We briefly review advances in understanding the initial stages of a heavy ion collision. In particular the focus is on
moving from parametrizing the initial state to calculating its properties from QCD, consistently with the description
of hard probes and dilute-dense scattering experiments. Modeling the event-by-event fluctuating nuclear geometry in
initial state calculations has significantly improved in recent years. We also discuss prospects of directly seeing effects
of particle correlations created in the initial state in the experimental observables.