DASHDoorDash, Inc.

“Well north of half of our code—probably closer to two-thirds of our code—is written by AI today. But that alone does not articulate how workflows and team setups ought to change. It means that we are being more productive and shipping more code, but the ultimate question I have is, are we actually delivering better outcomes for customers?”

“where I have found AI to be helpful—especially now with more powerful models that can reason in a multi-turn fashion—is that you can start looking at repetitive processes that are stitched together and actually get them done with perfect quality every single time, using just an agent. Even six to eight months ago, this was less true because you had to build a lot of backup or redundant systems to make sure agents do not go off the rails and can actually finish the task. That has happened with onboarding, for example—whether it is helping you with your menu or your catalog as a restaurant or retailer, or with your photos and your metadata and the annotation of that data.”

“not just building agentic ordering experiences on DoorDash, Inc. to make discovery or search easier, but also building a catalog, a digital catalog of structured information for the physical world: collecting where every banana sits or every ripe or unripe avocado, to every size shoe in whatever color and style a customer is looking for. All of that information about the physical world—of which there are billions of items, tens of millions per city—and getting that annotated and having that unique and proprietary to DoorDash, Inc., which we do not have to share with anybody.”

“we are using it across the board and seeing productivity improvements. The goal for us from a productivity improvement perspective is as it has always been: we want to do more with more. We want to drive more features. We want to do more for our audiences, and we want to do more internally as well. Ultimately, we channel productivity improvement into developing more features. If it is purely from a modeling perspective, I would expect, in the near term, OpEx to roughly be in the 2% range that I have talked about before.”

“We are already seeing benefits to the P&L from some of the AI work that we are doing—some of it on our own products, like the AI ordering agent, and some on tools related to merchants, customer support, and Dashers.”

“The second thing we have to do is build structure and cleanliness out of what is inherently very messy and constantly changing, which is a challenge. If we can do both across every category as we march from restaurants to grocery to different categories within retail—and do that through the merchant's channel online, the DoorDash, Inc. channel online, and the merchant's channel offline or in-store—I think that builds a really rich dataset that is nonexistent anywhere, extremely valuable for the merchant to have a full view of all the different types of customers and occasions, and really interesting for DoorDash, Inc. to build both products as well as businesses.”

“Take, for example, Google food ordering, which allowed you to order through various Google channels—Google Maps, Google Search, and I believe a few others—where you could order restaurant delivery. That started in the mid-2010s and went for about eight years before they shut it down. From a traffic perspective, they absolutely could drive a lot more traffic than virtually anyone else could to any one of these restaurants. Yet the retention of that traffic was a fraction of what platforms like DoorDash, Inc. saw, and as a result, customers effectively moved all of their shopping experiences to DoorDash, Inc.”

“Ads are just a means to connect consumers with merchants who are hoping to be discovered and making sure that you do that in the best possible way. With respect to agentic commerce, that is just one way of shopping. I do not think it will change our ability to advertise. It may increase some of the in-surface areas, but I think a lot of that remains to be seen. I do not think the ideal agentic shopping experience is just going to be a chat assistant. I think it is going to take on various forms, and we are iterating on that.”

“we are the only company that has the most robust catalog, much of which is actually about the physical world that does not exist in any digital repository, that cannot be scraped, and that we ourselves uniquely own access to because of all the work that we do to actually build up a repository of the physical world. That is something that we will continue to build greater and greater advantage in, especially in the world of agentic commerce.”

“we absolutely will have agentic ordering experiences in which it will be a lot easier for customers to do many things that they do today with much lower friction, to discover things that they perhaps did not know existed on DoorDash, Inc., to formulate complicated queries and solve those in the best possible way.”

“The top priority for us right now is getting all teams onto a single tech stack. The second priority is making sure that everyone in the company—not just engineers—is as AI-capable as anyone else. Then we can start thinking about what workflows have to change to truly deliver things faster.”

“customer support, which I think is also having an agentic revolution in itself. You will see all of these things play out in the DoorDash, Inc. product experience.”