The cloud-scale physics of star formation and feedback represent the main uncertainties in galaxy formation and evolution simulations. I will present our group's efforts towards overcoming this problem by using empirical constraints on the molecular cloud lifecycle to motivate a new generation of sub-grid models in galaxy simulations. Specifically, I will show how we can use the multi-scale nature of the star formation relation between the gas mass and the star formation rate as a direct probe of the cloud-scale physics of star formation and feedback. Using this scale dependence, we can now measure a variety of fundamental quantities, such as the molecular cloud lifetime, star formation efficiency, feedback timescale, feedback terminal momentum, and coherence length scale. While these quantities were previously only accessible in the Local Group, it is now possible to measure them across a representative part of the galaxy population. I will present our group’s results showing that molecular clouds in nearby star-forming galaxies undergo universally fast and inefficient star formation, due to short molecular cloud lifetimes (10-30 Myr) and rapid cloud destruction by stellar feedback (1-5 Myr), causing them to reach integrated star formation efficiencies of only 2-10%. Applying these empirical findings as explicit sub-grid models in galaxy simulations, we find that early, pre-supernova feedback plays a crucial role in structuring the interstellar medium of galaxies, which in turn has a significant impact on the initial clustering of stars at birth and the resulting galactic-scale, feedback-driven outflow rate. Together, these observational, theoretical, and numerical results sketch a physical picture in which the large-scale properties of the galaxy population are shaped by cloud-scale baryonic physics.