In most popular galaxy formation models AGN feedback is required in order to reproduce the observed galaxy mass function and the tight correlations between SMBH mass and host properties. Large scale AGN winds are currently the preferred physical process able to explain this feedback, as they are observed in different forms (molecular, ionized, disk winds) in a number of low and high redshift systems. I will revise recent results obtained by our group on this topic in the COSMOS filed, both from targeted observations of peculiar objects, and from populations studies, focusing on highly obscured sources, and the correlations (or lack thereof) between AGN and host properties.