Take something like Augmented Reality, there's a lot for the imagination to run with like reduced dependencies on physical location for high-fidelity communication (less text, more full-body expression). Or we could imagine dynamic tutorials: pop the hood in your car and see a live overlay of the parts and ways to replace them.

So we take this imagined future and think about how aspects might be implemented. What problems and barriers are in the way? If I'm interested in hardware maybe I could tinker with Lidar sensors and software to model physical spaces. If I'm interested in machine learning and statistics maybe focus on object detection. Maybe you're more about design and UI and can focus on the way people will want to interact with the digital overlay. Maybe you're into architecture and can spend time thinking about how to design physical spaces to integrate with the new augmented reality capabilities.

You're not going to solve every problem across those diverse domains, but you can still imagine the possibilities and find areas of interest that you're motivated to spend time in. I feel like this is what they mean by being 'upwind' of opportunity - you're not consuming paths set by others (downstream effects) but what has a likelihood of future value and possibilities.