For work, my team has been moving legacy architecture to a new stateless model while continuing to support billions of events going through our system. Our software runs on thousands of machines. It is very much switching out parts on a moving vehicle. High scale, high availability, updating arch, etc. I think it is interesting. You can't just start a green field thing and move over wholesale. You have to move over parts and for subsets of users. You have to incorporate years of bug fixes and learnings in the legacy stack to the new stack, and the new stack has new and different scaling challenges.