Computer systems are evolving beyond classical notions of desktops/laptops, servers and even smartphones. They are distributed, embedded, capable of learning and can modify our perception of the world. Securing such systems requires an end-to-end perspective. I will demonstrate the utility of this perspective by discussing our recent results on: (1) building least-privilege distributed systems with applications to the Internet of Things; and (2) establishing threat models for systems that learn. Specifically, I will focus on the class of low-code systems that allow end-users to create small automations that connect their digital and physical resources. Our techniques allow building such automations with least privilege. I will also briefly discuss our efforts at devising threat models for learning-enabled systems that interact with the physical world. One example is a new class of invisible-to-humans physical attack on computer vision.