The Agent Moved Into the Shopping Cart

An AI agent can now remember your usual lunch, narrow a marketplace, build the cart, ask for approval and place the order. The shopping trip is becoming a conversation.

The change is easy to miss because it does not look dramatic. There is no robot in the aisle. A person describes what they want, the agent removes most of the available choices, and one manageable answer appears. For routine purchases, the agent may also prepare the transaction and complete it after approval.

That puts the agent in the part of shopping that matters most: between wanting something and deciding what deserves attention.

Lunch becomes a conversation

A New York Times episode of The Daily, also available as an official public video, made the shift feel mundane.

Host Natalie Kitroeff asked Meta’s Muse agent to find a salad-like lunch with protein and deliver it to the office. Reporter Eli Tan entered the request, supplied the address and authorized a card on file. Muse proposed a Sweetgreen harvest bowl and asked Tan to approve it. Kitroeff called it her favorite order. The agent placed the order while the reporters watched it operate a browser, and the delivered bowl contained protein.

Nothing spectacular happened. That is what made the demonstration useful. A loose preference became one merchant, one item and a prepared transaction. The agent handled the work that usually produces a collection of tabs, filters and nearly identical menu listings.

DoorDash puts the agent inside the marketplace

DoorDash is building that behavior into its own service. Its September 30 announcement, “Skip The App: Now You Can Just Text DoorDash,” describes a U.S. beta that lets customers text DoorDash the way they might text someone already standing in a store.

A customer can describe a craving, request a usual order or ask for something that works for a group. DoorDash says the system can remember modifiers, dietary preferences and routines, suggest a cart and manage checkout inside the text thread. Its AI assistant help page describes a limited pilot with controls for updating or forgetting preferences and opting out of proactive prompts.

DoorDash estimates that an average U.S. user has access to more than 800,000 restaurant menu items and grocery products through the service. The agent’s value is not displaying the whole shelf. It is deciding what the customer will not have to inspect.

The company says Ask grocery baskets measured from August through early September averaged nearly 50 percent higher value and about 60 percent more unique items than the same customers’ traditional orders. Those are company-reported associations, not evidence that the assistant caused customers to spend more or buy more items.

A DoorDash spokesperson told the show on background that Ask DoorDash currently contains no ads or sponsored listings. The spokesperson said recommendations can use stated preferences, order history, budget, diet, household size, allergies, brand and price preferences, mood and real-time availability. Customers can change preferences conversationally, ask the assistant to forget specific information and opt out of proactive prompts.

That statement is paraphrased in accordance with the source’s terms. The correspondence is not linked or reproduced.

The agent can narrow a purchase without completing it

The same pattern applies to purchases that require more deliberation.

In a WIRED field test, reporter Reece Rogers and his partner asked one of OpenAI’s Dots agents to help replace a couch. After correcting a name error, they supplied doorframe dimensions and a budget. The agent returned four options with prices, measurements, product links, return policies and photos. They later added style and color preferences and requested ten choices scored against a rubric.

They did not buy a couch. The useful result was the narrowing. The agent converted a household discussion, physical constraints and taste into a shortlist without requiring the couple to repeat the same search across retailer sites.

The storefront remained. More of the decision happened before the shoppers reached it.

Who pays the personal shopper?

The companies entering agentic commerce are not using one business model.

CNBC reported that Meta chief executive Mark Zuckerberg said Meta would take a small fee on Muse transactions. DoorDash says its current assistant has no ads or sponsored listings.

OpenAI’s current public description of product discovery in ChatGPT says merchants retain their own checkout experiences while OpenAI concentrates on discovery. An OpenAI spokesperson told the show on attributable background that current ChatGPT shopping sends buyers to merchant sites, OpenAI takes no cut of those sales and commercial partnerships do not influence product recommendations.

That statement is paraphrased in accordance with the source’s terms. The correspondence is not linked or reproduced.

OpenAI has tested another model. Marketplace Pulse reported in January that Shopify merchants using the earlier Instant Checkout path would pay OpenAI a 4 percent transaction fee. Reuters reported that OpenAI ended Instant Checkout in March.

That history is not a contradiction. It shows a young category testing where the agent ends, where the store begins and who pays for the handoff.

Convenience needs a receipt

Shopping agents will become familiar because they remove an unpleasant part of shopping: opening tabs, turning vague preferences into search terms and comparing listings that differ mostly in presentation.

The risk is not that every agent is quietly rigged. It is that convenience can make the agent’s choices feel neutral even when every shopping service has boundaries, business relationships and its own definition of a good result.

A useful shopping agent should explain what it searched, why it selected the options it showed and whether money or partnerships shaped the result. That is not a demand for suspicion by default. It is a basic receipt for delegated judgment.

A personal shopper can work for the buyer and still work for a business. The buyer should be able to see both roles.

Listen to Episode 69

Episode 69, “The Agent Moved Into the Shopping Cart”, is available now.

Download the episode or subscribe to the feed.

Sources

If you have asked an agent to research, build a cart or complete a purchase, tell the show what you delegated and what you still wanted to control. Did it save time? Did it understand your preferences? Email [email protected]. Use the subject line “The agent in my shopping cart.” Anonymous notes and source-protection requests are welcome.