Tom Verrilli: preference

Transcript context

…And just to fully understand why this is happening and why this should happen, what I'm hearing, there's kind of many combinations. One is AI is enabling this, which is great. Perfect timing. What are the other motivations to do this? Is it just the product ends up being better? Is it fewer people? I certainly think the product ends up being better. One of the things that folks have been telling us for a long time is, yeah, that won't scale. Oh, leadership isn't going to be able to stay hands-on with what's going on. You're going to need to go and hire tons more layers. And what we found is that's actually not true. It requires a different muscle. You have to make an effort to make sure you genuinely understand ground truth. An example we use all the time is we'll be talking in a growth meeting and someone will say, "Oh yeah, but that was fraud." Turn around and be like, "How do you know that was fraud?" "Oh, it's labeled in the dataset as fraud. "Okay. Do you know how it gets labeled?" "I assume someone in ops does it." "Okay, do you know the SOP or how they label that?" "No." "Okay, so you don't know it's fraud." And if you push a really experienced product director like that, they're going to go, "Good point, I don't. I'm going to go find out." And then you invariably end up strengthening the system that agents are using to data label because suddenly there's a very smart person who's very invested in understanding how we do that and helping guide it. And so just that attitude that says we're going to do fewer things, we're going to make sure we execute the hell out of them, and we're going to push our best people to be in the weeds everywhere means that you fix lots of things as you go and you don't end up just making loads of trade-offs. And I think there's a general belief that it's too easy otherwise for growth to hide all sins. You get bigger and you just end up scaling and everybody's working off of averages. So culturally, I think you've got to be really committed to let's whole ass a few things, as I tend to say sometimes. Shout out Ron Swanson. And then push your best people to be really in the weeds of stuff. And actually what you find over time is you end up being more efficient by doing that because you actually understand how things work the first time and you make the best decisions. And then listen, AI, huge leverage for all of this. I can't think of how much time I spent as a junior PM asking my engineers how hard something would be, and distracting actual velocity in order to help scope future stuff. And now I can sit and talk to Claude and understand roughly LOEs. I can sit there and go through and be like, it feels like there's some car crash of modals that must be hitting new users as they open up the app. And you can literally just go through and be like, let's load up the feed and talk to me about the logic of who sees what in what order and when does this thing fire? And you can get answers really quickly. And having, as I said before, folks with enough tenure and enough reps that they can see that and say, "You know what? That's bad. Let's just make a good decision and change the ordering of those." y. And having, as I said before, folks with enough tenure and enough reps that they can see that and say, "You know what? That's bad. Let's just make a good decision and change the ordering of those." I've saved now a kickoff meeting, an alignment meeting, a week writing PRDs, experiment time, all of that by just having somebody who's kind of empowered to go and make a decision.…

Stored transcript either side of the excerpt. The highlighted words are the published quote; the surrounding text is unedited source, never generated.