Abstracting an effective theory from a complicated process is central to the study of complexity. Even when the underlying mechanisms are understood, or at least measurable, the presence of dissipation and irreversibility in biological, computational and social systems makes the problem harder. Here we demonstrate the construction of effective theories in the presence of both irreversibility and noise, in a dynamical model with underlying feedback. We use the Krohn-Rhodes theorem to show how the composition of underlying mechanisms can lead to innovations in the emergent effective theory. We show how dissipation and irreversibility fundamentally limit the lifetimes of these emergent structures, even though, on short timescales, the group properties may be enriched compared to their noiseless counterparts. Many systems, especially those with large numbers of underlying degrees of freedom, are best described by what are known as effective theories. These theories allow us to describe the relevant high-level phenomena while remaining ignorant, or at least agnostic, about the fine-grained details of a system's state. Here, we show how to construct effective theories of phenomena that may show irreversibility or dissipation, a question particularly relevant for biological, social and computational systems. We use a hierarchical decomposition technique from semigroup theory that allows one to take finite state automata (standard, if restricted, models of computation) and determine the irreducible components of the process. Truncating the hierarchy provides an effective theory. We study how different underlying mechanisms lead to qualitatively different effective theories, and show how noise and irreversibility interact to produce new computational phenomena on short timescales, which, in the presence of irreversibility, eventually dissipate at longer intervals.