The Frontier with Brendan Foody
00:06:41 · Brendan Foody, Andy Fang
[00:06:41] S0: Roughly 15 years later, almost that now, you guys have transformed DoorDash into a really category-defining company. And I'm curious how that mission continues to play a role in the AI economy that's emerging and how it's transforming commerce and the whole way the economy works and what role you see DoorDash playing in that. [00:07:04] S1: I mean, to be honest with you, I think we're all just learning and exploring here. And I think one thing that we're very keen on is learning more about agentic commerce and exploring like what that can look like and how like commerce isn't getting transformed. And I think we've done a ton of explorations in the space. I mean, when I first visited you, I think we were very keen on understanding how do we get these agents to like do what we want it to do. And then you were teaching me about evals and stuff and like kind of the nuances there. And I think for us, that's kind of how we landed on you know iterating in the space. And we launched Ask DoorDash a couple weeks ago. And that's been a very, very promising product for us. I guess the first vector there is seeing how we can introduce this type of agentique experience into the core product. And we're very fortunate that this product that we're rolling out has seen a ton of um early success.
00:07:04 · Andy Fang, Brendan Foody
[00:07:04] S1: I mean, to be honest with you, I think we're all just learning and exploring here. And I think one thing that we're very keen on is learning more about agentic commerce and exploring like what that can look like and how like commerce isn't getting transformed. And I think we've done a ton of explorations in the space. I mean, when I first visited you, I think we were very keen on understanding how do we get these agents to like do what we want it to do. And then you were teaching me about evals and stuff and like kind of the nuances there. And I think for us, that's kind of how we landed on you know iterating in the space. And we launched Ask DoorDash a couple weeks ago. And that's been a very, very promising product for us. I guess the first vector there is seeing how we can introduce this type of agentique experience into the core product. And we're very fortunate that this product that we're rolling out has seen a ton of um early success. [00:07:54] S0: I remember I would often like be looking for something on DoorDash and spend five to ten minutes looking around. And I've started using Ask DoorDash. And so I'm really nice, excited about all the progress there. Yeah. How do you see that evolving? Like if everything goes well with Ask DoorDash and the variations on that product direction over the next five years, what will the future look like for DoorDash customers and more broadly?
00:08:17 · Andy Fang
[00:08:17] S1: I mean, five years, I have no idea what things are gonna look like. But even just like thinking like six to 12 months out for a sec. I mean, I think what we're seeing in the early rollout is like, you know, we're seeing our grocery basket sizes are like, I think 40% larger, which is like a number that's been really hard for us to move historically. And then on the restaurant side, we're seeing like 50% of conversations of people in Ask DoorDash there are ordering from restaurants they've never ordered from before. And so these what we call new merchant trials at DoorDash. And that's also been a very hard metric for us to move. Um, and so for us to see that type of impact in this early rollout and see it sustain as we've been increasing the rollout size, that's very exciting. But I think one thing that we've been so caught up on the past couple quarters is like, how do we get the intelligence to be good enough? Once we felt like the intelligence is good enough, now the question is how do we actually make it easier for people to get introduced to the experience and know what to do inside the experience? Because I think a lot of times, like if people just see like a chat box uh and they're like, okay, like what do I enter or what do I type into this? It's can be a little intimidating. But what we're seeing is like people who engage with it, we're seeing really promising results. So we're trying to figure out ways to make that a little less intimidating.
00:09:26 · Brendan Foody, Andy Fang
[00:09:26] S0: When you talk about the intelligence being good enough, are you guys using closed source models under the hood or open source? And how are you seeing that evolving? [00:09:33] S1: Primarily we're using closed source, but we are using some like open weight models for like some like tool calls. But I would say like the primary intelligence of the agent is through closed weight models right now. [00:09:46] S0: Especially for like search over a giant catalog is actually a really surprisingly challenging problem. That exactly I agree that mainly the distribution is strong in the uh closed source models, but not yet the open source ones. One thing I've been thinking about a lot is that right now the these agents are mainly reactive. If we go to and we say we want a high protein meal, and then we're able to get that. And over time they'll start to become more proactive, right? Where it might identify what are the things that you might want to buy in the future, paying you around the time that you usually order dinner, all that kind of stuff. Yes. How are you thinking about that future developing? Definitely.
00:09:46 · Brendan Foody, Andy Fang
[00:09:46] S0: Especially for like search over a giant catalog is actually a really surprisingly challenging problem. That exactly I agree that mainly the distribution is strong in the uh closed source models, but not yet the open source ones. One thing I've been thinking about a lot is that right now the these agents are mainly reactive. If we go to and we say we want a high protein meal, and then we're able to get that. And over time they'll start to become more proactive, right? Where it might identify what are the things that you might want to buy in the future, paying you around the time that you usually order dinner, all that kind of stuff. Yes. How are you thinking about that future developing? Definitely. [00:10:24] S1: I mean, we're actually working on some uh concepts like that internally right now. But I think like if I were thinking a little bit further out, I think that's kind of where things go. It's like where you have these agents in the background actually doing stuff for you. And I think for us, what we're really excited about is not only are there these agents that kind of do stuff in the cloud, but DoorDash actually has a huge operational network that can actually do stuff in the real world. And so I think there's some interesting ideation that we've had internally about like, hey, what could be possible if these agents can perform things in the background? And so, you know, early innings there, but we'll see where things go. But I think it's gonna be a really interesting time to be disrupting consumer experiences, I'd say, over the next 12 months.
00:10:24 · Andy Fang, Brendan Foody
[00:10:24] S1: I mean, we're actually working on some uh concepts like that internally right now. But I think like if I were thinking a little bit further out, I think that's kind of where things go. It's like where you have these agents in the background actually doing stuff for you. And I think for us, what we're really excited about is not only are there these agents that kind of do stuff in the cloud, but DoorDash actually has a huge operational network that can actually do stuff in the real world. And so I think there's some interesting ideation that we've had internally about like, hey, what could be possible if these agents can perform things in the background? And so, you know, early innings there, but we'll see where things go. But I think it's gonna be a really interesting time to be disrupting consumer experiences, I'd say, over the next 12 months. [00:11:04] S0: I order my dinner at 6 p.m. on DoorDash every day. And so I feel like if it just said via text with here's the new meal or restaurant that I I should try that I haven't yet gone to, I uh it feels like there's so much impact to the user experience that could be possible there. What about allowing people to interact with it via other surfaces like Chat GPT? How are you thinking about that? I mean, I know this is a probably a controversial topic of how much to sort of use the own the end user interaction.