The paper discusses a new implementation of the Durham semi-analytic model of galaxy formation, which includes feedback from active galactic nuclei (AGN) to quench cooling flows in massive haloes. This mechanism naturally creates a break in the local galaxy luminosity function at bright magnitudes. The model is implemented within the Millennium N-body simulation, which provides accurate dark matter merger trees and a large number of realizations of the galaxy formation process. The authors adjust the physical parameters of the model to match the properties of the local galaxy population and investigate the evolution of the K-band luminosity and galaxy stellar mass functions. They find that the model robustly predicts a substantial population of massive galaxies out to redshifts of about 5 and a star formation rate density that rises at least up to redshifts of about 2 in objects of all masses. The inclusion of AGN feedback in the model provides a good match to observational data on the luminosity function and the 'antihierarchical' nature of star formation in massive ellipticals. The model also successfully explains the evolution of the galaxy stellar mass function out to redshifts greater than 2 and the 'antihierarchical' behavior observed in the high-redshift Universe.The paper discusses a new implementation of the Durham semi-analytic model of galaxy formation, which includes feedback from active galactic nuclei (AGN) to quench cooling flows in massive haloes. This mechanism naturally creates a break in the local galaxy luminosity function at bright magnitudes. The model is implemented within the Millennium N-body simulation, which provides accurate dark matter merger trees and a large number of realizations of the galaxy formation process. The authors adjust the physical parameters of the model to match the properties of the local galaxy population and investigate the evolution of the K-band luminosity and galaxy stellar mass functions. They find that the model robustly predicts a substantial population of massive galaxies out to redshifts of about 5 and a star formation rate density that rises at least up to redshifts of about 2 in objects of all masses. The inclusion of AGN feedback in the model provides a good match to observational data on the luminosity function and the 'antihierarchical' nature of star formation in massive ellipticals. The model also successfully explains the evolution of the galaxy stellar mass function out to redshifts greater than 2 and the 'antihierarchical' behavior observed in the high-redshift Universe.