The Internet of Things has a problem: batteries. They are large, difficult to replace, and hard to dispose of. Ambient energy harvesting could save the day by providing theoretically infinite energy supplies. However, extreme energy scarcity frequently causes executions to become intermittent, necessitating new hardware and software designs. In this talk, I’ll discuss several challenges and solutions toward achieving an optimal intermittent execution model for transiently-powered systems that can save persistent state on non-volatile memories (NVMs) with minimal overhead to ensure the application's forward progress. Through first-hand experience in multi-year deployments of transiently-powered systems, we argue that the virtuous cycle of theoretical investigation, system implementation, and real-world deployment must be applied to intermittent computing as well, or its potential is bound to remain unexpressed. Furthermore, persistent states on NVMs may be vulnerable to security threats such as stealing sensitive data or tampering with configuration data, which may eventually corrupt the device state and render the system unusable. I will conclude the talk by showing you an experimental study conducted to investigate the impact on typical intermittent computing workloads to protect the persistent state, including software and hardware implementations of staple encryption algorithms and the use of ARM TrustZoneM protection mechanisms.