I discuss what can be learned about galaxy formation theories from studying how active galactic nuclei
(AGN) occupy galaxies. In simulations, AGN play the role of quenching and preventing further star
formation in the most massive galaxies, preventing them from becoming even higher stellar mass than they
are. The cosmological simulations require subgrid physics to describe star formation processes as well as
black hole growth and feedback, both of which are complicated, nonlinear processes occurring at AU scales,
within simulations with at best about 10 to 100 pc resolution. Among the major cosmological simulation
codes, there are many different implementations of this subgrid physics that all lead to predictions of the
stellar mass function and star formation rates of galaxies that agree with observations, and this agreement is
taken very seriously as validation of the basic picture. These different implementations nevertheless lead to
quite different predictions for how AGN populate galaxies as a function of mass and star formation rate.
These predictions, in contrast to those regarding stellar properties of galaxies, are almost never compared
against observations. I describe a research program to perform these tests. This program requires a major
revision of the statistics of AGN demographics in local galaxies, which in many cases have been performed
previously without regard to the dramatic selection effects on AGN samples in galaxies. Although there are
many methodological difficulties in interpreting the AGN predictions from galaxy formation simulations, I
argue that observations of the AGN population can be used to distinguish between the simulations.