Electromagnetic (EM) sensing is a powerful tool to non-invasively monitor internal states of systems that are otherwise hard to access. This thesis leverages EM sensing across two distinct domains: biological plants and embedded systems. First, we perform EM modeling and measurements to non-destructively track plant water content. We show that Ka-band EM transmission loss strongly correlates with leaf water content. To overcome the practical limitations of transmission measurements, we develop a model and measurement technique with a single antenna that tracks the loss due to leaf water content via backscattering. The second part of this thesis shifts the focus to systems that are themselves sources of EM signals that can be used to reveal internal operational states. Specifically, we demonstrate how to accurately classify serial protocols used by an embedded device for internal communications via unintentional EM emanations. Measuring the magnetic near field, we are able to distinguish different protocols, data rates, and even message contents with high accuracy. We are also able to perform similar classification using far-field data recorded at lower sampling rates in a noisy laboratory environment. Finally, we consider an integrated embedded platform periodically executing tasks. Instead of passively sensing EM leakage, we create the side channel by intentionally transmitting an EM wave toward the device. Just as the backscatter reflecting off leaves reveals the internal state of the underlying plant, the embedded device modulates the incident wave based on its physical operation. We are able to accurately profile the embedded platform in an unsupervised fashion by analyzing the backscatter. We further characterize the backscattering link under various configurations, and find that our unsupervised framework is highly robust to changes in the backscattering link. This approach can serve as a foundation for profiling inaccessible embedded systems. Ultimately, this thesis demonstrates how EM sensing can serve as a tool to non-invasively obtain hard to access information across disciplines.