The main thesis for Faust is: make machines that make us more human. The more expanded version is, help make machines that make machines that make us more human. The first set of machines is "machines that make us more human" - in this context Roc Camera is the first product. I wanted to build something that I thought would make us more human. A way to capture moments that are verifiably real in the age of AI. Which to be is less about whether it really is real or not (though that matters) and more about whether it feels real. The experience of using a roc camera is what is ultimately the most important in the long run - or i'd like to believe is the case if we are making machines that make us more human. We want to continue to make more machines (like Roc Camera) that can do this/ The second set of machines, which I think Roc Explorer, is increasingly going to fall under is "machines that make machines" with the aim that they will make us more human. Part of that is about creating infrastructure like manufacturing (we put a lot of time into how to build the machine that produces the roc camera, i'd say half our effort in v1 of Roc Camera went into how to make the machine that makes the machines.) So in this sense, part of our bet with Roc Explorer is - can it be a place where the value of the photos themselves (that they can be proved they were taken with roc camera) provides a place where agents / robots / machines / vehicles can start querying and using these photos (it's really just sensor data) as anchor points about what is the case. Agents themselves don't care about epistemic truth of a certain photo being taken as you say, but the the people prompting the agents are ultimately humans, still. And humans that want these agents ultimately to do something (I presume) would still want them rooted in reality, not some made up thing. In this context, in the longer run, I think of it in the agent context as: how do we create photos or sets of data that are injectable into contexts? If you can reference real-world data points that you know and can trust are real, it essentially improves queries for agents significantly better. In the longer run, that's part of the thing: if you can understand your context significantly better with data that essentially you can trust, it's just easier to have better results. It's sort of like the difference between having a map that references reality at a higher fidelity versus asking people around about how to get around at some location. You just don't necessarily know if they're local or not. So in this sense, while the main goal continues to be to make machines that make us more human, I also want to increasingly experiment and open up the second part of our aim, which feeds into the first goal anyway. That is to make machines that make machines more human. In this context, making Roc Explorer is a way to help get better context for the machines that make machines about what's going on. Yes, a lot of the 'how' is still further out in practice, but this is more of the motivation behind the why and the thinking behind why we built Roc Explorer