Harley Finkelstein45:38
But the point is in it he was basically like, 'Yo, AI is a baseline expectation. You got to learn these tools. The way to get good at them is to use them a bunch.' And you got to start solving problems using this. That was the general ethos. I'm curious, you personally Harley, day-to-day, what's your AI usage? Are you like a lot, a little, just ChatGPT, other tools? Like what are you actually personally using either professionally or personally?
Two things. One is I use ChatGPT. I really like the fact that now it has the history side of it. So for example, back to longevity, every time I get a blood test I upload it to ChatGPT in a folder called longevity and health. And so now I can ask questions. Although I do find if I say I had a blood test last week, I uploaded my new one, I said compare this to my last one and it compared it to one from a long time ago and I was able to spot it. So it's not 100% but I like using it quite a bit. I gave a keynote at Summit last week. I literally fed my entire keynote script into ChatGPT and I said, 'Hey, what parts of these do you think are confusing if any? Like where am I losing the thread here?' So I think that I use ChatGPT and particularly the 03 model on a daily basis. I have two screens here, so based on my screen next to me I always have that open. I really like Claude for other projects. So for investing, for example, my wife Lindsay and I do a little bit of investing. We own some apartment buildings. So I have a file for real estate which is every time I get an update from the property management company I feed it in. I find Claude is much better for analyzing existing documents. I find that ChatGPT is much better for reaching out into the public sphere like the web and bringing things in together. Those are the two I use the most. And then back to the earnings calls, one of the things we did was we basically made all of Shopify legible and we built about a dozen MCPs inside of Shopify. So usually in preparation for an earnings call I will go to, let's say I want to talk about Shopify capital. I'll go to the capital leader and say, 'Hey, give me a couple of the highlights for UK. It's a new one for us. I want to talk about how it's going.' This time I just went to the vault and literally went into the capital program because everything is legible. I was able to pull most of the data on my own. The personal jury's still out whether or not I miss because when I talk to the leader of capital, I get some of the nuances and I'm like, 'Oh, you just said this sort of thing and I'm going to take that and run with that even though you didn't intentionally mean to leak it to me.' Whereas using the Shopify vault, I find that I'm basically getting a lot of headline stuff. But I do want to make one point. One of the things that got missed on Tobi's memo or the email now is that there was a term he used that I'm surprised more people haven't picked up on, which is AI being reflexive. And part of what he's trying to get the company to do is to be reflexive about the usage of AI so that rather than sort of make it a project and then use AI, this idea of making it reflexively part of our day-to-day work is really what he's trying to do. And in fact, so much so that he created this new friction, which is before you can go and hire someone on your team, you first have to substantiate to Shopify to the hiring manager to the talent team why AI cannot do it better. And one of the things we're noticing there is by actually creating this friction of saying before you go and hire, you must first do it, what ends up happening is people by the end of substantiating why they need to hire someone that AI can't do, realize that they actually should have AI can do a bunch of that as well. That is very different than what you're hearing in most companies, which is let's just use AI as part of our, if I have to design something in Figma or in Canva, I'm going to use AI to help with that. This idea of it being reflexive is what we're trying to do.