Modelling active galactic nuclei: ongoing problems for the faint-end of the luminosity function