Agent learning is about observing sequences and trying to draw conclusions about structure that might be present and how this structure might be represented. The idea of typical sequences and a typical set are fundamentals in information theoretic developments of lossless source coding, channel capacity, and rate distortion theory. However, researchers in time series analysis and agent learning may not be familiar with these ideas. In this chapter, typical sequences and the ideas of typical and nontypical sets are introduced and then extended to jointly typical sequences and sets. These ideas fall out of the basic results from the weak law of large numbers. It appears that, although typical sequences have not been exploited in agent learning analyses or to explore structure in sequences, there is much to be gained by translating these ideas into the agent learning problem.