In communication system design, it is commonly assumed that user dynamics are much slower than physical-layer resource allocation. However, with the emergence of applications with intermittent and short connectivity requirement, combined with the capabilities of new generation codes, this assumption need not hold. In this work, we develop a framework in which these two time scales are flipped: users encode information across their entire activity period to average out the user dynamics at the physical layer. To that end, we use stochastic geometry to show that the interference decorrelates during the activity period of a user. Next, we devise a coding scheme that maximizes the amount of communicated bits/user/joule. Our work has implications in underlay cognitive radio network design.