordering, with more interpretation attached

There is a retro twist to this experiment. Ordering by message predates the smartphone-app era, even if it was far less flexible. Domino’s introduced an emoji-based ordering method in 2015, allowing customers to use a single symbol to place an order. That approach was tied to one restaurant company and one very tightly defined action.

DoorDash’s proposed system has a broader job. It must navigate a collection of local restaurants, account for a user’s past orders and respond to a written request that may be casual, incomplete or undecided. The medium—sending a message—looks old-fashioned. The intended interpretation of that message is the newer part.

That does not make it automatically better. A one-symbol pizza order works because its meaning is deliberately constrained. A dinner request in plain language can be richer, but it can also be unclear. “My usual” may be simple for a returning customer with a stable routine and much less simple for someone whose prior orders vary widely. The beta will be the place where the usefulness of that memory-driven approach becomes clearer.

Why the iOS-only beta matters

The test currently runs through iMessage and is available only to iOS users who sign up through DoorDash’s site. Android users are expected to receive support in the future, but no timetable has been provided. That limitation is significant because the feature’s defining idea is not merely DoorDash using AI; it is DoorDash placing the interaction in a messaging channel.

For iOS participants, the appeal may be that dinner planning can begin in a familiar text thread rather than a dedicated shopping-and-delivery interface. For Android users, the ordinary DoorDash app remains the route for now. Until support expands, the experiment is also a test of whether iMessage is the right home for this type of assistant.

It is worth separating the beta from a full launch. A beta is a testing stage in which access can be limited while a feature is evaluated before broader availability. DoorDash has not said when this test will exit beta. Anyone interested should treat the iOS sign-up as an opportunity to test a developing feature, not a promise of universal access or a final version of the experience.

What gamers may recognize about the design

The underlying interface debate will feel familiar to anyone who has watched games add increasingly aggressive shortcuts for getting back to the action. A clean menu is useful when a player knows the exact destination. Recommendations, recents and presets become useful when the choice is routine. A chat-based helper is a different proposition: it asks the user to describe a goal and trusts the system to navigate the menus behind the scenes.

That can be appealing when the task feels repetitive. It can be less appealing when people want to inspect choices themselves. Food orders are personal in ways that make this especially noticeable: a suggestion is only useful if it matches both the immediate craving and the repeat-order details the person expects to be retained.

The larger games business is also continually changing its structure and tools, as seen in this recent Ubisoft organizational shift. DoorDash’s test is not a games announcement, but it belongs to the same wider conversation about how digital services increasingly try to reduce menus, searches and routine clicks into a single prompt.

Convenience is not the same as eliminating choice

The most persuasive case for DoorDash’s agent is not that tapping through an app is impossibly difficult. It plainly is not. The case is that deciding among options can be tiring, particularly for frequent customers who have established patterns but do not want to reconstruct them every time.

Its weakest case is the customer who already has a precise favorite. If the app puts a recent order in easy reach, an iMessage conversation may be an extra layer rather than an improvement. DoorDash appears aware of that tension: the same past-order data that enables a conversational “usual” is also what makes conventional reordering quick.

The feature will ultimately have to earn its place through relevance. It needs to be useful enough that people choose to text the agent even though they could open the app, search directly, scan recent orders or browse on their own. If it can accurately bridge the moment between “I want food” and “this is the order,” it may become a welcome shortcut. If it turns a quick repeat purchase into a dialogue box, it may be remembered as a very modern solution to a problem users had already mostly solved.