Uncapped with Jack Altman
· Tobi Lutke, Jack Altman
[00:00:00] S1: It is happening with AI. All the software we built was a bootstrapper for AI. Software becomes, due to this achievement, something that can become personal again. Now you can have basically a web browser, but the websites don’t even need to exist for what you would like to see. [00:00:00] S0: It’s so true when you say it like that. The easiest example to identify with for most of us is probably an iPhone. We’ve probably all just, randomly one day, looked at our iPhone and been like, “How is this possible? How does this thing exist?” It’s crazy. [00:00:00] S1: As beautiful as it is, and as reliable as it is, and as personal as it is. The iPhone is a lucky exception in that you can appreciate it as a physical thing. Everyone is right when they say, “Why is everything just standing still or not getting better around me?” Because from our ability to observe it, all the infrastructure is digital. But we are almost done with this. We are at the end of the opening chapters. Now I think we will see vastly more impact in the real world. [00:00:00] S0: We put all of our energy against this thing because it was the foundation, but now all these people will, in some sense, be free to not write software. [00:00:00] S1: That’s exactly right. So I think we will have a huge influx of the brightest and most creative and driven people to make things which are going to be much easier to relate to and have much more direct impact on people’s lives.
· Tobi Lutke, Jack Altman
[00:00:00] S1: That’s exactly right. So I think we will have a huge influx of the brightest and most creative and driven people to make things which are going to be much easier to relate to and have much more direct impact on people’s lives. [00:00:00] S0: How important is it for you, as CEO of Shopify, to have an opinion on where AI will be in two years? There’s one worldview you could take, which is, “This is so unpredictable. I’m going to take it as it comes. I’m going to try to know six months ahead, but I’m going to do what I can do.” And then there would be another which is, “Now I’m actually going to spend a lot of time with the labs, and I’m going to try to have an understanding of two, three years out and try to care about the farthest out I can see.” What do you think is the right position? [00:00:00] S1: Yeah, if my friends would listen to this question, they would laugh, just because it’s my predominant obsession. I try to have as many data points on as many people, which I can then try to match to the right superlinear or sublinear curves, and just figure out how they all connect and what therefore will happen in the future. To me it’s the most fun game in the world: to have a pretty clear-eyed view of what the future is like. [00:00:00] S0: What in particular are you trying to figure out?
· Tobi Lutke
[00:00:00] S1: The AI memo is a good example. It was probably slightly too early to write it, but if you read it now, it says nothing that would be surprising. So being able to give my company the gift of a head start, that’s what we are trying to do. I can tell everyone something that is not quite clearly true now about the importance of these systems. We can rebuild our systems to really reinforce this and support everyone in their own tinkering and exploration. That means our view of the future is going to be more accurate, and so will the things we are building. Because you’re always building for a future point. Software is getting faster to build, but you still have to aim at a future point of value. It’s not our customers’ job to tell us what they need. It’s our customers’ job to tell us what the problems are that they’re experiencing, and we fall in love with those problems, adopt them as our own, and just solve them in an ideal way. But that’s our job, to figure out what the ideal way is. That comes into contact with the customers, and they give us further feedback for refinement. But in my mind, it’s a complete abdication to just build what your customers are asking for. It’s an abdication of product responsibility. So it’s hugely important. I try to live in everyone else’s relative future, so I do a lot of it. While it’s incredibly fun and helpful to talk with the labs, and sometimes there’s really important information you can get from this, what I found, and we touched on it earlier, is that one of my favorite things, especially for product teams and engineering, is to hire people who actually know Shopify really well from outside. Merchants make some of my best product managers, because they actually understand what the software feels like when it’s being used. It’s the same with people who build apps on the Shopify platform, having them come to Shopify to help make the app platform better. I actually don’t even think being in a lab is the best position. It’s actually using everything and paying a lot of attention to how everyone else uses the gifts that the labs release, and then being in the conversation, which usually happens on X, about what everyone’s figuring out. Building something and seeing everyone find it useful, I think this sort of learning by doing is actually how to get the clearest view of how everything works.
· Jack Altman, Tobi Lutke
[00:00:00] S0: Does that give you a clear view on the present, or does it also give you a clear view on what’s coming in a year or two? [00:00:00] S1: If you do it over a while, you get— [00:00:00] S0: You see a trajectory. [00:00:00] S1: You get trajectories, and the trajectories have a point. If you are exactly on the state-of-the-art meta of the best thought on stuff, and you have other data points, the future gets very simple to predict. Now it’s the hardest time ever to do it because right now the time horizons are so short. I did this very same thing throughout my entire life. It used to be that the future was absolutely trivial to predict. You just looked at the first couple numbers on mobile browser usage, and you knew what was happening on a cell phone and that we needed to make mobile websites. That just sounds insane now. But at some point, people didn’t believe that. [00:00:00] S0: Are there parts of AI right now that you think are under-hyped in terms of capabilities? And are there parts that you think are over-hyped?
· Tobi Lutke
[00:00:00] S1: Obviously, the labs really care about programming because that’s the problem they need to solve for themselves. Again, it’s always easier to build for various reasons. Opus is unbelievably good at programming. Right now it’s easy to go from that and assume it’s equally competent in everything else. Quite often it isn’t. If you want to discuss how to do a public talk or something, it just doesn’t have great theoretical rooting of this and why. It knows about all the different ways to construct tension and so on, but it’s not able to then look at something and make it meaningfully better. Whereas even with an incredibly well-optimized piece of code, it often finds abilities to do better. What is happening a lot is that we are bringing more types of problems into the domain of programming, which is really what OpenClaw is, if you think about it. Make it a file system, give it tools like a programmer uses, make a couple of files, and in the files just tell it the soul and the memory and so on. Then it uses the normal programming tools to interrogate this. Because it stays a little bit in the domain of programming, it now actually is remarkably better at this other out-of-domain thing, which is really funny. I think we should assume though that we are going through the entire radial chart of things that are valuable and bringing them to the same point we’ve honed programming to. So there’s a bit of over-hype in having to work so hard on bringing things into the programming domain, because all of it is just going to get much, much more natural and easier as the models appear. Where it’s under-hyped is just in deployment in companies, and what role it should play. There’s a huge—
· Jack Altman, Tobi Lutke
[00:00:00] S0: No one’s using it enough. [00:00:00] S1: No one’s using it enough. Partly because everyone just starts with, “Hey, help me do the things that I’ve been doing all along slightly better”, or actually vastly better, which is valuable, obviously. But if you do the more first-principles thing… If AI had been around ever since Alan Turing first wrote about it and had just worked, and we’d be in the presence of these superintelligences all the time, and we would just invent the job we are thinking about right now, how would we do it? Just invert the whole thing. It’s so much more fun to have those conversations. I’m exceptionally excited about this because I think it will create the environments that the most creative and best people will really appreciate, because everything around them happens at the speed that they want to work. That has never been true for any company that required some bureaucracy for doing stuff. [00:00:00] S0: On the point of talent, have you changed anything about the types of people you’re looking for, pre and post-AI? Are the people who were successful before still the most successful people, or is there any attribute that you are scanning for in a different weighting than you used to?