DoorDash has cut the wait for its AI-powered grocery reorder flow from about 16 seconds to 2 seconds by preparing recurring shopping needs in advance. The company says the change doubled shopping list views and increased the flow’s order rate by approximately 34%. The feature runs through Ask DoorDash, its conversational shopping assistant.
Instead of asking an AI model to reconstruct a customer’s likely needs each time they open a reorder request, DoorDash analyzes 90 days of purchase history along with profile signals ahead of time. A large language model (LLM) – the kind of AI behind chatbots – suggests items that may be recurring purchases. A separate set of fixed checks validates the suggestions before the reusable list data is saved.
What happens when a customer opens a list?
The live request retrieves the prepared list and checks each item against the current product catalog. The system first tries the original store item, then a catalog product ID, and finally a search term. That sequence helps it handle products that have changed or become unavailable without repeating the historical analysis for every request.
Not every action requires another AI response. Customers can change quantities or remove products directly, and the updated list becomes the starting point for the next agent turn. DoorDash says the Grocery agent is still used when a request needs more judgment, such as finding cheaper options based on current prices, promotions, package sizes, stock and the customer’s preferences.
A shared platform for other agents
The grocery system uses DoorDash’s shared agent platform, which provides tools for coordinating agents, memory, model access, tracing, evaluation and controlled rollouts. Individual teams still manage their agents’ instructions, tools and evaluation criteria. DoorDash engineer Twisha Jain described the approach as deciding which work belongs in the agent’s live loop and which can happen earlier.
DoorDash says it took one week to add its Reservations agent to the shared platform, about ten times faster than building its first domain agents. That points to a second goal beyond speeding up grocery reorders: making it easier for teams to build agents on common infrastructure. The company’s design also treats the shopping list as an editable object, not just a record of chat messages, so a customer’s direct changes take precedence.
Why it matters
For DoorDash customers, the reported change means a grocery reorder list can load more quickly, while customers can still edit it directly and get current catalog matches. DoorDash reports higher list views and order rates, but the figures do not show how the flow performs for every customer or how much of the increase came from the shorter wait.
