Lenny's Podcast: Product | Career | Growth
01:01:39 · Lenny Rachitsky
[01:01:39] S1: There's a concept that I heard come up a bunch when I talked to people who work with you. The phrase was play the accordion. Explain.
01:01:48 · Tom Verrilli
[01:01:48] S0: So I mentioned earlier, I'm not huge on frameworks, but I do think there's a mental model which is quite useful that goes to the thing that you and I discussed earlier on like think through a problem. Do the mental work even if you're not shipping at all, which is like obviously the magic of code is that you can kind of ship things and iterate really quickly and get data as you go. And I think one of the trappings that comes with that is you're just iterating forward and throwing spaghetti at a wall and you don't really know what direction you're going in. I think there's a second failure mode, which is people sit down and write out these long roadmaps and strategic vision docs of what we'll do over the next two or three years that kind of loses the comparative advantage we have over every other industry, which is learning. AB tests mean that you can update your understanding of a problem and therefore change your direction as you should go. So the point of the analogy of play the accordion is if you think about a piano accordion, before you can play a note, you've got to stretch it all the way out, bring the air in. So what is it we're trying to get done here? But you don't make music until you press the key and push it all the way back into V1. And then when you go to play the next kind of progression, you pull it all the way out again. Okay, given what we just learned, what do we do? And then you go and build the next thing again. And you've got to get used to this motion that says this isn't creating value. Pulling it out doesn't actually do anything. All the value is created here. But until you are going through this motion constantly of saying, well, reevaluate what we understand. How does this change what we're doing? You're probably not playing the right thing. And I know I'm probably bastardizing. Somewhere in the comments, there's going to be a musician who points out to me that this is actually also one of the ways you make music. But I found it just a generally very valuable memory device for PMs to be like, you've got to constantly be zooming out and pushing back in and zooming out and pushing back in.
01:03:38 · Lenny Rachitsky, Tom Verrilli
[01:03:38] S1: Yeah, that is such a fun analogy because even though you make music expanding it, it's like inward or it's like learning from the compression. [01:03:49] S0: What do we learn? How do you communicate to everyone [inaudible 01:03:51] back to shipping? [01:03:52] S1: It's a different kind of music. Internal music, external experiment. [01:03:55] S0: Yeah. Because genuinely there are people who are really big on roadmaps and there are people who are really big on iteration. I think the honest answer is you've got to do both. You can't over-index on either. [01:04:05] S1: So the lesson here is run experiments, but then make sure you think about the implication of the result at the bigger picture. And then with that [inaudible 01:04:15]. [01:04:14] S0: What are we trying to do? Have a plan. I think the system works this way. I have this belief that if we change the way that discovery algorithms work for XYZ reasons, it will have this impact. What's the smallest thing I can do to prove that? Okay, it worked. Great. No change to plan. V2. Oh shit, that didn't work. V3 is going to have to be different. And you just want to always be going through that motion. [01:04:36] S1: People not watching on YouTube, Tom is moving his hands. [01:04:39] S0: Gesticulating wildly. [01:04:42] S1: What's a context where you have said to someone, "Hey, go play the accordion," or, "We're playing the accordion." What are they usually doing wrong?
01:04:50 · Tom Verrilli
[01:04:50] S0: It tends to be that you're shipping something in a very local sense without necessarily understanding impact. So an example right now is the core of most marketplaces is listings. If you try and think about Amazon.com, without listings, there's basically nothing there. It's a left rail or right rail and some videos. In video commerce and live commerce at Whatnot, you don't really historically need listings. If I want to sell you a pair of AirPods, I could literally hold them up to screen and show you them and describe them and say they're AirPods. I'm going to start them at a dollar. And as a buyer, you now have all the information you need in order to make a purchase decision, which is great. And so you may not have to invest in making listings the way that someone else does. And that's probably net good for a seller because it takes three and a half minutes to make a listing and it takes zero minutes to describe a thing and hold it up. But you then zoom out, and say, "Oh shit, new buyers are going to come to live commerce and expect search to function. If I don't know what you're selling until after you've sold it, there's no possible way that I can put somebody in the right stream for AirPods because we didn't know you have them and I therefore can't direct people to it." And so you go through this exercise of like, "We don't need to fix it. It works." And then you're like, "Okay, great. What does that have implications for other things?" Okay, then what would we do? Oh, we're going to make everyone make listings. And you're like, "Zoom back out again. If every seller is now spending three minutes for every listing that they're making, the number of things they can sell per hour is now dramatically, dramatically lower, which means it's bad for sellers. Okay, shit, we can't do that again." And so you have to keep stretching through the what are the longer term implications or what are the knock-on effects of things we're building?
01:06:36 · Lenny Rachitsky
[01:06:36] S1: That is extremely helpful. What's interesting about Whatnot is there's this spectrum of Whatnot, and then there's this whole trend of agentic commerce where agents are going to be buying the things for you and just working with each other, create a whole new agentic economy. And this is the opposite. Humans live talking to each other, buying from each other. Thoughts on that trend and how that might impact you guys?
01:07:00 · Tom Verrilli
[01:07:00] S0: Listen, I, for one, welcome our agentic overlords for things like I don't want to have to think about light bulbs, air filters, any of the programmatic stuff that I need to run my life. I'm also pretty great with it for really high intent things. I need a particular cable for my computer. I'm looking for outdoor lights for my house. Things where there are really taxing searches. I think about this, you got to go to a wedding, which means I need black shoes, but delivered by Thursday because if they don't get here before Thursday, they're no good for me kind of things. Absolutely. But most of commerce in America isn't actually high intent. How many years are we into e-comm now? 30 years into e-comm. And e-commerce has never exceeded 20% of retail spend in America. The vast, vast, vast majority of retail shopping in the United States is still people going in person and buying things. In the UK, it's similar. It's like 75-25. And it turns out actually that a lot of shopping is low intent. I'm going to the mall because I've got a wedding coming up, and I don't have anything to wear and I'm going to wander around and see what people have available because I don't actually know specifically what I want. And actually the value of stores is that that person who runs that shoe store has agency and taste, and she curates a selection of shoes. She has a level of customer service. She has a window display, which can show me the types of things she has. And I can wander around the mall and actually work out what it is I want to buy or be educated by it. And as it turns out, it's quite pleasant. There's a reason that the mall is a social thing. And so I think that agentic is going to be huge. Retail is a $7.5 trillion industry in the US though, so I don't think it's a winner takes all thing. What I think Live Commerce does is it's the first time that we've ever managed to bring together scale and convenience of the internet and also that same social cultural experience of physical shopping. Because at Twitch, we used to basically assume that if it's less than 1,000 people in a stream, it's non-economic because you're operating on CPMs. But you can go into a stream on Whatnot and see 30, 50 people in that stream. And you imagine for a moment that you're running a shoe store at a mall and you had 50 people in your store. You'd never close.