DoorDash moves checkout into Apple Messages for 20,000 pilot users

Andy Fang, DoorDash's co-founder, made the pitch in two sentences. "Every other AI agent is meeting you for the first time. We've known you for years," he said ^[15]. The Messages agent runs on the same technology as Ask DoorDash, the in-app chatbot launched in June 2026 ^[3]. Its claimed advantage is the customer file behind it. At the unveiling, during DoorDash's Dash Forward event at its San Francisco headquarters ^[3], technical lead Ayush Shah asked for his usual breakfast burrito. The agent came back with the order and his modifications already loaded ^[4].

The board-deck version says agents grow baskets. By DoorDash's figures, grocery orders built with Ask DoorDash had nearly 50% higher basket value than the same consumers' usual orders from August through early September 2026 ^[18]. A Citizens analyst note summarising comments from chief marketing officer Tim Castree said people who clicked through to DoorDash from AI chatbots converted at higher rates and filled bigger baskets than standard traffic ^[7].

That same note shows why the board-deck version is incomplete. It said orders from early agent traffic came with baskets two-thirds smaller ^[8]. That is a basket a third the size of whatever it was compared with ^[25]. DoorDash has released no comparable figures for the Messages pilot, and the published summary of the note does not say which agents, what period or what base the drop was measured against ^[19][9]. On this record, the two-thirds figure cannot be read as a forecast for the Messages channel.

The only testing of the Messages product itself found accuracy problems. Bloomberg News tried an early version and found prices that differed from those in the app, and photos that did not match what the agent recommended ^[16]. A DoorDash spokesperson said people joining the waitlist will receive a newer version ^[17]. A wrong price matters more in the thread than in the app. The agent quotes the price and completes checkout with stored payment details in the same conversation ^[13].

I think the real trade-off is between reorders and discovery. Castree said about half of customers arrive without knowing what they want ^[20]. From June through August 2026, nearly half of the in-app chatbot's restaurant orders went to places the customer had never tried, according to DoorDash's own usage figures ^[21]. Castree has also said the company does not want to outsource discovery to AI agents ^[14]. Most of the Messages features lean on what the customer already orders: the usual burrito and, with permission, proactive messages about their usual coffee ^[4][5]. The fridge photo is the feature that tests discovery inside the thread. A customer sends a picture, and the agent can suggest recipes and fill a cart with the missing ingredients ^[6].

Ownership is the slower question. Bloomberg Intelligence said tools such as Instinct or Meta's Muse could shift "traffic and customer ownership away from established platforms" ^[11]. Uber and Instacart have built plug-ins for assistants such as ChatGPT or Gemini, with varying levels of integration ^[10]. DoorDash has put its own agent, with its own customer file, inside Apple's app ^[1][13]. A separate DoorDash-hosted connector built on the Model Context Protocol lets workplace agents find items, build carts, place orders and track deliveries ^[22]. SpaceXAI, Cognition and Mercor have used it internally ^[23].

The decision this quarter is how far to open the waitlist, which went live on September 30 ^[2]. The consequence next quarter is whether DoorDash can put a basket figure for Messages beside its in-app one, or has to defend the channel on convenience alone. Castree said the outcome of AI agents is undetermined ^[24].