We present a new implementation of the GAlaxy Evolution and Assembly (GAEA) semi-analytic model, that features an improved modelling of the process of cold gas accretion onto supermassive black hole (SMBHs), derived from both analytic arguments and high-resolution simulations. We consider different scenarios for the loss of angular momentum required for the available cold gas to be accreted onto the central SMBHs, and we compare different combinations of triggering mechanisms, including galaxy mergers and disc instabilities in star forming discs. We compare our predictions with the luminosity function (LF) observed for Active Galactic Nuclei (AGN) and we confirm that a non-instantaneous accretion timescale (either in the form of a low-angular momentum reservoir or as an assumed light curve evolution) is needed in order to reproduce the measured evolution of the AGN-LF and the so-called AGN-downsizing trend. Moreover, we also study the impact of AGN feedback, in the form of AGN-driven outflows, on the SF properties of model galaxies, using prescriptions derived both from empirical studies or from numerical experiments. We show that AGN-driven outflows are effective in suppressing the residual star formation rate in massive galaxies ($> 10^{11} M_odot$) without changing their overall assembly history. These winds also affect the SFR of lower mass galaxies, resulting in a too large fraction of passive galaxies at $< 10^{10} M_odot$. Finally, we study the Eddington ratio distribution as a function of SMBH mass, showing that only objects more massive than $10^8 M_odot$ are already in a self-regulated state as inferred from observations.